CN118966478B

CN118966478B - Energy consumption prediction method and system based on green smart lighting for cultural tourism

Info

Publication number: CN118966478B
Application number: CN202411448243.1A
Authority: CN
Inventors: 邹岩坤
Original assignee: Liangye Technology Group Co ltd
Current assignee: Liangye Technology Group Co ltd
Priority date: 2024-10-17
Filing date: 2024-10-17
Publication date: 2025-01-21
Anticipated expiration: 2044-10-17
Also published as: CN118966478A

Abstract

The present invention provides a method and system for predicting energy consumption of green intelligent lighting based on culture and tourism, which relates to the technical field of energy consumption prediction, including obtaining historical lighting energy consumption data, real-time energy consumption data and influencing factor data of cultural and tourism scenes, and constructing a deep probability graph model based on the preprocessed influencing factor data set using a variational inference algorithm, combining the pre-trained lighting energy consumption model with the deep probability graph model to obtain a prediction model adapted to the current scene; using a graph neural network algorithm, taking the topological structure information, functional zoning information and lighting equipment characteristic information of the cultural and tourism scene as input, constructing a scene knowledge graph, generating energy consumption prediction values through a multi-scale prediction algorithm, and generating green intelligent lighting control strategies based on the energy consumption prediction values and in combination with the lighting equipment characteristic information in the scene knowledge graph using an intelligent decision-making algorithm, applying the generated green intelligent lighting control strategies to the lighting systems of cultural and tourism scenes to achieve intelligent control of lighting equipment.

Description

Green intelligent illumination energy consumption prediction method and system based on travel

Technical Field

The invention relates to an energy consumption prediction technology, in particular to a green intelligent illumination energy consumption prediction method and system based on a hotel.

Background

In the modern travel industry, management of illumination energy consumption is becoming a key factor for improving resource utilization efficiency and realizing sustainable development. Travel scenes often have complex functional compartments and dynamic guest flows, which place higher demands on the design and regulation of lighting systems. The existing illumination energy consumption prediction method is mostly dependent on a simple statistical model, and lacks depth excavation and dynamic adaptation capability to influencing factors. Therefore, how to effectively integrate the historical energy consumption data, the real-time energy consumption data and the multidimensional influence factor data, and accurately predict the historical energy consumption data, the real-time energy consumption data and the multidimensional influence factor data through advanced machine learning and graph neural network technology becomes a problem to be solved urgently in the industry.

Disclosure of Invention

The embodiment of the invention provides a green intelligent illumination energy consumption prediction method and system based on a hotel, which can solve the problems in the prior art.

In a first aspect of an embodiment of the present invention,

The utility model provides a green wisdom illumination energy consumption prediction method based on travel, including:

acquiring historical illumination energy consumption data, real-time energy consumption data and influence factor data of a travel scene, preprocessing the data by utilizing a data cleaning algorithm, an interpolation algorithm and an anomaly detection algorithm to obtain a preprocessed energy consumption data set and an influence factor data set, extracting features by adopting a time sequence decomposition algorithm based on the preprocessed historical illumination energy consumption data, and generating an energy consumption time sequence feature set;

According to the preprocessed influence factor data set, weighting calculation is carried out on influence factors of different time scales and space scales by utilizing a multi-layer attention network to obtain a weighted influence factor feature set, the energy-consuming time sequence feature set and the weighted influence factor feature set are input into a multi-mode feature fusion network to generate unified feature representation, a depth probability map model is built by utilizing a variation inference algorithm based on the unified feature representation, and a pre-trained illumination energy consumption model is combined with the depth probability map model to obtain a prediction model adapting to a current scene;

The method comprises the steps of using a graph neural network algorithm, taking topological structure information, functional partition information and lighting equipment characteristic information of a traveling scene as inputs, constructing a scene knowledge graph, taking a prediction model adapting to the current scene and the scene knowledge graph as inputs, generating an energy consumption prediction value through a multi-scale prediction algorithm, generating a green intelligent lighting regulation strategy through an intelligent decision algorithm based on the energy consumption prediction value and combining the lighting equipment characteristic information in the scene knowledge graph, and applying the generated green intelligent lighting regulation strategy to a lighting system of the traveling scene to realize intelligent control of lighting equipment.

In an alternative embodiment of the present invention,

Based on unified feature representation, constructing a depth probability map model by using a variation inference algorithm, combining the pre-trained illumination energy consumption model with the depth probability map model, and obtaining a prediction model adapting to the current scene comprises the following steps:

based on unified feature representation, constructing a depth probability map model comprising hidden variables and observed variables, inputting the hidden variables and context information related to the observed variables into a multi-layer perceptron, calculating to obtain attention weight vectors corresponding to each observed variable, carrying out weighted summation on the hidden variables by using the attention weight vectors, and generating the observed variables through nonlinear activation function processing;

Taking the generated observation variable as the mean value of Gaussian distribution, simultaneously utilizing a neural network to learn to obtain distribution variance, defining probability distribution of the observation variable, multiplying prior distribution of hidden variable with conditional probability distribution of all the observation variable, constructing joint distribution of a depth probability map model, defining a log likelihood function based on the constructed joint distribution, simultaneously selecting standard normal distribution as basic distribution, designing a plurality of reversible neural network layers comprising a standardization layer, a reversible mixing layer and an affine coupling layer, and converting the basic distribution through the reversible neural network layer to construct variation distribution;

Deducing a variation lower bound expression by using the constructed variation distribution and the combined distribution, calculating a variation lower bound, taking the calculated variation lower bound as an optimization target, acquiring a pre-trained large-scale illumination energy consumption model, combining the pre-trained large-scale illumination energy consumption model with a depth probability map model, constructing an initial combination prediction model, and inputting a unified feature representation into the initial combination prediction model to obtain a prediction result;

Calculating the prediction loss based on the prediction result and the actual energy consumption data, constructing an overall optimization objective function by combining the prediction loss and the calculated variation lower bound, adopting a random gradient descent optimization algorithm based on the overall optimization objective function, iteratively updating parameters of the combined prediction model, and repeating iteration until the maximum iteration times are met, so as to obtain the final prediction model adapting to the current scene.

