CN113656691B

CN113656691B - Data prediction method, device and storage medium

Info

Publication number: CN113656691B
Application number: CN202110943383.6A
Authority: CN
Inventors: 耿东阳; 张建申
Original assignee: Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2021-08-17
Filing date: 2021-08-17
Publication date: 2025-09-19
Anticipated expiration: 2041-08-17
Also published as: CN113656691A; WO2023020257A1

Abstract

The present invention provides a data prediction method, device and storage medium, which obtains hierarchical time series data; the hierarchical time series data is multiple groups of data corresponding to the time series of each level, wherein the sum of the data of the sub-levels of each level in each level is equal to the data of the corresponding parent level; the hierarchical time series data is predicted using a preset data prediction model to determine the prediction results within a preset time period after multiple historical time periods; wherein the preset data prediction model is trained based on the prediction errors of multiple groups of training data within the historical preset time period in the hierarchical time series data, as well as the errors between the various levels. Since the preset data prediction model not only takes into account the accuracy of the prediction error during training, but also takes into account the errors between the various levels, the trained preset data prediction model is more accurate in predicting data.

Description

Data prediction method, device and storage medium

Technical Field

The embodiment of the invention relates to the technical field of prediction models, in particular to a data prediction method, a data prediction device and a storage medium.

Background

Time series prediction has wide application fields such as demand prediction in retail industry, financial market quotation prediction, logistics goods amount prediction and the like. In many business processes for implementing automation and intelligence, time series prediction plays a very important role, such as online shopping websites, and the sales of each type of commodity in a future period of time needs to be predicted to be a series of business decisions. Such as stock, promotions, etc., and thus the ability to predict technology ultimately has a significant impact on sales revenue, inventory costs, etc. Meanwhile, the number of commodities sold by large-scale online shopping websites can reach millions, and a large-scale time sequence creates new challenges for modern time sequence prediction technology.

In the prior art, the time sequence prediction is performed by using single-level data, and then other-level prediction results are obtained by splitting or aggregating, so that the time sequence prediction is simple and convenient to use, but the prediction accuracy is generally relatively low. In addition, since the results obtained using different single levels are different, not only is manual experience relied upon in selecting the level, but also accuracy loss results.

Disclosure of Invention

The data prediction method, the data prediction device and the storage medium can improve the accuracy of data prediction.

The technical scheme of the invention is realized as follows:

The embodiment of the invention provides a data prediction method, which comprises the following steps:

The hierarchical time sequence data is a plurality of groups of data corresponding to each hierarchical time sequence, wherein the sum of data of sub-hierarchies of each hierarchy is equal to data of a corresponding father hierarchy;

predicting the layered time series data by using a preset data prediction model to determine a prediction result in a preset time period after a plurality of historical time periods, wherein,

The preset data prediction model is obtained by training together based on prediction errors of multiple groups of training data in a historical preset time period and errors among various layers in layered time sequence data.

In the above scheme, the method further comprises, before predicting the layered time series data by using the preset data prediction model and determining the prediction results in the preset time period after the plurality of historical time periods, acquiring the layered time series data after the prediction results in the preset time period are obtained:

Carrying out standardization processing on a plurality of groups of data of the layered time sequence data, and dividing the plurality of groups of data subjected to the standardization processing into a training set and a testing set according to a preset historical time period, wherein the training set comprises a plurality of groups of training data;

Calculating the prediction error of the training set and the error among all the layers by using the loss function of the initial prediction model, and carrying out iterative adjustment on model parameters of the initial prediction model according to the prediction error and the error among all the layers until the training condition is met, and stopping to obtain a first prediction data set corresponding to the testing set;

and comparing the plurality of groups of test data with the first prediction data set to determine a preset data prediction model.

In the above scheme, calculating the prediction error of the training set and the error between each level by using the loss function of the initial prediction model, and performing iterative adjustment on the model parameters of the initial prediction model according to the prediction error and the error between each level until the training condition is satisfied, to obtain a first prediction data set corresponding to the test set, including:

Inputting multiple groups of training data into an initial prediction model to obtain a second prediction data set, wherein the second prediction data set comprises prediction data of each level of multiple historical time periods;

calculating a prediction error and an error between each level in combination with the loss function based on the second prediction data set and the plurality of sets of training data;

carrying out gradient solving on the loss function to obtain model parameters in the iterative process, thereby obtaining an updated prediction model;

Continuously training a plurality of groups of training data by using the updated prediction model until the training condition is met, and stopping to obtain a final prediction model, thereby obtaining a plurality of prediction models in the iterative process;

And extracting the prediction data of each level of each historical time period of the corresponding test set from each corresponding second prediction data set by adopting a plurality of prediction models, and further obtaining a first prediction data set in the iterative process.

In the above scheme, calculating the prediction error and the error between each level based on the second prediction data set and the plurality of sets of training data in combination with the loss function includes:

Calculating a prediction error based on first prediction data in the second prediction data set and a plurality of groups of training data, wherein the first prediction data is prediction data of each level in a plurality of first time periods in the second prediction data set, and the plurality of first time periods are time periods before a preset historical time period in a plurality of historical time periods;

The method comprises the steps of calculating errors among all levels based on second prediction data in a second prediction data set, wherein the second prediction data are prediction data of all levels in a plurality of second time periods in the second prediction data set, and the plurality of second time periods are time periods after a preset historical time period in a plurality of historical time periods.

In the above scheme, calculating the prediction error based on the first prediction data in the second prediction data set and the plurality of sets of training data includes:

and calculating the square sum of the difference between the first prediction data in the same first time period and the training data of the corresponding level, further obtaining the first sum of each level in the same first time period, and adding a plurality of first sums corresponding to a plurality of first time periods to obtain the prediction error.

In the above scheme, calculating the error between each hierarchy based on the second prediction data in the second prediction data set includes:

calculating the sum of squares of the difference between the prediction data of each father level of each layer in the second prediction data in the same second time period and the prediction data sum of each corresponding child level, and adding the square sums of a plurality of second time periods to obtain a second sum;

and multiplying the second sums by the harmonic error penalty term super-parameters to obtain the error between each level.

In the above scheme, comparing the plurality of sets of test data with the first prediction data set to determine a preset data prediction model, including:

Comparing the multiple groups of test data with multiple prediction data in the first prediction data set respectively, and determining multiple comparison errors corresponding to the multiple prediction data;

determining a target comparison error in a preset error range from the multiple comparison errors;

Determining a target iteration time corresponding to target secondary prediction data corresponding to the target comparison error;

And determining a prediction data prediction model corresponding to the target iteration number in the plurality of prediction models.

In the scheme, the plurality of sets of training data comprise a plurality of sets of first processed data, and the plurality of sets of test data comprise a plurality of sets of second processed data;

Performing standardization processing on multiple groups of data of the layered time sequence data, and dividing the multiple groups of data after the standardization processing into a training set and a testing set according to a preset historical time period, wherein the method comprises the following steps:

deleting the abnormal value of each level in the plurality of groups of data, and filling the average data of the level corresponding to the abnormal value;

Filling blank data corresponding to each level in the plurality of sets of data by using average data corresponding to the level with the blank data, so as to obtain a plurality of sets of processed data corresponding to the time sequence of each level;

Determining a preset historical time period in the plurality of historical time periods, combining a plurality of groups of first processed data corresponding to a plurality of first time periods before the preset historical time period into a training set, and combining a plurality of groups of second processed data corresponding to a plurality of second time periods after the preset historical time period into a test set.

