CN110503570A - A method, system, device and storage medium for detecting abnormal power consumption data - Google Patents

Abstract

The invention provides a method for detecting abnormal electricity consumption data, which includes the steps of: acquiring data, cleaning data, reducing data dimension, establishing a model, and screening abnormal users. The invention also relates to an abnormal power consumption data detection system, electronic equipment and storage medium. The invention effectively solves the problems existing in the line loss management, can conduct data mining research and analysis on the abnormal line loss of the power consumption system in the station area, makes the line loss management more transparent and efficient, and can exert its comprehensive management application, and finally realize the The goal of energy saving and loss reduction and standardized management.

Description

A method, system, device and storage medium for detecting abnormal power consumption data

technical field

The invention relates to the technical field of electricity consumption information collection, in particular to a method for detecting abnormal electricity consumption data.

Background technique

With the rapid development of the information age, the Internet and information and communication industries are the first to carry out research related to big data. For the power industry, big data also has far-reaching research significance and bright application prospects. With the gradual evolution of the next-generation power system, the data-driven power supply chain will gradually replace the traditional power supply chain. Among them, the promotion of electricity consumption information collection system provides the necessary data foundation for the management and operation decision-making and power supply service optimization based on power data analysis in my country's power industry. At the same time, with the exponential growth of power consumption data such as power data, working condition data, and event information, the characteristics of big data are becoming more and more prominent, and the application demand of power consumption big data is becoming more and more urgent. Massive power consumption data mainly comes from various metering devices and systems. Due to various equipment failures, communication failures, power grid fluctuations and management and other reasons, a large number of abnormal power consumption data have appeared. In the face of this massive increase in electricity consumption data, most power departments only use traditional statistical methods to analyze abnormal data, and most of them need to rely on on-site inspections. Due to the limitation of human, material and financial resources, the underlying reasons behind abnormal data cannot be effectively extracted, but it has brought about "data disaster" and "data waste". Therefore, it is difficult to meet the requirements with traditional analysis methods. We need to discover the deeper laws of abnormal electricity consumption data through data mining, eliminate the contingency of the data, and refine the inevitability of the data.

Due to the large number of low-voltage customer groups and frequent changes, there are currently common line loss abnormalities in Taiwan district line loss management due to unclear household change relationship, poor meter reading quality, electricity theft, metering failures and other management reasons. In recent years, many domestic power supply companies have faced a common dilemma to varying degrees, that is, "large investment and low return" in the treatment of Taiwan line losses. It turns into a loss in management, and the direction of investment in transformation remains unchanged.

SUMMARY OF THE INVENTION

In order to overcome the deficiencies of the prior art, the present invention provides a method for detecting abnormal power consumption data. The present invention achieves high throughput of big data batch processing tasks by using efficient parallel computing technology through application integration based on the integration of real-time database, cloud computing, and cloud real-time storage platform technology. Using the isolated forest algorithm with good stability and strong anti-noise performance, it can effectively mine abnormal data users, analyze the cause of line loss, and strengthen the management of line loss in the station area.

The present invention provides a method for detecting abnormal electricity consumption data, comprising the following steps:

Obtain data, and obtain electricity consumption data through electricity consumption information collection;

Data cleaning, cleaning the collected electricity consumption data, and detecting the type of dirty data in the electricity consumption data to obtain valid electricity consumption data; the types of dirty data include: missing values, duplicate values, maximum and minimum values, Load burr, negative impact value;

Data dimensionality reduction, using the daily load characteristic index to perform feature dimension reduction on the effective power consumption data, the daily load characteristic index includes load rate, peak-valley difference rate, maximum utilization hour rate, peak load rate, and average load rate , valley load rate;

Model establishment, several isolated trees are constructed into an isolated forest, the first analysis model is established by using the isolated forest algorithm, and the model is evaluated by using the evaluation curve;

Screen abnormal users, use the first analysis model to screen target data, and perform data mining on the screened data to screen out abnormal electricity users.