In an alternative embodiment of the present invention,

The calculation formula for constructing the overall optimization objective function by combining the predicted loss and the calculated variation lower bound is as follows:

;

Where L represents the overall optimization objective function, N represents the number of data samples, Y _n represents the true value of the nth observed variable, Represents the predicted value of the nth observation variable, λ represents the weight parameter, E _q(Z|Y) logp (z|y) represents the expected value of the log likelihood, q (z|y) represents the posterior distribution of the hidden variable Z given the observation variable Y, D _KL (·) represents the KL divergence, p (z|y) represents the probability of the hidden variable Z given the observation variable Y, and p (Z) represents the prior distribution.

In an alternative embodiment of the present invention,

Using a graph neural network algorithm, taking topology structure information, functional partition information and lighting equipment characteristic information of a travel scene as inputs, constructing a scene knowledge graph, taking a prediction model adapting to a current scene and the scene knowledge graph as inputs, and generating an energy consumption predicted value through a multi-scale prediction algorithm comprises:

Acquiring topology structure information, functional partition information and lighting equipment characteristic information of a travel scene as input data, utilizing a graph neural network algorithm, introducing an attention mechanism, adaptively aggregating characteristic information of neighbor nodes by learning importance weights of different neighbor nodes, introducing convolution operation at each layer of the graph neural network, extracting high-order characteristic representation of the nodes by characteristic aggregation and transformation of local neighborhood, and generating a scene knowledge graph containing scene structural representation and semantic information;

Acquiring historical energy consumption data and influence factor data of a travel scene, inputting a prediction model adapting to the current scene by combining a scene knowledge graph, and decomposing an output result of the prediction model in a time dimension by a multi-scale prediction algorithm to obtain short-term, medium-term and long-term prediction subsequences respectively;

The method comprises the steps of combining structural representation and semantic information in a scene knowledge graph, applying a sequence prediction model based on an attention mechanism to a short-term prediction subsequence to generate a short-term illumination energy consumption prediction result, fusing functional partition information in the scene knowledge graph, applying a prediction model based on time sequence decomposition and periodic detection to a middle-term prediction subsequence to generate a middle-term illumination energy consumption prediction result, integrating topological structure information in the scene knowledge graph, applying a prediction model based on causal inference and scene analysis to a long-term prediction subsequence to generate a long-term illumination energy consumption prediction result;

And introducing a self-adaptive multi-scale fusion mechanism, and self-adaptively fusing the prediction results of different scales by learning the importance weights of the short-term, medium-term and long-term illumination energy consumption prediction results to obtain a final energy consumption prediction value.

In an alternative embodiment of the present invention,

Applying a causal inference and scene analysis based prediction model to the long-term prediction subsequence, generating a long-term illumination energy consumption prediction result includes:

acquiring historical illumination energy consumption data, scenario factor data and a long-term prediction subsequence of a travel scene, wherein the scenario factor data comprises weather conditions, time characteristics and energy-saving measure implementation conditions;

Performing missing value filling and abnormal value processing on the historical illumination energy consumption data, performing standardized processing on the processed historical illumination energy consumption data and scene factor data to obtain preprocessed input data, inputting the preprocessed input data into a pre-constructed countermeasure generation network model, wherein the generator model adopts a long and short time memory network structure, and the discriminator model adopts a convolutional neural network structure, and optimizing the generator and the discriminator in an alternate training mode to obtain a trained countermeasure generation network model;

Constructing a causal graph model based on a trained generator for generating a network model by antagonizing, taking a scenario factor as a node, determining causal relation among the nodes through a structure learning algorithm, quantifying causal strength among the nodes by using the causal graph model and applying a structural equation model to obtain a causal relation quantification result, integrating the causal relation quantification result into the generator model, and finally obtaining a causal enhanced long-term illumination energy consumption prediction model;

generating a plurality of groups of scene factor combinations based on historical data distribution and expert knowledge, arranging the plurality of groups of scene factor combinations into a scene matrix, wherein each row represents a future scene, each column corresponds to one scene factor, inputting a long-term prediction subsequence and the scene matrix into a causally enhanced long-term illumination energy consumption prediction model, and generating long-term illumination energy consumption prediction results under a plurality of scenes.

In an alternative embodiment of the present invention,

Based on the energy consumption predicted value, combining the lighting equipment characteristic information in the scene knowledge graph, generating the green intelligent lighting regulation strategy by using the intelligent decision algorithm comprises the following steps:

extracting characteristics and relation information of lighting equipment in a scene knowledge graph to form a decision support knowledge base, and constructing an initial decision model of multi-objective optimization based on the decision support knowledge base and an obtained energy consumption predicted value, wherein an objective function comprises energy consumption minimization, user comfort maximization and equipment service life maximization;

Designing an intelligent optimization algorithm based on reinforcement learning, taking historical energy consumption data, real-time environment parameters and user demand data as state input, taking lighting control parameters as action output, solving the initial decision model to generate an initial regulation strategy, simultaneously developing a simulation platform of a virtual model of a lighting system, inputting the initial regulation strategy into the simulation platform, performing first-round simulation, and generating a first-round simulation result data set;

The method comprises the steps of designing a multi-dimensional evaluation index system, comprehensively evaluating a first round of simulation result data set to obtain a first round of evaluation result, optimizing an initial decision model by using a self-adaptive parameter adjustment algorithm based on the first round of evaluation result to generate an optimized decision model, applying the optimized decision model to a simulation platform, performing multi-round iterative simulation, performing strategy evaluation based on evaluation indexes after each round of simulation, and correspondingly adjusting the optimized decision model until a preset termination condition is met to obtain a final decision model and a corresponding green intelligent lighting regulation strategy.

In an alternative embodiment of the present invention,

The calculation formula of the objective function is as follows:

;

Wherein F (x) represents an objective function, β represents a temperature parameter, r _i (t) represents a prize value for an ith target in a t-th step decision, r _j (t) represents a prize value for a j-th target in a t-th step decision, F _i (x) represents an ith optimization target, wherein F ₁ (x) represents an energy consumption minimization target, F ₂ (x) represents a user comfort maximization target, F ₃ (x) represents a device lifetime maximization target, Representing the ideal point of the i-th object,Representing the negative ideal point of the i-th object.