In the above scheme, the method further comprises:

acquiring a plurality of sets of logistics cargo volume data corresponding to a plurality of historical time periods;

And processing the plurality of groups of logistics cargo amount data by using a preset data prediction model to obtain predicted logistics cargo amount data of a preset time period after a plurality of historical time periods.

The embodiment of the invention also provides a data prediction device, which comprises:

The data acquisition unit is used for acquiring layered time sequence data, wherein the layered time sequence data is a plurality of groups of data corresponding to each layer of time sequence, and the sum of data of sub-layers of each layer is equal to the data of a corresponding father layer;

A prediction unit for predicting the hierarchical time series data by using a preset data prediction model to determine the prediction results in a preset time period after a plurality of historical time periods,

The preset data prediction model is obtained by training together based on the prediction errors of a plurality of groups of training data in a historical preset time period in the layered time sequence data and the errors among all the layers.

The embodiment of the invention also provides a data prediction device, which comprises a memory and a processor, wherein the memory stores a computer program which can be run on the processor, and the processor realizes the steps in the method when executing the program.

The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when being executed by a processor, implements the steps of the above method.

In the embodiment of the invention, layered time sequence data is acquired, the layered time sequence data is a plurality of groups of data corresponding to each layer of time sequence, wherein the sum of data of sub-layers of each layer is equal to data of a corresponding parent layer, the layered time sequence data is predicted by a preset data prediction model to determine a prediction result in a preset time period after a plurality of historical time periods, and the preset data prediction model is obtained by training together based on prediction errors of a plurality of groups of training data in the historical preset time period in the layered time sequence data and errors among the layers. The preset data prediction model is obtained based on the prediction errors of a plurality of groups of training data in a historical time period and the errors among all the layers, and not only the accuracy of the prediction errors but also the errors among all the layers are considered during training, so that the preset data prediction model obtained through training is more accurate in data prediction.

Drawings

FIG. 1 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

FIG. 2 is a schematic diagram showing an optional effect of the data prediction method according to the embodiment of the present invention;

FIG. 3 is a schematic diagram showing an optional effect of the data prediction method according to the embodiment of the present invention;

FIG. 4 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

FIG. 5 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

FIG. 6 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

FIG. 7 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

FIG. 8 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

FIG. 9 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

FIG. 10 is a schematic structural diagram of a logistics cargo amount prediction device according to an embodiment of the present invention;

FIG. 11 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention;

fig. 12 is a schematic structural diagram of a data prediction apparatus according to an embodiment of the present invention;

FIG. 13 is a second schematic structural diagram of a data prediction apparatus according to an embodiment of the present invention;

fig. 14 is a schematic diagram of a hardware entity of a data prediction apparatus according to an embodiment of the present invention.

Detailed Description

The technical solution of the present invention will be further elaborated with reference to the accompanying drawings and examples, which should not be construed as limiting the invention, but all other embodiments which can be obtained by one skilled in the art without making inventive efforts are within the scope of protection of the present invention.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with one another without conflict.

If a similar description of "first/second" appears in the present document, the following description is added, in which the terms "first/second/third" merely distinguish similar objects and do not represent a specific ordering of the objects, it being understood that the "first/second/third" may, where allowed, interchange a specific order or precedence order such that the embodiments of the invention described herein can be implemented in an order other than that illustrated or described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to be limiting of the invention.

In the prior art, for example, a nationwide fast-food manufacturing enterprise needs to predict future sales of a certain product nationally and in various provinces at the same time so as to make an inventory layout and a stock plan. The prediction scheme is to perform single time sequence prediction on sales time sequences of each province and each province respectively, so that the prediction results of different levels often cannot automatically meet consistency, namely the total sum of sales predictions of each province is not equal to the total sum of sales predictions of each province, and the 'inconsistent' prediction result cannot be used in the collaborative decision flow of each level.

The main prediction methods at present comprise Top-Down (Top-Down), bottom-Up (Bottom-Up), middle breakthrough (Middle-Out) and optimal blending. As the name suggests, "top-down" refers to predicting the highest level time series first, then splitting the prediction result into lower levels according to a fixed ratio, and "bottom-up" refers to predicting the finest granularity time series first, then aggregating the prediction result upward. The "middle break-through" approach combines bottom-up and top-down approaches. First, a "middle level" is selected and predictions are generated for all sequences at that level. For series above the intermediate level, a bottom-up approach is used to generate consistent predictions by summarizing the predictions of the "intermediate level" upward. For sequences below "intermediate level", consistent predictions are generated using a top-down approach by decomposing the predictions of "intermediate level" downward. The optimal blending method is to obtain all levels of predicted results, process the predicted results in an optimal linear weighted blending mode, and obtain final results.

The three methods of top-down, bottom-up and middle breakthrough are the most used methods at present, the methods are used for predicting by using only single-level data, and then obtaining prediction results of other levels by splitting or aggregation, and although the method is simple and convenient to use, the prediction accuracy is usually relatively low, and the main defects are that firstly, the three prediction methods are used for predicting the data by using only single-level substantially, the information contained in the prediction data of other levels is not utilized, the accuracy loss is caused, and secondly, the prediction errors are additionally introduced when the prediction results are summarized upwards or decomposed downwards. In addition, since the results obtained using different single levels are different, not only is manual experience relied upon in selecting the level, but also accuracy loss results.

In order to solve the above technical problem of low prediction accuracy of the prediction model, the embodiment of the present invention further provides a data prediction method, referring to fig. 1, which is an optional flowchart of the data prediction method provided in the embodiment of the present invention, and will be described with reference to the steps shown in fig. 1.

And S101, acquiring hierarchical time sequence data, wherein the hierarchical time sequence data are a plurality of groups of data corresponding to each hierarchical time sequence, and the sum of data of sub-hierarchies of each hierarchy is equal to the data of a corresponding parent hierarchy.

In the embodiment of the invention, a server acquires layered time sequence data. The hierarchical time series data are multiple groups of data corresponding to each hierarchical time series. The sum of the data of the child levels of each of the respective levels is equal to the data of the corresponding parent level.

In the embodiment of the invention, the server establishes communication connection with the client corresponding to each hierarchy in advance. The server acquires, from the clients, a plurality of sets of data of each hierarchy corresponding to a plurality of time periods of history through communication connection with the clients of each hierarchy. That is, the server obtains multiple sets of data corresponding to each hierarchy time sequence from clients of each hierarchy.

In the embodiment of the invention, a server acquires pre-stored layered time sequence data from a database of the server.

Any one set of data in the plurality of sets of data can comprise any combination of data corresponding to each level of the time sequence. That is, any one of the sets of data may include a combination of data of respective levels corresponding to any one of the historical plurality of time periods. Wherein, one data in a group of data can be any one of sales volume data, flow volume data and user age data of the corresponding level. Wherein the flow data may be one of a corresponding total number of streams, a total weight of streams, and a total volume of streams.

For example, the time series may be three time series corresponding to three months before the current time. The time sequence may also be three time sequences corresponding to three days before the current time, and in the embodiment of the present invention, the time sequence is not limited.