Preferably, the electricity consumption information collection method includes cloud storage, and the cloud storage is used for scattered storage of the electricity consumption data on a plurality of independent storage servers, and the types of the storage servers include metadata management services, volume Management Services and Block Data Management Services.

Preferably, in the step data cleaning, it also includes: the calculation formula for filling empty missing values according to the fluctuation period characteristics of the electricity load is as follows:

Among them, X _i represents the electricity load at the current moment, i is the moment when the load data is missing, ranging from 1 to 24, and a1 and a2 are the weighting coefficients of the load at the corresponding moment of the two days before and after the current moment and the two time points before and after the current moment.

Preferably, before the step of acquiring data, it also includes the steps of:

Establish a management plan and establish line loss management indicators in the station area. The status identifiers of the station area line loss management indicators include coverage type, household change type, recoverable type, data type, and line loss type; The electrical data is used to identify the state, and corresponding control measures are taken for different states to form a line loss management plan in the station area.

Preferably, in the establishment of the step model, it also includes: using the receiver operating characteristic ROC curve, the area under the curve AUC, the cumulative recall rate curve and the P-R curve, taking the precision as the ordinate and the recall as the abscissa to model the model Evaluate.

Preferably, the isolated forest algorithm includes a first-stage algorithm and a second-stage algorithm, the first-stage algorithm includes constructing a plurality of isolated trees to form an isolated forest; the second-stage algorithm includes using the generated isolated forest to evaluate and test data, and calculate the abnormal score for the detected data.

An electronic device, comprising: a processor;

a memory; and a program, wherein the program is stored in the memory and configured to be executed by a processor, the program including an abnormal power usage data detection method.

A computer-readable storage medium on which a computer program is stored, the computer program being executed by a processor includes a method for detecting abnormal power consumption data.

An abnormal power consumption data detection system, comprising a data acquisition module, a data cleaning module, a data dimension reduction module, a model building module, and a screening abnormal user module; wherein,

The acquiring data module is used for acquiring electricity consumption data by means of electricity consumption information collection;

The data cleaning module is used to clean the collected electricity consumption data, and detect the type of dirty data in the electricity consumption data to obtain valid electricity consumption data; the types of dirty data include: missing values, duplicate values, extremely large Minimum value, load burr, negative impact value;

The data dimensionality reduction module is used to perform feature dimensionality reduction on the effective power consumption data by using a daily load characteristic index, wherein the daily load characteristic index includes a load rate, a peak-valley difference rate, a maximum utilization hour rate, a peak load rate, Load rate during flat period and load rate during valley period;

Described model establishment module is used for several isolated trees to construct into isolated forest, utilizes the isolated forest algorithm to establish the first analysis model, utilizes the evaluation curve to carry out model evaluation;

The abnormal user screening module is used to screen the target data by using the first analysis model, and perform data mining on the screened data to screen out abnormal electricity users.

Preferably, it also includes an establishment management scheme module, the establishment management scheme module is used to establish the station area line loss management index, and the status identifiers of the station area line loss management index include coverage type, household change type, recoverable type, data Class, line loss class; state identification of the collected power consumption data of multiple stations, and take corresponding control measures for different states to form a station area line loss management plan;

The data acquisition module includes a cloud storage unit, and the cloud storage unit is configured to store the power consumption data in a distributed manner on multiple independent storage servers, and the types of the storage servers include metadata management services, volume management services, and block data management services;

The data cleaning module includes a missing value filling unit, and the calculation formula of the missing value filling unit for filling empty missing values according to the fluctuation period characteristic of the electricity load is as follows:

Among them, X _i represents the electricity load at the current moment, i is the moment when the load data is missing, ranging from 1 to 24, and a1 and a2 are the weighting coefficients of the load at the corresponding moment of the two days before and after the current moment and the two time points before and after the current moment.

Compared with the prior art, the beneficial effects of the present invention are:

1), a detection method of abnormal power consumption data, as a new station line loss management method that adapts to the development needs of smart grid, effectively solves the problems existing in the current station area management, and makes the station area line loss management more transparent , high efficiency, give full play to its comprehensive management role in marketing management, and ultimately achieve the goals of energy saving, loss reduction, and standardized management.