In a second aspect of an embodiment of the present invention,

Provided is a green intelligent illumination energy consumption prediction system based on a hotel, which comprises:

The first unit is used for acquiring historical illumination energy consumption data, real-time energy consumption data and influence factor data of a travel scene, preprocessing the data by utilizing a data cleaning algorithm, an interpolation algorithm and an anomaly detection algorithm to obtain a preprocessed energy consumption data set and an influence factor data set, extracting features by adopting a time sequence decomposition algorithm based on the preprocessed historical illumination energy consumption data, and generating an energy time sequence feature set;

The second unit is used for carrying out weight calculation on influence factors of different time scales and space scales by utilizing a multi-layer attention network according to the preprocessed influence factor data set to obtain a weighted influence factor feature set, inputting the energy-consuming time sequence feature set and the weighted influence factor feature set into a multi-mode feature fusion network to generate a unified feature representation, constructing a depth probability map model by utilizing a variation inference algorithm based on the unified feature representation, and combining the pre-trained illumination energy consumption model with the depth probability map model to obtain a prediction model adapting to the current scene;

The third unit is used for using a graph neural network algorithm, taking topological structure information, functional partition information and lighting equipment characteristic information of a travel scene as inputs, constructing a scene knowledge graph, taking a prediction model adapting to the current scene and the scene knowledge graph as inputs, generating an energy consumption prediction value through a multi-scale prediction algorithm, generating a green intelligent lighting regulation strategy through an intelligent decision algorithm based on the energy consumption prediction value and combining the lighting equipment characteristic information in the scene knowledge graph, and applying the generated green intelligent lighting regulation strategy to a lighting system of the travel scene to realize intelligent control of lighting equipment.

In a third aspect of an embodiment of the present invention,

There is provided an electronic device including:

A processor;

a memory for storing processor-executable instructions;

Wherein the processor is configured to invoke the instructions stored in the memory to perform the method described previously.

In a fourth aspect of an embodiment of the present invention,

There is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the method as described above.

In the embodiment, by introducing multi-source heterogeneous data and adopting algorithms such as time sequence decomposition, multi-layer attention network and the like to perform feature extraction and fusion, the time-space correlation and the action mechanism of influencing factors contained in the data can be fully mined, and the accuracy of energy consumption prediction is improved. The pre-trained large-scale illumination energy consumption model is utilized and combined with the depth probability map model constructed for the current scene, knowledge migration can be carried out among different scenes, and generalization capability and robustness of the model are improved. Through the multi-scale prediction algorithm, short-term, medium-term and long-term energy consumption prediction results can be generated, and decision requirements under different time scales are met, such as short-term prediction is used for real-time scheduling, and long-term prediction is used for energy planning and the like. Based on the energy consumption prediction result and the scene knowledge graph, an optimal illumination control strategy can be generated by utilizing an intelligent decision algorithm, so that the illumination quality is ensured, the energy consumption is minimized, and the green intelligent illumination is realized.

Drawings

FIG. 1 is a flow chart of a method for predicting energy consumption of intelligent illumination based on green travel in accordance with an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a system for predicting energy consumption of intelligent illumination based on green travel according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The technical scheme of the invention is described in detail below by specific examples. The following embodiments may be combined with each other, and some embodiments may not be repeated for the same or similar concepts or processes.

Fig. 1 is a flow chart of a method for predicting energy consumption of intelligent illumination based on green travel according to an embodiment of the present invention, as shown in fig. 1, the method includes:

S101, acquiring historical illumination energy consumption data, real-time energy consumption data and influence factor data of a travel scene, preprocessing the data by using a data cleaning algorithm, an interpolation algorithm and an anomaly detection algorithm to obtain a preprocessed energy consumption data set and an influence factor data set, extracting features by using a time sequence decomposition algorithm based on the preprocessed historical illumination energy consumption data, and generating an energy time sequence feature set.

The historical lighting energy consumption data comprise lighting equipment electricity consumption data under different time granularities, the real-time energy consumption data are lighting equipment electricity consumption data at the current moment, and the influence factor data comprise meteorological data, tourist number data, holiday data and the like. And cleaning the original data by adopting a data cleaning algorithm, and removing abnormal values and missing values. For missing values, interpolation algorithms are used for filling, common interpolation algorithms include linear interpolation, spline interpolation, and the like. Outlier detection may use statistical-based methods such as 3σ principle, box-plot methods, etc. And obtaining the preprocessed energy consumption data set and the influence factor data set through data preprocessing.

Next, based on the preprocessed historical illumination energy consumption data, a time series decomposition algorithm is adopted to extract features, and an energy-consuming time series feature set is generated. The time sequence decomposition algorithm can decompose the original time sequence data into trend items, seasonal items and random items, wherein the trend items reflect long-term change trend of the data, the seasonal items reflect periodic change rules of the data, and the random items are residual items after the trend items and the seasonal items are removed. Common time series decomposition algorithms include STL decomposition, HP filtering, and the like.

In the embodiment, by introducing multi-source heterogeneous data and adopting algorithms such as time sequence decomposition, multi-layer attention network and the like to perform feature extraction and fusion, the time-space correlation and the action mechanism of influencing factors contained in the data can be fully mined, and the accuracy of energy consumption prediction is improved. In addition, through the analysis of the influence factors, a manager can better predict and allocate resources, and the overall energy efficiency is improved. These effects not only promote sustainable development of travel scenes, but also provide scientific basis for formulation of related policies.

S102, according to the preprocessed influence factor data set, weighting calculation is carried out on influence factors of different time scales and space scales by using a multi-layer attention network to obtain a weighted influence factor feature set, the energy-consuming time sequence feature set and the weighted influence factor feature set are input into a multi-mode feature fusion network to generate unified feature representation, a depth probability map model is built by using a variation inference algorithm based on the unified feature representation, and a pre-trained illumination energy consumption model is combined with the depth probability map model to obtain a prediction model suitable for a current scene.

First, the influencing factor data set includes a variety of variables that may influence the illumination energy consumption, such as weather, time, activity type, etc. These influencing factors may exhibit different influence intensities on different time scales (e.g. hours, days, weeks) and spatial scales (e.g. different locations, regions). Thus, the use of a multi-layer attention network can effectively capture the importance of these factors and calculate their specific contribution to energy consumption. The core of the attention mechanism is that it can dynamically adjust weights according to the characteristics of the input data, thereby highlighting important factors and suppressing unimportant factors. The preprocessed influencing factor data is input into the multi-layer attention network. The network calculates the weight of each influence factor through the operation of a plurality of layers, and generates a weighted influence factor characteristic set. This feature set effectively reflects the impact of factors on energy consumption under different temporal and spatial conditions.