In the embodiment of the present invention, the child level may be a city level, and the parent level may be a provincial level corresponding to the child level. One province level may correspond to multiple city levels. The parent level may also be a primary agent and the child level may be a plurality of secondary agents corresponding to the parent level. One primary agent may correspond to a plurality of secondary agents. The sum of the data of the plurality of child levels is the data of the corresponding parent level.

In the embodiment of the invention, a server firstly collects layered time sequence data to be predicted. The i-th time series observed by the server at time 1-T is noted as y _i＝(y^t _i,…,y^T _i)^T, i=1. Wherein y _i＝(y^t _i,…,y^T _i)^T characterizes data for each level of 1-T time periods, 1.

Wherein the hierarchical time series data satisfies that the sum of the data of each child level is equal to the data of the corresponding parent level. In connection with FIG. 2, the hierarchy satisfies y ₁＝y₂+y₃. Where y ₁ is parent level data corresponding to y ₂ and y ₃, and y ₂ and y ₃ is child level data corresponding to y ₁.

In connection with fig. 3, the hierarchical structure of the hierarchical time series data satisfies y ₁＝y₂+y₃,y₂＝y₄+y₅,y₃＝y₆+y₇. Where y ₂ is the parent level data corresponding to y ₄ and y ₅ and y ₃ is the parent level data corresponding to y ₆ and y ₇. The task of hierarchical time series prediction is to predict the value of all time series in the future period t+h given the observed data at time 1Where y is data in time-series data. When the time series is in the day dimension, then the daily inventory is the amount of goods. For example, for 7 months 1 day to 5 days, y_beijing city= (10, 20,30,40, 50), y_northbound province= (30,40,50,60,10). In hierarchical time series research and application, the constraint condition is expressed in a commonly shown hierarchical structure image. Such constraints are fundamental features of hierarchical timing, and are also reflected in the meaning of "hierarchical", and are natural rules that each variable satisfies in a statistical range, such as y_nationwide=sum (y_beijing, y_northly, and.+ -.), y_beijing city=sum (y_sealake, and.+ -.), and y_western city).

S102, predicting the hierarchical time sequence data by using a preset data prediction model to determine a prediction result in a preset time period after a plurality of historical time periods, wherein the preset data prediction model is obtained by training based on prediction errors of a plurality of groups of training data in the hierarchical time sequence data and errors among all the layers.

In the embodiment of the invention, a server predicts hierarchical time sequence data by using a preset data prediction model to determine a prediction result in a preset time period after a plurality of historical time periods, wherein the preset data prediction model is obtained by training a plurality of groups of prediction errors of training data in the historical preset time period and errors among all layers together in the hierarchical time sequence data.

In the embodiment of the invention, the server divides the multiple groups of data into a training set and a testing set. The server carries out iterative training on the initial prediction model through the training data and the loss function. The server obtains a plurality of prediction models corresponding to the iteration for a plurality of times through iterative training. And the server compares the predicted data of each iteration with corresponding real data to obtain the predicted error of each iteration. And the server determines the prediction model corresponding to the iteration with the minimum error as a preset data prediction model.

In the embodiment of the invention, hierarchical time sequence data is obtained, the hierarchical time sequence data is a plurality of groups of data corresponding to each hierarchical time sequence, wherein the sum of data of sub-hierarchies of each hierarchy is equal to data of a corresponding father hierarchy, the hierarchical time sequence data is predicted by using a preset data prediction model, and a prediction result in a preset time period after a plurality of historical time periods is determined, wherein the preset data prediction model is obtained by training together based on prediction errors of a plurality of groups of training data in the historical preset time period in the hierarchical time sequence data, and errors among the hierarchies. The preset data prediction model is obtained based on the prediction errors of a plurality of groups of training data in a historical time period and the errors among all the layers, and not only the accuracy of the prediction errors but also the errors among all the layers are considered during training, so that the preset data prediction model obtained through training is more accurate in data prediction.

In some embodiments, referring to fig. 4, fig. 4 is a schematic flow chart of an alternative data prediction method provided in the embodiment of the present invention, and S101 shown in fig. 1 further includes S103 to S105 for implementation, and will be described in connection with the steps.

S103, carrying out standardization processing on a plurality of groups of data of the layered time sequence data, and dividing the plurality of groups of data after the standardization processing into a training set and a testing set according to a preset historical time period.

In the embodiment of the invention, the server performs standardization processing on multiple groups of data of the hierarchical time sequence data, and divides the multiple groups of standardized data into a training set and a testing set according to a preset historical time period. The training set comprises a plurality of groups of training data. The test set includes a plurality of sets of test data.

In the embodiment of the invention, the server can delete redundant data in the plurality of groups of data and fill the redundant data with average data of corresponding layers, or the server can fill blank data of each layer in the plurality of groups of data through the average data of the corresponding layers, so as to obtain the processed plurality of groups of data. Since the sets of data correspond to a plurality of historical time periods. The server determines a preset historical time period from a plurality of historical time periods, and the server determines several groups of training data corresponding to the preset historical time period as a training set. The server determines several groups of test data corresponding to the preset historical time period as a test set.

Wherein the training set is a data set used to train the initial predictive model. The test set is a data set used to determine a pre-set data prediction model.

In the embodiment of the invention, the server preprocesses a plurality of groups of data, deletes abnormal values and missing value filling, and standardizes the data. The preprocessed data is then taken at a certain time T0 and divided into training sets (t=1,..t 0) and test sets (t=t0+1,..t) according to the usage.

S104, calculating the prediction error of the training set and the error among all the layers by using the loss function of the initial prediction model, and carrying out iterative adjustment on the model parameters of the initial prediction model according to the prediction error and the error among all the layers until the training condition is met, and stopping to obtain a first prediction data set corresponding to the testing set.

In the embodiment of the invention, a server calculates the prediction error of the training set and the error among all the layers by using the loss function of the initial prediction model, and iteratively adjusts the model parameters of the initial prediction model according to the prediction error and the error among all the layers until the training condition is met, and then the first prediction data set corresponding to the testing set is obtained. Wherein the first set of prediction data comprises a plurality of prediction data in an iterative process corresponding to each level of each historical time period of the test set.

In the embodiment of the invention, the server inputs a plurality of groups of training data in the training set into the initial prediction model. A second set of prediction data is obtained. The second set of prediction data includes prediction data for each level of the plurality of historical time periods. The server calculates the prediction errors of the plurality of groups of training data and the errors among the layers based on the second prediction data set and the plurality of groups of training data by combining the loss function. And solving the loss function by the server to obtain the model parameters of the training. And the server adjusts the initial prediction model according to the model parameters to obtain a new prediction model. And the server continues to train a plurality of groups of training data through the new prediction model until the training condition is met, and the final prediction model is obtained. Simultaneously, a first prediction data set corresponding to the test set in the iterative process is also obtained.

The training condition may be that a preset training frequency is reached or a loss function value converges.

S105, comparing the plurality of groups of test data with the first prediction data set to determine a preset data prediction model.

In the embodiment of the invention, the server compares a plurality of groups of test data with the first prediction data set to determine the preset data prediction model.

In the embodiment of the invention, the first prediction data set comprises a plurality of prediction data corresponding to a plurality of iterations of the test set. The server compares the data of each level in each time period in the plurality of groups of test data with the corresponding data in each prediction data, determines the error of each level, and then adds the errors of each level to obtain the error of each prediction data. And then multiple errors corresponding to the multiple prediction data can be determined. The server determines that the prediction model with the minimum error corresponding to the one-time prediction data after iterative adjustment is the preset data prediction model.