2) The established line loss management index system in the station area has five states: coverage, household change, recoverable, data, line loss, and five state identifiers and their hierarchical relationships. For stations in different states, according to different management and control priorities, formulate different management and control methods, management and control cycles, and responsible departments, and ultimately promote the progress of the stations in a healthy state.

3) The present invention effectively solves the problems existing in the line loss management, and can conduct data mining research and analysis on the abnormal line loss of the power consumption system in the station area, so that the line loss management is more transparent and efficient, and its comprehensive management application can be exerted. , and ultimately achieve the goals of energy saving and loss reduction and standardized management;

4) Cloud computing technology can provide high-quality services on demand by using distributed software and hardware resources and information, and has been successfully applied in many fields such as search engines, social networks, and communications. In the field of smart grid informatization construction, cloud computing's unique large-scale data efficient access and parallel computing capabilities enable it to provide high-quality data processing services for information systems including power consumption information collection systems, and provide intelligent The information system in the power grid era provides solid technical support.

The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, and to implement it according to the content of the description, the preferred embodiments of the present invention are described below in detail with the accompanying drawings. Specific embodiments of the present invention are given in detail by the following examples and the accompanying drawings.

Description of drawings

The accompanying drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the attached image:

Fig. 1 is the overall flow chart of a kind of abnormal electricity consumption data detection method of the present invention;

Fig. 2 is the progressive schematic diagram of the station area line loss management index state of a kind of abnormal power consumption data detection method of the present invention;

3 is a schematic diagram of constructing an isolated tree of a method for detecting abnormal electricity consumption data according to the present invention;

4 is a schematic diagram of data dimensionality reduction processing of a method for detecting abnormal electricity consumption data according to the present invention;

5 is a schematic diagram of screening abnormal users in a method for detecting abnormal electricity consumption data according to the present invention;

Fig. 6 is the overall structure schematic diagram of the service-oriented architecture system of a kind of abnormal power consumption data detection system of the present invention;

FIG. 7 is a schematic diagram of the overall structure of an abnormal power consumption data detection system of the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be noted that, on the premise of no conflict, the embodiments or technical features described below can be combined arbitrarily to form new embodiments. .

A method for detecting abnormal electricity consumption data, as shown in Figure 1, includes the following steps:

S0. Establish a management plan, and establish a line loss management index in the station area. The status identifiers of the line loss management index in the station area include coverage type, household change type, recoverable type, data type, and line loss type; According to the power consumption data, the status is identified, and corresponding control measures are taken for different statuses to form a line loss management plan in the station area. In one embodiment, as shown in Figure 2, 1.1-Initial station area: data preparation, included in hierarchical progressive management; 1.2-Installation station area: Reasonable arrangement of equipment installation, coverage rate reaches 100%; 1.3-Household change Class station area: check the relationship between households in the station area, the accuracy rate reaches 100%; 1.4-recoverable class station area: collect multiple times, analyze faults, and the recoverable rate reaches 95%; 1.5-data class station area: collect multiple times, Analyzing the error, the recoverable rate reaches 95%; 1.6-line loss type station area: analyze the cause of abnormal line loss rate and formulate loss reduction measures; 1.7-standard station area: take solid optimization measures to maintain the standard station status.

Specifically, the established station area line loss management indicators include five state identifiers and their hierarchical relationships: coverage type, household variable type, recoverable type, data type, and line loss type; The following status identification is carried out, and corresponding control measures are formulated for the management and control focus of different types of stations, so as to form a line loss management method based on the power consumption information collection system. The specific measures are as follows:

Coverage category: If the installation rate of collection equipment in the station area does not reach 100%, the collection equipment installation plan should be reasonably arranged;

Household change category: The collection coverage has reached 100% in the Taiwan area, but the household change relationship is still inaccurate, and the household change relationship should be approved through the combination of internal inspection data and on-site inspection;

Recoverable category: The collection coverage rate has reached 100%, but the recoverable rate has not yet reached 95%. The recoverable rate should be counted, and the reasons for missed and miscollected collections should be analyzed;

Data type: The coverage rate reaches 100%, the recovery rate reaches 95%, and the household change relationship is correct, but the error between the collected data and the manual meter reading data is greater than the average value, and a reasonable meter reading plan is formulated;

Line loss category: The coverage rate, recoverability rate, and accuracy rate have all reached 100%, and the household change relationship is correct, but the line loss rate is abnormal. The reason for the abnormal line loss rate should be analyzed in time, and loss reduction measures should be formulated.