And then, combining the weighted influence factor feature set with the energy-consuming time sequence feature set, and inputting the multi-mode feature fusion network. The goal of the network is to integrate data from different sources into a unified feature representation that can capture the dynamic features of the time series and the weighted effects of the spatial influencing factors simultaneously. By fusing, a more comprehensive, representative feature set can be obtained.

In actual operation, it is assumed that data of influence factors such as weather of a certain travel scene, tourist traffic and the like are collected. Through the multi-layer attention network, the influence weight of weather factors on energy consumption in different seasons can be found to be obviously higher than the flow of tourists. After weighting, the features are fused into a unified feature representation, and then the parameters of the model are adjusted by combining the existing illumination energy consumption model and applying a variation inference algorithm, so that a new illumination energy consumption prediction model which is suitable for the current scene is finally formed.

In an alternative embodiment of the present invention,

Illustratively, a depth probability map model is first constructed that contains hidden and observed variables based on a unified feature representation. In this model, hidden variables represent potential modes of energy consumption and implicit roles of influencing factors, and observed variables represent energy consumption characteristics and influencing factor characteristics. In order to capture the dynamic dependency between the observed variables, an attention mechanism is introduced to calculate the attention weight vector corresponding to each observed variable. Specifically, hidden variables and context information (such as time, spatial position, etc.) related to the observed variables are input into a multi-layer perceptron, and an attention weight vector is calculated through nonlinear transformation and softmax functions. The hidden variables are then weighted summed using the attention weight vector and processed by a nonlinear activation function (e.g., reLU) to generate the observed variables. The generating process based on the attention mechanism can adaptively adjust the influence of hidden variables on different observed variables, and improves the expression capacity of the model.

When defining the probability distribution of the observed variable, taking the generated observed variable as the mean value of the Gaussian distribution, and learning by using a neural network to obtain the variance of the distribution. Thus, the conditional probability distribution of the observation variable can be expressed as a gaussian distribution of the learned variance with the generated value as the mean. Meanwhile, the prior distribution of hidden variables is assumed to be a standard normal distribution. The joint distribution of the depth probability map model is constructed by multiplying the prior distribution of hidden variables with the conditional probability distribution of all observed variables.

For model inference and learning, a variation inference method is adopted. First, a log likelihood function, i.e., the log probability of the observed variable, is defined based on the constructed joint distribution. However, it is difficult to directly calculate the log likelihood function due to the presence of hidden variables. Thus, a variational distribution is introduced to approximate the posterior distribution of the hidden variables. In order to construct a more flexible and accurate variational distribution, a standard normal distribution is selected as a base distribution, and then a variational distribution generation process comprising a plurality of reversible neural network layers is designed. Specifically, the input is normalized using a normalization layer (Normalization Layer), and then transformed by a reversible hybrid layer (Invertible Mixing Layer) and an affine coupling layer (Affine Coupling Layer). Through the combination of these reversible neural network layers, the base distribution is converted into a complex variational distribution to better fit the true posterior distribution of hidden variables.

In the learning process of the model, the constructed variation distribution and the joint distribution are utilized to deduce the expression of the variation lower bound (Evidence Lower Bound, ELBO). The lower variation bound is a lower bound of the log-likelihood function, consisting of the KL divergence between the variation distribution and the joint distribution and the expected log-likelihood of the observed variable under the variation distribution. By maximizing the lower bound of variation, the difference between the variation distribution and the true posterior distribution can be minimized simultaneously, and the log likelihood of the observed variable can be maximized.

To further improve the predictive performance, a pre-trained large-scale illumination energy consumption model was introduced. And combining the pre-training model with the depth probability map model to construct an initial combined prediction model. Specifically, the unified feature representation is input into an initial combined prediction model, and a prediction result of illumination energy consumption is obtained through feature extraction of a pre-training model and an inference process of a depth probability map model.

In the optimization process of the model, a prediction loss (such as an average absolute error or a mean square error) is calculated based on the prediction result and the actual energy consumption data. Meanwhile, the total optimization objective function is constructed by combining the predicted loss and the variation lower bound obtained through the previous calculation. The objective function takes into account both the predictive performance and the model's generative power and inferred quality. And finally, adopting a random gradient descent optimization algorithm to iteratively update parameters of the combined prediction model based on the overall optimization objective function. And repeating the iteration process until the preset maximum iteration times are reached or other termination conditions are met, and obtaining a final prediction model adapting to the current scene.

In an alternative embodiment of the present invention,

;

In this embodiment, by introducing an attention mechanism, the influence of hidden variables on different observation variables is adaptively adjusted, so as to improve the expressive power and interpretation of the model. And the reversible neural network layer is adopted to construct variation distribution, so that flexibility and accuracy of variation inference are improved. And a pre-training model is introduced, and the prediction performance and generalization capability are improved through transfer learning. And constructing an overall optimization objective function, considering the prediction performance and the generation capacity of the model, and improving the comprehensive performance of the model. The accuracy and the reliability of energy consumption prediction are improved, the adaptability of the model to different situations is enhanced, and a decision maker is helped to make a more effective energy management strategy. In addition, the deep learning characteristics of the model can also learn and optimize itself so that future adjustments and adaptations can be made more quickly as new data arrives. The method is not only helpful for realizing reasonable utilization of resources, but also can promote sustainable development of the travel scene.

S103, using a graph neural network algorithm, taking topological structure information, functional partition information and lighting equipment characteristic information of a travel scene as inputs, constructing a scene knowledge graph, taking a prediction model adapting to the current scene and the scene knowledge graph as inputs, generating an energy consumption prediction value through a multi-scale prediction algorithm, generating a green intelligent lighting regulation strategy by using an intelligent decision algorithm based on the energy consumption prediction value and combining the lighting equipment characteristic information in the scene knowledge graph, and applying the generated green intelligent lighting regulation strategy to a lighting system of the travel scene to realize intelligent control of lighting equipment.