Illustratively, the server subtracts the data of each level in each time period in the plurality of sets of test data from the corresponding data in the predicted data, and obtains the error of the data of each level in each time period. The server adds the errors of the data of each level of each time period, and can obtain the errors corresponding to the secondary prediction data.

In the embodiment of the invention, the server iteratively adjusts the prediction model through the prediction error and the error between each level, and a plurality of prediction models in the iterative process are obtained. And the server compares the plurality of groups of test data with the first prediction data set to determine a preset data model. The preset data prediction model is obtained based on the prediction errors of a plurality of groups of training data in a historical time period and the errors among all the layers, and not only the accuracy of the prediction errors but also the errors among all the layers are considered during training, so that the preset data prediction model obtained through training is more accurate in data prediction.

In some embodiments, referring to fig. 5, fig. 5 is a schematic flow chart of an alternative data prediction method provided in the embodiment of the present invention, and S104 shown in fig. 4 may also be implemented through S106 to S110, and will be described in connection with the steps.

S106, inputting a plurality of groups of training data into the initial prediction model to obtain a second prediction data set.

In the embodiment of the invention, the server inputs a plurality of groups of training data into the initial prediction model to obtain a second prediction data set of the first iteration in the iteration process. Wherein the second set of prediction data comprises prediction data for each level of the plurality of historical time periods.

In the embodiment of the invention, the server inputs a plurality of groups of training data into the initial prediction model to obtain a second prediction data set for the first training. The server calculates a prediction error and an error between each layer in accordance with the first second prediction data set in combination with the loss function. And the server obtains model parameters according to the prediction errors and the errors among all the levels, and adjusts the initial prediction model to obtain the prediction model updated next time. The server inputs the multiple sets of training data into the prediction model updated next time again, and then the process is executed, so that iteration is completed.

And S107, calculating the prediction error and the error between each level by combining the loss function based on the second prediction data set and the plurality of groups of training data.

In the embodiment of the invention, the server calculates the prediction error and the error between each level by combining the loss function based on the second prediction data set and the plurality of groups of training data. Wherein the loss function is a function corresponding to the initial predictive model.

In the embodiment of the invention, the server calculates the prediction errors corresponding to the multiple groups of training data in the multiple first time periods by combining the loss function based on the second prediction data set and the multiple groups of training data. The plurality of first time periods are time periods before a preset historical time period in the plurality of historical time periods. The prediction error characterizes an error between the prediction data and corresponding data in the plurality of sets of training data.

In an embodiment of the present invention, the server calculates, based on the second set of predicted data, an error between the respective levels within the second combination of predicted data for the plurality of second time periods in combination with the loss function. The plurality of second time periods are time ends after the preset historical time period in the plurality of historical time periods. Errors between the respective levels characterize errors of data of a parent level and corresponding child level data sums in the second prediction data set.

S108, carrying out gradient solving on the loss function to obtain model parameters in the iterative process, thereby obtaining an updated prediction model.

In the embodiment of the invention, the server performs gradient solving on the loss function to obtain the model parameters in the iterative process, thereby obtaining the updated prediction model.

In the embodiment of the invention, the server solves the gradient of the loss function after each iteration in the iteration process, and the model parameters of each iteration in the iteration process are obtained. The server adjusts the prediction model of the time through model parameters of each time to obtain an updated prediction model.

And S109, training the plurality of groups of training data by using the updated prediction model until the training conditions are met, and stopping to obtain a final prediction model, thereby obtaining a plurality of prediction models in the iterative process.

In the embodiment of the invention, the server uses the updated prediction model to continuously train a plurality of groups of training data until the training condition is met, and the final prediction model is obtained, so that a plurality of prediction models in the iterative process are obtained.

In the embodiment of the invention, the network structure of the prediction model is based on the prediction error and the error between each level, and the prediction error is transmitted back to the middle layer and the input layer by layer, and the weight of each layer is corrected in a gradient descending mode. After the weight of each layer of the network structure of the prediction model is corrected, a new prediction model is obtained. The new network structure of the prediction model can continue to train the training set until the training condition is met, and a plurality of prediction models in the iterative process are obtained.

S110, extracting the prediction data of each level of each historical time period of the corresponding test set from each corresponding second prediction data set by adopting a plurality of prediction models, and further obtaining a first prediction data set in the iterative process.

In the embodiment of the invention, the server extracts the prediction data of each level of each historical time period of the corresponding test set from each corresponding second prediction data set by adopting a plurality of prediction models, and further obtains the first prediction data set in the iterative process.

In the embodiment of the invention, the server extracts the predicted data of each level of each historical time period of the corresponding test set from each second predicted data set, and obtains one predicted data set corresponding to each iteration. The server combines the one prediction data set for each iteration to form a first prediction data set.

In the embodiment of the invention, the server inputs a plurality of groups of training data into the initial prediction model to obtain a second prediction data set. And the server calculates the prediction error and the error between each level through the second prediction data set. The server iteratively adjusts the prediction model through the prediction error and the error between each level, and a plurality of prediction models in the iterative process are obtained. Meanwhile, the server can extract the first prediction data set from the plurality of second prediction data sets in the iterative process to perform comparison. The preset data prediction model is obtained based on the prediction errors of a plurality of groups of training data in a historical time period and the errors among all the layers, and not only the accuracy of the prediction errors but also the errors among all the layers are considered during training, so that the preset data prediction model obtained through training is more accurate in data prediction.

In some embodiments, referring to fig. 6, fig. 6 is a schematic flow chart of an alternative data prediction method provided in the embodiment of the present invention, S107 shown in fig. 5 may also be implemented through S111 to S112, and the description will be made with reference to the steps.

S111, calculating a prediction error based on the first prediction data in the second prediction data set and multiple sets of training data.

In the embodiment of the invention, the server calculates the prediction error based on the first prediction data in the second prediction data set and a plurality of groups of training data.

The first prediction data is prediction data of each level in a plurality of first time periods in the second prediction data set. The plurality of first time periods are time periods before a preset history time period among the plurality of history time periods.

S112, calculating errors among all levels based on second prediction data in the second prediction data set.

In the embodiment of the invention, the server calculates the error between each level based on the second prediction data in the second prediction data set.

The second prediction data is prediction data of each level in a plurality of second time periods in the second prediction data set. The plurality of second time periods are time periods after a preset history time period among the plurality of history time periods.

In the embodiment of the invention, a server constructs a layered time sequence prediction model based on DeepAR. The DeepAR model is a time sequence prediction model based on a cyclic neural network, and can be used for general time sequence prediction, but cannot be directly used for layered time prediction. Thus for the hierarchical time series prediction task, the improved loss function (1) for hierarchical time series prediction designed by the present invention is:

Wherein, the In order to predict the error loss,As a loss function of DeepAR model, without loss of generality, let l (x, y) = (x-y) ²,Is the inter-level harmonic error loss, where λ is the harmonic error penalty term super-parameter. C is a set of constraints derived from the hierarchy.Is the predicted value of the "parent node" time series in constraint c at time t,Is the predicted value of the time sequence of the 'leaf nodes' in the constraint condition c at the time t, and J (c) is the number of the 'leaf nodes'. For example, the hierarchical time series data of the structure shown in FIG. 3, the constraint satisfied by the hierarchy is C＝{y₁＝y₂+y₃,y₂＝y₄+y₅,y₃＝y₆+y₇}.