S1. Acquire data, and acquire power consumption data by means of power consumption information collection. In one embodiment, for the line loss type station area, the cloud storage technology is used to realize the collection, classification and processing of power consumption information data of multiple line loss type station areas. The distributed file storage mechanism of cloud storage is adopted to store electricity information data in multiple independent storage servers. It includes volume management, metadata management, and block data management services;

Metadata refers to the file name, attributes, and data block location information. Because metadata is accessed frequently, the system loads and caches the metadata into memory for management to improve access efficiency.

Block data refers to file data divided into multiple data blocks according to a certain size, which are distributed and stored on different storage node servers. The storage space provided by a pair of metadata servers and the storage server nodes managed by them is called a data block. volume space;

The volume management server is responsible for virtualizing and integrating multiple volumes and providing a unified overall access to the cloud real-time storage platform space.

S2. Data cleaning, cleaning the collected electricity consumption data, and detecting the type of dirty data in the electricity consumption data to obtain valid electricity consumption data; the types of dirty data include: missing values, duplicate values, extremely large and extremely small data. value, load glitch, shock negative value. In one embodiment, the types of dirty data are analyzed and summarized, and then targeted measures are taken according to their manifestations to delete redundant data in the data set and maintain the integrity of the data set. Common dirty data types are: 1), missing values: empty values in the table; 2), duplicate values: the user's electricity load data is repeated at a certain moment; 3), maximum and minimum values: the electricity load data is too large or Too small; 4), load burr: the data suddenly increases or decreases between adjacent periods; 5), negative impact value: the reading data drops in a continuous period of time.

According to the fluctuation cycle characteristics of the electricity load, for the data with serious missing data, according to the fluctuation cycle characteristics of the electricity load, calculate the load at the same time point in the two adjacent days before and after, and the average value of the load at the two time points before and after the current moment, and then The load change rate method of one day relative to the previous day fills the vacant value with the mean value plus the load change amount. The calculation method is as follows:

In the formula, X _i represents the electricity load at the current moment, i is the moment when the load data is missing, and takes a value from 1 to 24. α ₁ and α ₂ represent the corresponding moments of the two days before and after the current moment and the weighted load of the two time points before and after the current moment. coefficient. For abnormal noise point data, the rectangular method is used to integrate the load data at each collection time of the day to calculate the repair value of the electricity. The calculation formula is as follows:

In the formula, X _i is the power restoration value, F is the number of load data collection in one day, P _i is the load data at time i, and ΔT is the load data collection time interval.

S3. Data dimensionality reduction, using a daily load characteristic index to perform feature dimension reduction on the effective power consumption data, where the daily load characteristic index includes a load rate, a peak-valley difference rate, a maximum utilization hour rate, a peak load rate, and a flat period. Load rate, valley load rate. In one embodiment, as shown in Figure 4, for the load curve as a time series, the electricity load data is easily affected by various factors such as temperature, income, and electricity price policy. The distance cannot be fully reflected, and the similarity of the shape or contour of the time series cannot be fully guaranteed. Moreover, for curves with obvious load shapes such as daily load curves, they will show unsatisfactory isometric properties in high-dimensional situations. In order to fully reflect the similarity between loads and take into account the computing efficiency, this embodiment selects six commonly used daily load characteristic indicators: load rate, peak-valley difference rate, maximum utilization hour rate, peak load rate, average load rate, The load rate during the valley period reflects the power consumption characteristics of various users more comprehensively from the four perspectives of the whole day, peak period, flat period and valley period. Feature dimension reduction is performed on the effective load curve matrix using six daily load characteristic indexes.