First, topology information, functional partition information, and lighting device characteristic information of a travel scene are the basis for constructing a scene knowledge graph. The topology information reflects the spatial relationship between the elements (e.g. buildings, paths, etc.) in the scene, the functional partition information describes the functional use (e.g. leisure area, exhibition area, etc.) of the areas in the scene, and the lighting device characteristic information contains technical parameters (e.g. power, light intensity, energy consumption efficiency, etc.) of the lighting devices. The information is organized in the form of a graph to form a multidimensional knowledge graph, and a structural basis is provided for subsequent data analysis and prediction.

And then, taking a scene knowledge graph as input, and analyzing through a graph neural network in combination with an energy consumption prediction model adapting to the current scene. The graph neural network is a deep learning model specially processing graph structure data, and can effectively capture complex relations between nodes (such as lighting equipment). Through the process, the model can learn the association characteristics among different nodes from the atlas, so that the accuracy of energy consumption prediction is improved.

And after the energy consumption predicted value is obtained, generating a green intelligent lighting regulation strategy by using an intelligent decision algorithm based on the values and the lighting equipment characteristic information in the scene knowledge graph. The intelligent decision algorithm aims at making a reasonable lighting regulation and control scheme according to the predicted energy consumption value and the existing equipment characteristics. This may include adjusting the brightness, switching state, or timing control of the device to achieve optimal utilization of energy and reduce carbon emissions.

It is assumed that in a certain travel scenario, the topology shows guest flow conditions for different areas, while the functional partition information shows a high usage of a certain area for a certain period of time. Through the graph neural network, it may be found that the lighting demand of the area fluctuates with the change of the tourist flow, thereby generating the predicted value. If the prediction shows that the energy consumption of the area increases significantly during the eight to nine night, the intelligent decision algorithm may make a strategy to enhance the illumination during this period and gradually decrease the brightness after the tourist is reduced, thereby achieving the goal of energy saving.

In an alternative embodiment of the present invention,

For example, topology information, functional partition information, and lighting device characteristic information of a travel scene are first acquired as input data. The topology structure information describes the connection relation between each area in the scene and the equipment, the functional partition information reflects the functional attributes and purposes of the different areas in the scene, and the lighting equipment characteristic information comprises the type, power, light source parameters and the like of the lamp.

The input data is modeled by using a graph neural network algorithm, and the scene is abstracted into a graph structure, wherein nodes represent areas or devices in the scene, and edges represent association relations between the areas or devices. Attention mechanisms are introduced in the message passing and aggregation operation of the graph neural network, and feature information of neighbor nodes is adaptively aggregated by learning importance weights of different neighbor nodes. Meanwhile, convolution operation is introduced into each layer of the graph neural network, and high-order characteristic representation of the nodes is extracted through characteristic aggregation and transformation of local neighborhood. And generating a scene knowledge graph containing scene structural representation and semantic information through learning and reasoning of the multi-layer graph neural network.

And acquiring historical energy consumption data and influence factor data of a travel scene, such as weather, temperature, humidity, flow of people and the like, and inputting a prediction model adapting to the current scene by combining a scene knowledge graph. And decomposing the output result of the prediction model in the time dimension by a multi-scale prediction algorithm to obtain short-term, medium-term and long-term prediction subsequences respectively. Short-term prediction focuses on energy consumption changes from several hours to one day in the future, mid-term prediction focuses on energy consumption changes from several days to several weeks in the future, and long-term prediction focuses on energy consumption changes from several months to one year in the future.

In combination with the structured representation and semantic information in the scene knowledge graph, a sequence prediction model based on an attention mechanism is applied to the short-term predictor sequence. The model dynamically focuses on key time steps in historical energy consumption data through a focusing mechanism, captures the characteristics and modes of short-term energy consumption change and generates a short-term illumination energy consumption prediction result. And fusing the functional partition information in the scene knowledge graph, and applying a prediction model based on time sequence decomposition and periodic detection to the middle-period prediction subsequence. The model decomposes the mid-term energy consumption data into trend items, period items and residual items, predicts the mid-term illumination energy consumption by analyzing the energy consumption periodicity rules of different functional areas and combining scene knowledge, and generates mid-term illumination energy consumption prediction results. Integrating topological structure information in the scene knowledge graph, and applying a prediction model based on causal inference and scene analysis to the long-term prediction subsequence. The model utilizes the topological structure and the association relation of the scene to infer causal dependency between different areas and devices, and predicts the change trend of the long-term illumination energy consumption by combining scene analysis and hypothesis verification.

And introducing a self-adaptive multi-scale fusion mechanism, and self-adaptively fusing the prediction results of different scales by learning importance weights of the short-term, medium-term and long-term illumination energy consumption prediction results. And dynamically adjusting the weight of each time scale prediction result by adopting an attention mechanism, wherein the calculation of the weight considers the confidence coefficient, the historical performance, the scene characteristics and other factors of the prediction result, and finally obtaining the fused energy consumption prediction value.

In the embodiment, the structural information and the semantic information of the scene are fully utilized by fusing the graph neural network and the knowledge graph, so that the accuracy and the interpretability of the energy consumption prediction are improved. And an adaptive prediction model is designed according to the short-term, medium-term and long-term energy consumption change characteristics by adopting a multi-scale prediction method, so that the time scale coverage range and the adaptability of prediction are improved. Attention mechanism and self-adaptive fusion strategy are introduced, key information and prediction results of different scales are dynamically focused, and self-adaptive capacity and robustness of a prediction model are improved. And combining causal inference and scene analysis, considering the influence of scene internal factors and external environment factors on energy consumption, and enhancing the interpretability and decision support capability of the prediction result.

In an alternative embodiment of the present invention,

Illustratively, the historical illumination energy consumption data is first preprocessed, including missing value padding, outlier processing, and data normalization, to obtain preprocessed input data. And then inputting the preprocessed input data into a pre-constructed countermeasure generation network model for training, wherein the generator model adopts a long and short time memory network structure, the discriminator model adopts a convolutional neural network structure, and the generator and the discriminator are optimized in an alternate training mode, so that the trained countermeasure generation network model is finally obtained.

On the basis, a trained generator for generating a network model by using the antagonism is further utilized to construct a causal graph model, scene factors are taken as nodes, and causal relations among the nodes are determined through a structure learning algorithm. The structure learning algorithm deduces the causal relationship among the nodes by analyzing the condition independence and the dependency among the nodes, and generates a causal graph. Common structure learning algorithms include constraint-based algorithms, score-based algorithms, bayesian-based algorithms, and the like. And carrying out causal structure learning by adopting a constraint-based PC algorithm, wherein the algorithm gradually deletes edges in the causal graph through condition independence test, and finally, the causal graph meeting all condition independence constraints is obtained.