Wherein, the Is a predicted value.Is thatCorresponding training data. n is the number of each level and t ₀ is the number of the plurality of first time periods. T is the number of the plurality of second time periods.

In some embodiments, referring to fig. 7, fig. 7 is a schematic flow chart of an alternative data prediction method provided in the embodiment of the present invention, and S111 to S112 shown in fig. 6 may also be implemented through S113 to S115, and each step will be described in connection with the description.

S113, calculating the square sum of differences between the first prediction data in the same first time period and training data of corresponding levels, further obtaining first sums of all levels in the same first time period, and adding a plurality of first sums corresponding to a plurality of first time periods to obtain prediction errors.

In the embodiment of the invention, the server calculates the square sum of the difference between the first prediction data and the training data of the corresponding level in the same first time period, so as to obtain the first sum of all levels in the same first time period. The server adds the first sums corresponding to the first time periods to obtain a prediction error.

The plurality of first time periods includes, for example, two first time periods. Each hierarchy includes a parent hierarchy (primary agent) and two corresponding child hierarchies (two secondary agents). The server calculates the sum of squares of the differences between the data of the parent level and the corresponding prediction data in the first time period, calculates the sum of squares of the differences between the data of the two child levels and the corresponding prediction data, and adds the sum of squares of the corresponding differences of the parent level and the sum of squares of the differences of the corresponding two child levels to obtain the first sum corresponding to the first time period. Similarly, the server calculates a first sum corresponding to the second time period by the same method. The server adds the two first sums to obtain a prediction error.

S114, calculating the sum of squares of the differences between the prediction data of each parent level of each layer in the second prediction data in the same second time period and the corresponding prediction data sum of each child level, and adding the sum of squares of the second time periods to obtain a second sum.

In the embodiment of the invention, the server calculates the sum of squares of the difference between the prediction data of each parent level of each layer in the second prediction data in the same second time period and the sum of the prediction data of each corresponding child level, and adds the square sums of a plurality of second time periods to obtain a second sum.

The plurality of second time periods includes, for example, two second time periods. Each hierarchy includes a parent hierarchy (primary agent) and two corresponding child hierarchies (two secondary agents). The server calculates a sum of squares of differences between data of the parent level and a sum of predicted data of the corresponding respective child levels in the first and second time periods. Similarly, the server calculates the sum of squares corresponding to the second time period by the same method. The server adds the two sums of squares to obtain a second sum.

S115, multiplying the second sums by the harmonic error penalty term super-parameters to obtain the error between the layers.

In the embodiment of the invention, the server obtains errors among all levels by using a plurality of second sum and harmonic error penalty term superparameters.

The harmonic error penalty term super-parameter may be any positive number.

Compared to normal time series prediction, hierarchical time series prediction essentially adds a constraint on the consistency between levels of the final prediction result, namely:

The problem of large-scale optimization is very difficult to directly solve, and after constraint conditions are added into a loss function as penalty terms, the method of random gradient descent and the like can be used for solving the formula (1), so that the inconsistency of prediction results among layers can be reduced along with the reduction of the loss function value in the training process when any given difference penalty term exceeds the parameter lambda.

For hierarchical time series prediction, future levels of timing must meet the inter-level consistency. By adding a penalty term for the harmonic error loss, from the process of parameter iteration, the method is equivalent to the fact that the network parameters requiring DeepAR can give consideration to both prediction deviation and hierarchical structure deviation in the optimization process. From the results, this is equivalent to the lower bound on the optimum test set error. Taking the hierarchical timing in fig. 2 as an example, according to the cauchy inequality:

It can be seen that the hierarchical deviation term in the loss function is essentially a lower bound of prediction error, and it is intuitively understood that, although the prediction result satisfying the inter-level consistency does not necessarily guarantee that the prediction accuracy is the highest, since the future real data must satisfy the consistency, if the error between the levels of the prediction result is large, it is explained that the prediction accuracy is not high, and thus adding the term to the loss function can help to improve the prediction performance of the hierarchical time series prediction.

In the embodiment of the invention, the server calculates the prediction error and the error between each level respectively through the first prediction data and the second prediction data in the second prediction data set. In the process of calculating the error by combining the loss function, the server considers the error among all the levels, and further predicts the data more accurately through the prediction model after the model parameter adjustment of the loss function.

In some embodiments, referring to fig. 8, fig. 8 is a schematic flow chart of an alternative data prediction method provided in the embodiment of the present invention, S103 shown in fig. 3 may be implemented through S116 to S118, and the description will be made with reference to the steps.

S116, deleting the abnormal value of each level in the plurality of groups of data, and filling the abnormal value with average data of the level corresponding to the abnormal value.

In the embodiment of the invention, the server deletes the abnormal value of each level in the plurality of groups of data and fills the average data of the level corresponding to the abnormal value.

The average data is an average value of a plurality of data of the hierarchy corresponding to a plurality of historical time periods of the hierarchy.

And S117, filling the blank data corresponding to each level in the plurality of sets of data by using the average data corresponding to the level with the blank data, so as to obtain a plurality of sets of processed data corresponding to the time sequence of each level.

In the embodiment of the invention, the server fills the blank data corresponding to each level in the plurality of sets of data by using the average data corresponding to the level with the blank data, so as to obtain a plurality of sets of processed data corresponding to the time sequence of each level.

S118, determining a preset historical time period in a plurality of historical time periods, combining a plurality of groups of first processed data corresponding to a plurality of first time periods before the preset historical time period into a training set, and combining a plurality of groups of second processed data corresponding to a plurality of second time periods after the preset historical time period into a test set.

In the embodiment of the invention, a server determines a preset historical time period from a plurality of historical time periods, combines a plurality of groups of first processed data corresponding to a plurality of first time periods before the preset historical time period into a training set, and combines a plurality of groups of second processed data corresponding to a plurality of second time periods after the preset historical time period into a test set.

By way of example, the plurality of historical time periods may include 12 time periods corresponding to 1 month to 12 months. The server may determine 9 months as the preset history period. In the case that the server determines that 9 months are the preset historical time periods, the first time periods are 8 time periods corresponding to 1-8 months, and the second time periods are 3 time periods corresponding to 10-12 months.

In the embodiment of the invention, the server performs standardized processing on multiple groups of data, and removes abnormal values and fills blank data, so that the data structure of the multiple groups of data is more perfect, and model training is facilitated.

In some embodiments, referring to fig. 8, fig. 8 is a schematic flow chart of an alternative data prediction method provided in the embodiment of the present invention, and S105 shown in fig. 3 may be implemented through S119 to S122, and will be described in connection with the steps.

S119, comparing the plurality of groups of test data with the plurality of times of predicted data in the first predicted data set respectively, and determining a plurality of times of comparison errors corresponding to the plurality of times of predicted data.

In the embodiment of the invention, the server compares a plurality of groups of test data with a plurality of times of predicted data in the first predicted data set respectively, and determines a plurality of times of comparison errors corresponding to the plurality of times of predicted data.

In the embodiment of the invention, the server compares the test data of each level of each time period in a plurality of groups of test data with the corresponding prediction data in a certain prediction data of the first prediction data set. The server determines errors corresponding to the test data of each level in each time period. The server adds the errors corresponding to the test data of each level in each time period to obtain the errors corresponding to each time period, namely, the errors of each group of test data. And the server adds the errors corresponding to each group of test data to obtain the errors corresponding to the predicted data. Further, a plurality of comparison errors of the plurality of prediction data can be obtained.

S120, determining a target comparison error in a preset error range from the multiple comparison errors.

In the embodiment of the invention, the server determines the target comparison error within the preset error range from the multiple comparison errors.

S121, determining target iteration times corresponding to target sub-prediction data corresponding to target comparison errors.

In the embodiment of the invention, the server determines the target iteration times corresponding to the target secondary prediction data corresponding to the target comparison errors.

S122, determining a prediction data prediction model corresponding to the target iteration number in the plurality of prediction models.

In the embodiment of the invention, a plurality of prediction models are formed in the iterative process. And the server determines a preset data prediction model formed by corresponding target iteration times in the multiple prediction models.

In the embodiment of the invention, as a plurality of prediction models are formed in the iteration process, the server determines the preset data prediction model corresponding to the target iteration time with the minimum test data error, and the preset data prediction model has higher prediction precision on the test set, so that the layered time sequence data is processed through the preset data prediction model, and a prediction result with higher prediction precision can be obtained.

In some embodiments, referring to fig. 9, fig. 9 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention, and the steps will be described in connection with the following.

S123, acquiring a plurality of groups of logistics volume data corresponding to a plurality of historical time periods.

In the embodiment of the invention, a server acquires a plurality of sets of logistics cargo volume data corresponding to a plurality of historical time periods.

The plurality of sets of logistics cargo amount data comprise the national, regional and provincial cargo amount data and the hierarchical relationship.

S124, processing the multiple groups of logistics volume data by using a preset data prediction model to obtain predicted logistics volume data of a preset time period after multiple historical time periods.

In the embodiment of the invention, the server processes multiple sets of logistics cargo volume data by using a preset data prediction model to obtain predicted logistics cargo volume data of a preset time period after multiple historical time periods.

In the embodiment of the invention, the server processes a plurality of sets of logistics cargo volume data by using the preset data prediction model, and the preset data prediction model is obtained by training based on the prediction errors of a plurality of sets of training data in a historical preset time period and the errors among all levels. And then, the prediction model predicts a plurality of groups of logistics cargo volume data, so that a prediction result with higher precision can be obtained.

The embodiment of the invention also provides a logistics cargo amount prediction device 600, which is used for executing the data prediction method provided in fig. 9, please refer to fig. 10, which is a schematic structural diagram of the logistics cargo amount prediction device provided in the embodiment of the invention.

The embodiment of the invention provides a logistics cargo amount prediction device 600, which comprises a data acquisition module 601, a data preprocessing module 602, a target prediction model training module 603 and a data prediction model 604.

The data acquisition module 601 is configured to acquire a hierarchical relationship between the time series and the historical time series of the logistics cargo amount. Such as nationwide, regional, provincial shipping volume data and hierarchical relationships. The data acquisition module 601 is configured to execute S123.

The data preprocessing module 602 is configured to preprocess data, reject outliers and missing value padding, and normalize the data. The preprocessed data is then divided into training and testing sets.

The target prediction model training module 603 is configured to train the initial network model by using the historical time sequence data, so as to obtain a target prediction model of the time sequence data.

The data prediction module 604 is configured to predict data of the time-series data in a future time period by using a target prediction model to obtain a prediction result, and store and display the prediction result.

In some embodiments, referring to fig. 11, fig. 11 is a schematic flow chart of an alternative method for predicting data according to an embodiment of the present invention, and the steps will be described in connection with the following.

S201, collecting layered time sequence data to be predicted.

Illustratively, in conjunction with fig. 12, a data acquisition module 701 in the data prediction apparatus 700 is configured to acquire a hierarchical relationship between historical time series data and time series.

S202, preprocessing data, eliminating abnormal values and missing value filling, and segmenting data to divide a training set and a testing set.

The data preprocessing module 702 in the data prediction apparatus 700 is used for preprocessing the historical time series data, eliminating abnormal values and missing value filling, and normalizing the data. The preprocessed data is then divided into training and testing sets.

S203, constructing DeepAR a time sequence prediction model.

S204, training set input.

S205, setting hierarchical loss item super parameters.

S206, updating DeepAR model parameters by adopting a learning rate self-adaptive Adam optimization algorithm.

S207, whether training reaches the preset training times or not.

Illustratively, the target prediction model training module 703 in the data prediction apparatus 700 is configured to train the initial network model with the historical time series data to obtain a target prediction model of the time series data. I.e. the final model.

S208, taking the final model to output a future period prediction result.

The data prediction module 704 in the data prediction apparatus 700 is configured to predict, using a target prediction model (final model), data of time series data in a future time period to obtain a prediction result, and store and display the prediction result.

Because DeepAR time series prediction models are constructed based on the prediction errors of multiple groups of training data in the time series data in the historical time period and the errors among various layers, the accuracy of the prediction errors is considered and the errors among various layers are considered when the DeepAR time series prediction models are trained, so that the final model obtained through training predicts data more accurately.

An exemplary embodiment of the present invention further provides a data prediction apparatus 700 for executing the data prediction method provided in fig. 11, please refer to fig. 12, which is a schematic diagram of a structure of the data prediction apparatus provided in the embodiment of the present invention.

The embodiment of the invention provides a data prediction device 700, which comprises a data acquisition module 701, a data preprocessing module 702, a target prediction model training module 703 and a data prediction model 704.

The data acquisition module 701 is configured to acquire a hierarchical relationship between the historical time series data and the time series.

The data preprocessing module 702 is configured to preprocess the historical time series data, reject the outlier and missing value padding, and normalize the data. The preprocessed data is then divided into training and testing sets. The module is specifically in S202 in the above prediction method flow.

The target prediction model training module 703 is configured to train the initial network model by using the historical time sequence data to obtain a target prediction model of the time sequence data. The module details are S203 to S207 in the above prediction method flow.

And the data prediction module 704 is used for predicting the time sequence data in the future time period by utilizing the target prediction model to obtain a prediction result, and storing and displaying the prediction result. The module details are S208 in the above prediction method flow.

Referring to fig. 13, a second schematic structural diagram of the data prediction apparatus according to the embodiment of the present invention is shown.

The embodiment of the invention also provides a data prediction device 800, which comprises a data acquisition unit 803 and a prediction unit 804.

A data obtaining unit 803, configured to obtain hierarchical time series data, where the hierarchical time series data is a plurality of groups of data corresponding to each hierarchical time series, and a sum of data of child hierarchies of each hierarchical level is equal to data of a corresponding parent hierarchy;

A prediction unit 804, configured to predict the layered time series data by using a preset data prediction model, determine a prediction result in a preset time period after a plurality of historical time periods, where,

In the embodiment of the invention, the data prediction device 800 is used for performing standardization processing on multiple groups of data of layered time series data, dividing the multiple groups of standardized data into a training set and a testing set according to preset historical time periods, wherein the training set comprises multiple groups of training data, the testing set comprises multiple groups of testing data, a loss function of an initial prediction model is utilized to calculate prediction errors of the training set, errors among all levels, and model parameters of the initial prediction model are iteratively adjusted according to the prediction errors and the errors among all levels until training conditions are met, a first prediction data set corresponding to the testing set is obtained, the first prediction data set comprises multiple prediction data in an iterative process of all levels corresponding to all historical time periods of the testing set, and the multiple groups of testing data and the first prediction data set are utilized to conduct comparison, so that a preset data prediction model is determined.