After the causal graph is obtained, the causal intensity among the nodes is quantified by using a structural equation model. The structural equation model is a mathematical model describing causal relationships between variables that represents each node as a linear combination of its parent nodes and introduces noise terms to represent the effects of unobserved factors. By estimating coefficients in the structural equation model, causal intensities between nodes can be quantified. And carrying out parameter estimation on the structural equation model by adopting a maximum likelihood estimation method to obtain a causal relationship quantification result.

Next, causal relationship quantification results are integrated into the generator model, resulting in causally enhanced long-term illumination energy consumption prediction models. Specifically, the causal relationship quantification result is used as priori knowledge and is introduced into a loss function of the generator model, and the generator model is constrained in a regularization term mode, so that the generated data accords with the causal relationship constraint. Thus, the generator model not only can learn the distribution characteristics of the historical data, but also can consider the causal relationship among the scene factors, and improves the accuracy and the interpretability of prediction.

Finally, in order to generate long-term illumination energy consumption prediction results under a plurality of scenes, a plurality of groups of scene factor combinations are generated based on historical data distribution and expert knowledge, and are arranged into a scene matrix, each row represents a future scene, and each column corresponds to one scene factor. And then, inputting the long-term prediction subsequence and the scene matrix into a causally enhanced long-term illumination energy consumption prediction model to generate a long-term illumination energy consumption prediction result under the corresponding scene. Thus, the prediction results under a plurality of scenes can be obtained, and a more comprehensive reference is provided for decision making.

A set of historical illumination energy consumption data is assumed, including daily illumination energy consumption values and corresponding situational factors such as weather, temperature, holidays, etc. Firstly, preprocessing the group of data, filling the missing values, removing the abnormal values, and performing standardization processing to obtain preprocessed input data. And then, inputting the preprocessed input data into a pre-constructed countermeasure generation network model for training, and obtaining a trained countermeasure generation network model through alternate optimization of a generator and a discriminator.

Next, a causal graph model is constructed using the trained generator model. And taking the scenario factors such as weather, temperature, holidays and the like as nodes, and carrying out causal structure learning through a PC algorithm to obtain a causal relation graph among the scenario factors. And then quantifying the causal relationship by using a structural equation model, and estimating the causal strength between each node and the father node. And integrating the causal relationship quantification result into a generator model to obtain a causal enhanced long-term illumination energy consumption prediction model.

Finally, according to the historical data distribution and expert knowledge, a plurality of groups of scene factor combinations are generated, such as different weather conditions, temperature ranges, holiday schedules and the like, and are arranged into a scene matrix. And inputting the long-term prediction subsequence (such as a time sequence of one month in the future) and the scene matrix into a causally enhanced long-term illumination energy consumption prediction model to generate a long-term illumination energy consumption prediction result under the corresponding scene. Therefore, the predicted value of the illumination energy consumption under different conditions can be obtained, and a basis is provided for the optimal control of the illumination system.

In the embodiment, the historical illumination energy consumption data is preprocessed, and the countermeasures are utilized to train the network model, so that the distribution characteristics of the historical data can be effectively learned, the expression capacity and the generalization capacity of the prediction model are improved, and the accuracy of long-term illumination energy consumption prediction is improved. By introducing the causal graph model and the structural equation model, causal relation among scene factors can be explicitly modeled, causal strength is quantized, so that a prediction model can not only give a prediction result, but also explain reasons behind the prediction result, the interpretability of the prediction result is enhanced, and the credibility of the model is improved. By generating a plurality of groups of scene factor combinations and inputting the scene factor combinations into a causal enhanced prediction model, a long-term illumination energy consumption prediction result under different scenes can be obtained, more comprehensive reference information is provided for a decision maker, prediction analysis under multiple scenes is supported, and the scientificity and the accuracy of decision making are improved. The length of the long-term prediction subsequence is adjusted, so that the time granularity of prediction can be flexibly controlled, long-term trend prediction can be generated, short-term fine prediction can be generated, and different decision requirements can be met. By integrating the causal enhanced long-term illumination energy consumption prediction model into the illumination optimization control system, a prediction basis can be provided for optimization scheduling of the illumination system, and the operation strategy of the illumination system is dynamically optimized by combining real-time monitoring data and a prediction result, so that the energy utilization efficiency is improved, and intelligent illumination is realized.

In an alternative embodiment of the present invention,

Illustratively, a perfect decision support knowledge base needs to be established first. And (3) constructing a semantic association network of the scene lighting equipment by adopting a knowledge graph technology to form a structured and machine-understandable knowledge representation. Specifically, attribute characteristics of the lighting equipment, such as model, power, luminous flux, color temperature, color rendering index and the like, and relationship information, such as topological connection, control logic and the like, among the equipment are obtained from heterogeneous data sources, such as equipment parameters, spatial layout, use instructions and the like, through modes of manual labeling, automatic extraction and the like. And defining core concepts, attributes and relations in the lighting field by using an ontology modeling method, and constructing an ontology model of the scene knowledge graph. And then, mapping the extracted equipment characteristics and the relation information into an ontology model to form an instantiated knowledge graph.

Implicit device association and optimization potentials can be further mined and inferred based on the constructed scene knowledge graph. For example, the energy consumption characteristics and the use modes of different areas and different types of lighting equipment are analyzed through a graph reasoning technology to find an optimized dimming combination and a control strategy, and the low-dimensional semantic representation of equipment nodes is learned through a graph representation learning technology to characterize the similarity and complementarity among the equipment nodes so as to provide semantic enhanced state characteristics for subsequent intelligent optimization.

Finally, a scene knowledge graph formed by combining multidimensional information such as energy consumption statistics, equipment characteristics, associated semantics and the like is formed, and a decision support knowledge base generated by an intelligent illumination regulation strategy is formed. Based on the knowledge base, the environment perception capability, the user demand understanding capability and the global coordination capability of the lighting optimization decision can be remarkably improved.