In the embodiment of the invention, the data prediction device 800 is used for inputting multiple sets of training data into an initial prediction model to obtain a second prediction data set, wherein the second prediction data set comprises prediction data of each level of multiple historical time periods, calculating a prediction error and an error between each level based on the second prediction data set and multiple sets of training data in combination with a loss function, carrying out gradient solving on the loss function to obtain model parameters in an iteration process so as to obtain an updated prediction model, continuing training the multiple sets of training data by utilizing the updated prediction model until the training conditions are met, stopping to obtain a final prediction model so as to obtain multiple prediction models in the iteration process, and extracting the prediction data of each level of each historical time period of a corresponding test set in each corresponding second prediction data set by adopting the multiple prediction models so as to obtain a first prediction data set in the iteration process.

In the embodiment of the invention, the data prediction device 800 is configured to calculate a prediction error based on first prediction data in a second prediction data set and multiple sets of training data, where the first prediction data is prediction data of each level in multiple first time periods in the second prediction data set, the multiple first time periods are time periods before a preset history time period in multiple history time periods, calculate an error between each level based on second prediction data in the second prediction data set, the second prediction data is prediction data of each level in multiple second time periods in the second prediction data set, and the multiple second time periods are time periods after the preset history time period in the multiple history time periods.

In the embodiment of the present invention, the data prediction device 800 is configured to calculate the sum of squares of differences between the first prediction data and the training data of the corresponding level in the same first period, further obtain a first sum of levels in the same first period, and add a plurality of first sums corresponding to the plurality of first periods to obtain a prediction error.

In the embodiment of the present invention, the data prediction apparatus 800 is configured to calculate a sum of squares of differences between the prediction data of each parent level and the sum of prediction data of each corresponding child level in each layer of second prediction data in the same second time period, add the sums of squares of the second time periods to obtain a second sum, and multiply the second sums with a harmonic error penalty term superparameter to obtain an error between each level.

In the embodiment of the invention, the data prediction device 800 is configured to compare multiple sets of test data with multiple sets of predicted data in the first predicted data set, determine multiple comparison errors corresponding to the multiple sets of predicted data, determine a target comparison error within a preset error range from the multiple comparison errors, determine a target iteration number corresponding to target sub-predicted data corresponding to the target comparison error, and determine a predicted data prediction model corresponding to the target iteration number from the multiple prediction models.

In the embodiment of the invention, the plurality of sets of training data comprise a plurality of sets of first processed data, the plurality of sets of test data comprise a plurality of sets of second processed data, the data prediction device 800 is used for deleting abnormal values of all levels in the plurality of sets of data and filling the abnormal values with average data of the levels corresponding to the abnormal values, the average data of the levels corresponding to the blank data are used for filling the blank data corresponding to all levels in the plurality of sets of data, so as to obtain a plurality of sets of processed data corresponding to time sequences of all levels, a preset historical time period is determined in the plurality of historical time periods, a plurality of sets of first processed data corresponding to the plurality of first time periods before the preset historical time period are combined into a training set, and a plurality of sets of second processed data corresponding to the plurality of second time periods after the preset historical time period are combined into a test set.

In the embodiment of the present invention, the data obtaining unit 803 in the data predicting device 800 is configured to obtain multiple sets of logistics volume data corresponding to multiple historical time periods, and the predicting unit 804 in the data predicting device 800 is configured to process the multiple sets of logistics volume data by using a preset data predicting model, so as to obtain predicted logistics volume data in a preset time period after the multiple historical time periods.

In the embodiment of the invention, hierarchical time series data is acquired through a data acquisition unit 803, the hierarchical time series data is a plurality of groups of data corresponding to time series of each hierarchy, wherein the sum of data of sub-hierarchies of each hierarchy is equal to data of a corresponding father hierarchy, the hierarchical time series data is predicted through a prediction unit 804 by using a preset data prediction model, and a prediction result in a preset time period after a plurality of historical time periods is determined, wherein the preset data prediction model is obtained by training together based on prediction errors of a plurality of groups of training data in the historical preset time period in the hierarchical time series data and errors among the hierarchies. The preset data prediction model is obtained based on the prediction errors of a plurality of groups of training data in a historical time period and the errors among all the layers, and not only the accuracy of the prediction errors but also the errors among all the layers are considered during training, so that the preset data prediction model obtained through training is more accurate in data prediction.

It should be noted that, in the embodiment of the present invention, if the data prediction method is implemented in the form of a software functional module, and is sold or used as a separate product, the data prediction method may also be stored in a computer readable storage medium. Based on such understanding, the technical solution of the embodiments of the present invention may be embodied essentially or in a part contributing to the related art in the form of a software product stored in a storage medium, including several instructions for causing a data prediction device (which may be a personal computer or the like) to perform all or part of the methods described in the embodiments of the present invention. The storage medium includes various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read Only Memory (ROM), a magnetic disk, or an optical disk. Thus, embodiments of the invention are not limited to any specific combination of hardware and software.

Correspondingly, an embodiment of the invention provides a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the above-mentioned method.

Correspondingly, the embodiment of the invention provides a data prediction device, which comprises a memory 802 and a processor 801, wherein the memory 802 stores a computer program capable of running on the processor 801, and the processor 801 realizes the steps in the method when executing the program.

It should be noted here that the description of the storage medium and the device embodiments above is similar to the description of the method embodiments above, with similar advantageous effects as the method embodiments. For technical details not disclosed in the embodiments of the storage medium and the apparatus of the present invention, please refer to the description of the method embodiments of the present invention.

It should be noted that fig. 14 is a schematic diagram of a hardware entity of a data prediction apparatus according to an embodiment of the present invention, as shown in fig. 14, the hardware entity of the data prediction apparatus 800 includes a processor 801 and a memory 802, where;

the processor 801 generally controls the overall operation of the data prediction device 800.

The memory 802 is configured to store instructions and applications executable by the processor 801, and may also cache data (e.g., image data, audio data, voice communication data, and video communication data) to be processed or processed by each module in the processor 801 and the data prediction apparatus 800, which may be implemented by a FLASH memory (FLASH) or a random access memory (Random Access Memory, RAM).

It should be appreciated that reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. It should be understood that, in various embodiments of the present invention, the sequence numbers of the foregoing processes do not mean the order of execution, and the order of execution of the processes should be determined by the functions and internal logic thereof, and should not constitute any limitation on the implementation process of the embodiments of the present invention. The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

In the several embodiments provided by the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The above-described apparatus embodiments are merely illustrative, and the division of the units, for example, is merely a logical function division, and may be implemented in other manners, such as multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. In addition, the various components shown or discussed may be coupled or directly coupled or communicatively coupled to each other via some interface, device or unit, whether electrical, mechanical or otherwise.

The units described as separate components may or may not be physically separate, and components displayed as units may or may not be physical units, may be located in one place or distributed on a plurality of network units, and may select some or all of the units according to actual needs to achieve the purpose of the embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated in one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of hardware plus a form of software functional unit.