On the basis, a multi-objective-oriented initial lighting regulation and control decision model is further constructed. In particular, the objective function of the decision model includes three aspects, energy consumption minimization, user comfort maximization, and equipment life maximization. The energy consumption minimization aims to reduce the power consumption of the lighting system to the maximum extent, green energy-saving operation is realized, the user comfort maximization aims to create a light environment which meets the lighting requirements of users and is comfortable in vision, the user experience is improved, the equipment service life maximization aims to prolong the healthy service period of the lighting equipment, and the maintenance and replacement cost is reduced.

The key point of constructing the initial decision model is that the weight coefficient of each optimization target is reasonably set based on the energy consumption prediction result and scene knowledge, and the priority and the dependency relationship among different targets are balanced. The method comprises the steps of extracting parameters such as power and luminous flux of the lighting equipment from a scene knowledge graph, estimating space illuminance distribution and energy consumption level based on a physical model, measuring comfort level of different lighting strategies by combining knowledge such as user preference and visual health standard, and estimating health conditions of the lighting equipment under different modulation modes according to knowledge such as equipment life curves and maintenance histories. On the basis, methods such as weighted summation, analytic hierarchy process and the like are adopted to construct a multi-objective function, and the weight coefficient of each target is dynamically adjusted according to the energy consumption prediction trend, the scene semantics and the like to generate an initial decision model with scene adaptability. The target weight in the decision model can be flexibly configured according to the emphasis point of the actual scene. For example, the energy consumption target can be given higher weight for industrial workshops requiring energy conservation, the priority of the comfort target can be properly improved for elderly activity centers with heavy comfort, and the influence of the equipment life factor can be enhanced for museum exhibition halls requiring frequent dimming and sensitive life. Through knowledge-driven personalized decision modeling, the initial decision model can give consideration to a general optimization framework and specific scene requirements, and forms a regulation and control strategy foundation according to local conditions.

In order to further improve the adaptability of the intelligent lighting system to complex dynamic environments, an intelligent optimization algorithm based on reinforcement learning is designed, and the autonomous learning and dynamic evolution of a lighting regulation strategy are realized. Specifically, the intelligent lighting regulation problem is modeled as a Markov decision process, and the optimal lighting control strategy is learned through continuous interaction of an agent with the environment. The state space contains historical energy consumption data, real-time environmental parameters (e.g., illuminance, personnel activities, etc.), and user demand data (e.g., lighting preferences, visual health, etc.), reflecting the complex scene in which the lighting system is located. The action space is a lighting control parameter such as a lamp switch, brightness adjustment, color temperature adjustment and the like. The rewarding function is based on multidimensional indexes such as energy consumption, comfort level, service life and the like, and guides the intelligent agent to learn a multi-objective regulation strategy. Based on the method, a Deep Q-Network (DQN), DEEP DETERMINISTIC Policy Gradient (DDPG) and the like are adopted, and a mapping relation from a state to an action is directly learned through the approximation of a neural Network from end to generate a continuous and fine-granularity illumination control strategy.

To accelerate the convergence and robustness of strategy learning, knowledge-based reinforcement mechanisms are introduced in reinforcement learning optimization. On one hand, the scene knowledge graph is embedded into the state representation to provide global semantic information for the intelligent agent and accelerate environment understanding and strategy generalization, and on the other hand, the initial decision model is integrated into the exploration strategy to guide the intelligent agent to search in a potential strategy space so as to avoid blind attempts. In addition, a simulation platform based on a model is designed, the operation of the lighting system is simulated through a virtual environment, the strategy iterative optimization is safely and efficiently carried out on line, and the cost risk of real deployment is reduced.

And after multiple rounds of reinforcement learning iteration, the intelligent agent continuously optimizes the regulation and control strategy from the environment feedback, and finally, the optimal illumination control strategy meeting the multi-target requirement is obtained. It is worth mentioning that the learned strategy can be adaptively adjusted according to the real-time scene change without manual intervention. For example, when the illumination condition changes, the intelligent body can automatically adjust the brightness of the lamp to maintain constant illumination, when the activity of personnel is reduced, the intelligent body can automatically adjust the illumination power to save energy consumption, and when the illumination of a certain area is low for a long time, the intelligent body can properly adjust the illumination of the area to protect visual health. Through knowledge-guided continuous learning optimization, the intelligent lighting system can make real-time intelligent response to environmental changes and always keep the optimal state of energy conservation, comfort and health.

To objectively evaluate the performance of the intelligent lighting regulation strategy, a comprehensive fine-grained index evaluation system needs to be constructed. And an index set covering four dimensions of energy efficiency, comfort, health and intelligence is designed to form a quantitative basis for strategy evaluation and continuous optimization. In the energy efficiency dimension, indexes such as illumination power density (LPD), energy consumption efficiency (LPW) and the like are introduced, luminous flux output reflecting illumination energy consumption level and unit energy consumption of unit area is reflected, and the energy saving performance of the illumination system is evaluated. The comfort dimension introduces indexes such as illumination uniformity, glare index (UGR), color tolerance (CRI) and the like, measures the space uniformity of illumination distribution, the visual comfort of a light source and the fidelity of light color restoration, and evaluates the comfort of illumination quality. The health dimension introduces indexes such as a Flicker Index (FI), a blue light hazard (BHR), a circadian rhythm (CLA) and the like, the stroboscopic safety of a light source, the influence of blue light on eyesight and the interference of illumination on a biological clock are measured, and the health friendliness of an illumination scheme is evaluated. The intelligent dimension introduces indexes such as response time delay, dimming precision, optimizing effect and the like, examines the real-time response capability of the system to the change of the demand, the execution precision of the control instruction, and the balance optimizing level of multiple targets, and evaluates the intelligent degree of illumination control.

After the policy evaluation result is obtained, the difference between each index performance and the optimization target needs to be further quantized, and an interpretable and executable policy improvement scheme is generated. For example, if the energy consumption is not expected, the energy consumption hot spot can be diagnosed, the space-time interval and the regulation and control parameters with low energy efficiency are positioned, the optimization algorithm is guided to improve the consumption reduction degree of the problem interval, if the glare problem is prominent, the glare source can be analyzed, the high-brightness light spot is positioned, the optimization algorithm is guided to reduce the illumination intensity of the problem light spot, if the biological clock interference is out of standard, the circadian rhythm curve can be traced back, the time period of circadian deviation is positioned, and the optimization algorithm is guided to improve the color temperature regulation amplitude of the problem time period. Through fine-granularity policy evaluation and feedback, a closed-loop optimization guidance mechanism is formed, and pertinence and effectiveness of an optimization algorithm can be remarkably improved.