It will be appreciated by those of ordinary skill in the art that implementing all or part of the steps of the above method embodiments may be implemented by hardware associated with program instructions, where the above program may be stored in a computer readable storage medium, where the program when executed performs the steps comprising the above method embodiments, where the above storage medium includes various media capable of storing program code, such as a removable storage device, a Read Only Memory (ROM), a magnetic disk or an optical disk.

Or the above-described integrated units of the invention may be stored in a computer-readable storage medium if implemented in the form of software functional modules and sold or used as separate products. Based on such understanding, the technical solution of the embodiments of the present invention may be embodied essentially or in a part contributing to the related art in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the methods described in the embodiments of the present invention. The storage medium includes various media capable of storing program codes such as a removable storage device, a ROM, a magnetic disk or an optical disk.

The foregoing is merely an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily think about changes or substitutions within the technical scope of the present invention, and the changes and substitutions are intended to be covered by the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A data prediction method, comprising:

Obtaining hierarchical time series data; the hierarchical time series data is multiple sets of data corresponding to time series at each level, wherein the sum of the data of the sub-levels of each level in the level is equal to the data of the corresponding parent level; the levels include: regional level and agency level; the hierarchical time series data includes logistics cargo volume data of the corresponding level;

Using a preset data prediction model, the layered time series data is predicted to determine the prediction results within a preset time period after multiple historical time periods; wherein,

The preset data prediction model is obtained by training based on the prediction errors of multiple groups of training data within a historical preset time period in the layered time series data, as well as the errors between various levels.

2. The data prediction method according to claim 1, wherein before predicting the hierarchical time series data using a preset data prediction model and determining prediction results within a preset time period after a plurality of historical time periods, after obtaining the hierarchical time series data, the method further comprises:

Standardizing the multiple sets of data of the layered time series data, and dividing the standardized multiple sets of data into a training set and a test set according to a historical preset time period; the training set includes: multiple sets of training data; the test set includes: multiple sets of test data;

Calculating the prediction error of the training set and the error between each layer using the loss function of the initial prediction model, and iteratively adjusting the model parameters of the initial prediction model based on the prediction error and the error between each layer until the training condition is met, thereby obtaining a first prediction data set corresponding to the test set; the first prediction data set includes: multiple prediction data from the iterative process for each layer in each historical time period corresponding to the test set;

The plurality of test data sets are compared with the first prediction data set to determine the preset data prediction model.

3. The data prediction method according to claim 2, wherein the loss function of the initial prediction model is used to calculate the prediction error of the training set and the error between each layer, and the model parameters of the initial prediction model are iteratively adjusted according to the prediction error and the error between each layer until the training condition is met, thereby obtaining a first prediction data set corresponding to the test set, comprising:

Inputting the multiple sets of training data into the initial prediction model to obtain a second prediction data set; the second prediction data set includes: prediction data of each level of the multiple historical time periods;

Calculating the prediction error and the errors between the various levels based on the second prediction data set and the multiple sets of training data in combination with the loss function;

Performing gradient solving on the loss function to obtain model parameters in the iterative process, thereby obtaining an updated prediction model;

Using the updated prediction model, continue training the multiple sets of training data until the training conditions are met, and obtain a final prediction model, thereby obtaining multiple prediction models in the iterative process;

When using the multiple prediction models to obtain each corresponding second prediction data set, the prediction data of each level corresponding to each historical time period of the test set is extracted to obtain the first prediction data set in the iterative process.

4. The data prediction method according to claim 3, wherein the step of calculating the prediction error and the errors between the various levels based on the second prediction data set and the multiple sets of training data in combination with the loss function comprises:

Calculating the prediction error based on first prediction data in the second prediction data set and the multiple sets of training data; the first prediction data is prediction data of each level in multiple first time periods in the second prediction data set; the multiple first time periods are time periods before the historical preset time period in the multiple historical time periods;

Based on the second prediction data in the second prediction data set, the errors between the various levels are calculated; the second prediction data are the prediction data of each level in multiple second time periods in the second prediction data set; the multiple second time periods are the time periods after the historical preset time periods in the multiple historical time periods.

5. The data prediction method according to claim 4, wherein the step of calculating the prediction error based on the first prediction data in the second prediction data set and the plurality of training data sets comprises:

Calculate the sum of the squares of the differences between the first prediction data in the same first time period and the training data of the corresponding level, and then obtain the first sum of each level in the same first time period, and add the multiple first sums corresponding to the multiple first time periods to obtain the prediction error.

6. The data prediction method according to claim 4, wherein the step of calculating the errors between the various levels based on the second prediction data in the second prediction data set comprises:

Calculating the sum of squares of differences between the predicted data of each parent level of each layer in the second predicted data in the same second time period and the sum of the predicted data of each corresponding child level, and adding multiple square sums of the multiple second time periods to obtain a second sum;

The multiple second sums are multiplied by the harmonic error penalty term hyperparameter to obtain the errors between the various levels.

7. The data prediction method according to any one of claims 3 to 6, wherein comparing the plurality of test data sets with the first prediction data set to determine the preset data prediction model comprises:

Comparing the multiple test data sets with multiple prediction data in the first prediction data set, respectively, to determine multiple comparison errors corresponding to the multiple prediction data;

Determining a target comparison error within a preset error range from the multiple comparison errors;

Determine a target number of iterations corresponding to a target number of prediction data corresponding to the target comparison error;

The preset data prediction model corresponding to the target iteration number is determined among the multiple prediction models.

8. The data prediction method according to any one of claims 3 to 6, wherein the plurality of sets of training data include: a plurality of sets of first processed data; the plurality of sets of test data include: a plurality of sets of second processed data;

The step of normalizing the multiple sets of data of the layered time series data and dividing the normalized multiple sets of data into a training set and a test set according to a preset historical time period includes:

Deleting outliers at each level in the multiple sets of data and filling in the data with the average data of the level corresponding to the outliers;

Filling the blank data corresponding to each level in the plurality of sets of data with the average data corresponding to the level with blank data, thereby obtaining a plurality of sets of processed data corresponding to the time series of each level;

A historical preset time period is determined from the multiple historical time periods, and the multiple groups of first processed data combinations corresponding to the multiple first time periods before the historical preset time period are used as training sets, and the multiple groups of second processed data combinations corresponding to the multiple second time periods after the historical preset time period are used as test sets.

9. The data prediction method according to any one of claims 3 to 6, further comprising:

Obtaining multiple sets of logistics cargo volume data corresponding to the multiple historical time periods;

The plurality of groups of logistics cargo volume data are processed using the preset data prediction model to obtain predicted logistics cargo volume data for a preset time period after the plurality of historical time periods.

10. A data prediction device, comprising:

A data acquisition unit is configured to acquire hierarchical time series data; the hierarchical time series data is a plurality of sets of data corresponding to time series at each level, wherein the sum of the data of the sub-levels of each level in the hierarchical layer is equal to the data of the corresponding parent level; the levels include: regional level and agency level; the hierarchical time series data includes logistics cargo volume data of the corresponding level;

The prediction unit is used to predict the hierarchical time series data using a preset data prediction model to determine the prediction results within a preset time period after multiple historical time periods; wherein,

11. A data prediction device, characterized in that it comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the steps in the method according to any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.