Through simulation evaluation of the initial strategy, an initial decision model based on fixed parameters and weights can be found. In order to further improve the adaptivity and generalization of illumination optimization, an adaptive parameter adjusting mechanism is designed, and the on-line optimization and dynamic updating of a decision model are realized. The self-adaptive parameter adjustment driven based on scene knowledge is the key for optimizing the decision model. On the one hand, through map reasoning and semantic analysis, the change of scene environment, such as personnel flow, task switching, weather change and the like, is perceived in real time, and the weight distribution of the optimization target is dynamically adjusted according to the change. For example, when the number of people in a conference room suddenly increases, the weight of an illumination uniformity target is increased, the optimization force of local illumination is enhanced, and when outdoor illumination suddenly decreases due to sudden severe weather, the weight of an energy consumption target is increased, and the energy saving force under the premise of ensuring illumination quality is enhanced. On the other hand, through data mining and user feedback analysis, the dynamic change of the user demand is understood in real time, and the target vector of the optimization model is dynamically adjusted according to the dynamic change. For example, the target values of the illuminance level and the color rendering index can be appropriately adjusted for the elderly users with vision deterioration, and the minimum value of the color temperature of night illumination can be appropriately adjusted for office staff who frequently overtake at night to alleviate the interference on the biological clock.

After the adjustment schemes of the optimization parameters and the targets are obtained, the optimization steps are further refined to executable decision model optimization steps. The method mainly comprises the following three steps of firstly, evaluating contribution degrees of different parameter adjustments to an optimization target based on parameter sensitivity analysis, and accordingly implementing key optimization on key parameters, secondly, searching parameter combinations meeting new optimization targets in a decision space based on optimization algorithms such as gradient descent and the like, evaluating and screening candidate solutions with optimal optimization effects, thirdly, continuously fitting a relation model of the optimization parameters and the optimization effects based on an online learning algorithm, and implementing incremental update on the relation model by utilizing newly acquired data streams, so that real-time performance and accuracy of parameter optimization are continuously improved.

The self-adaptive parameter tuning mechanism can give the initial decision model the capability of continuous evolution and always keep close fit with the dynamic scene and real-time requirement by integrating knowledge-based parameter tuning, gradient-based target optimization, learning-based model updating and other means. The optimized decision model can flow back to the virtual simulation platform for closed test and is subjected to multi-scene and multi-demand tests. When the optimization model is subjected to comprehensive verification, the optimization model can be deployed in a real physical space, so that the intelligent illumination service for users can be brought to the condition of local conditions and needs to be applied.

In an alternative embodiment of the present invention,

The calculation formula of the objective function is as follows:

;

Fig. 2 is a schematic structural diagram of a system for predicting energy consumption of intelligent illumination based on green travel according to an embodiment of the present invention, as shown in fig. 2, the system includes:

In a third aspect of an embodiment of the present invention,

There is provided an electronic device including:

A processor;

a memory for storing processor-executable instructions;

In a fourth aspect of an embodiment of the present invention,

The present invention may be a method, apparatus, system, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for performing various aspects of the present invention.

It should be noted that the above embodiments are merely for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the technical solution described in the above embodiments may be modified or some or all of the technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the scope of the technical solution of the embodiments of the present invention.

Claims

1. The intelligent illumination energy consumption prediction method based on the hotel green is characterized by comprising the following steps of:

acquiring historical illumination energy consumption data, real-time energy consumption data and influence factor data of a travel scene, preprocessing data by utilizing a data cleaning algorithm, an interpolation algorithm and an anomaly detection algorithm to obtain a preprocessed energy consumption data set and an influence factor data set, extracting features by adopting a time sequence decomposition algorithm based on the preprocessed historical illumination energy consumption data to generate an energy consumption time sequence feature set, wherein the influence factor data comprises meteorological data, tourist number data and holiday data;

The method comprises the steps of using a graph neural network algorithm, taking topological structure information, functional partition information and lighting equipment characteristic information of a travel scene as inputs, constructing a scene knowledge graph, taking a prediction model adapting to a current scene and the scene knowledge graph as inputs, generating an energy consumption prediction value through a multi-scale prediction algorithm, generating a green intelligent lighting regulation strategy through an intelligent decision algorithm based on the energy consumption prediction value and combining the lighting equipment characteristic information in the scene knowledge graph, and applying the generated green intelligent lighting regulation strategy to a lighting system of the travel scene to realize intelligent control of lighting equipment;

Based on unified feature representation, constructing a depth probability map model comprising hidden variables and observed variables, inputting the hidden variables and context information related to the observed variables into a multi-layer perceptron, calculating to obtain attention weight vectors corresponding to each observed variable, carrying out weighted summation on the hidden variables by using the attention weight vectors, and processing by a nonlinear activation function to generate the observed variables, wherein the hidden variables represent potential modes of energy consumption and implicit actions of influencing factors, and the observed variables represent energy consumption characteristics and influencing factor characteristics;

calculating the prediction loss based on the prediction result and the actual energy consumption data, constructing an overall optimization objective function by combining the prediction loss and the calculated variation lower bound, adopting a random gradient descent optimization algorithm based on the overall optimization objective function, iteratively updating parameters of the combined prediction model, and repeating iteration until the maximum iteration times are met, so as to obtain a final prediction model adapting to the current scene;

2. The method of claim 1, wherein the calculation formula for constructing the overall optimization objective function in combination with the predicted loss and the calculated variance lower bound is as follows:

3. The method of claim 1, wherein applying a causal inference and context analysis based prediction model to the long-term predictor sequence to generate a long-term illumination energy consumption prediction result comprises:

4. The method of claim 1, wherein generating a green intelligent lighting regulation strategy using an intelligent decision algorithm based on the energy consumption prediction value in combination with lighting device characteristic information in the scene knowledge graph comprises:

5. The method of claim 4, wherein the objective function is calculated as:

6. A green intelligent lighting energy consumption prediction system based on a hotel for implementing the method of any of the preceding claims 1-5, characterized by comprising:

7. An electronic device, comprising:

A processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any of claims 1 to 5.

8. A computer readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the method of any of claims 1 to 5.