S4, model establishment, model establishment, a number of isolated trees are constructed into an isolated forest, a first analysis model is established by using the isolated forest algorithm, and model evaluation is performed by using an evaluation curve;

In one embodiment, as shown in Fig. 3, an isolated tree iTree is constructed, and the process is as follows: 1. Randomly select a feature from the six daily load characteristic indicators; 2. Randomly select a value k of the feature; 3. According to the feature Classify each record, put the records less than k in the feature on the left branch, and put the records greater than or equal to k on the right branch; 4. Then recursively construct the left branch and the right branch until the following conditions are met: a. Incoming The data set has only one record or multiple identical records; b. The height of the tree reaches the limit height.

Specifically, to construct an isolated forest composed of t iTrees, the steps are as follows:

1. Randomly select ψ sample points from the training data as a sub-sample set and put them into the root node of the tree;

2. Randomly specify a dimension, and randomly generate a cutting point P in the current node data;

3. Generate a hyperplane at this cutting point, divide the data space of the current node into two subspaces, put the data less than P in the specified dimension on the left of the current node, and put the data greater than or equal to p on the right of the current node.

4. Steps 1 and 2 are recursively performed in the child nodes, and new child nodes are continuously constructed until the data itself can no longer be divided or the depth of the tree reaches log ₂ ψ.

S5. Screen abnormal users, screen the target data by using the first analysis model in the step model establishment, and perform data mining on the screened data to screen out abnormal electricity users. In one embodiment, as shown in Figure 5, iForest is composed of several iTrees with differences, and the model is evaluated by using ROC curve and AUC, cumulative recall curve and P-R curve. When a single user evaluates, all iTrees need to be traversed during each evaluation process. The position of the leaf node that the query object falls on is counted, and the anomaly score is calculated by the average path length. Finally, according to the size of the abnormal score, the user is evaluated to determine whether the user to be tested is an abnormal user.

Specifically, for the receiver operating characteristic ROC curve, when the distribution of positive and negative samples in the test set changes, the ROC curve can remain unchanged. For the continuous values output by the binary classification model, the samples larger than the threshold are classified as positive, and the samples smaller than the threshold are classified as negative. Reducing the threshold value can identify more positive classes, improve the recall rate, and at the same time classify more negative samples as positive classes, thus increasing the false positive rate. The ROC curve visualizes this change process. In the ROC space coordinates, the point (0,1) represents the ideal classifier, and the closer the ROC curve is to the point (0,1), the better the classification effect. The value of AUC is the size of the area under the ROC curve. AUC=1 corresponds to an ideal classifier, and AUC=0.5 means that the model has no predictive value, just like the guessing of the follower machine. Between 0.5 and 1, it means that it is better than random guessing.

For the P-R curve, plot the precision rate on the vertical axis and the recall rate on the horizontal axis to obtain the curve of the precision rate and the recall rate, referred to as "P-R curve" as the classification threshold changes from large to small. The accuracy decreases and the recall increases. When evaluating the classifier, the closer the P-R curve is to the point (1,1), the better the classification effect.

Specifically, the generated iForest is used to evaluate the test data, and the abnormal score is calculated for the detected samples. For any data x, let it traverse each iTree, get the depth of x in the iTree and the average depth h(x) in each iTree, so as to calculate the abnormal score of the sample. The abnormal score of the detected sample x is defined as follows:

Among them: h(x) is the depth of the node retrieved by the detected sample x in the iTree; E(h(x)) is the average of all t iTrees; c(ψ) is the binary search tree constructed by ψ points The average path length of ;

H(k)=ln(k)+ζ, where ζ is Euler's constant.

Observing the definition of abnormal score, we can see that: when E(h(x))→0, s→1; when E(h(x))→ψ-1, s→0; when E(h(x)) →c(ψ),s→0.5. That is, the closer s(x) is to 1, the higher the possibility of abnormal data, and the closer to 0, the higher the possibility of normal points.

An electronic device, comprising: a processor;

a memory; and a program, wherein the program is stored in the memory and configured to be executed by a processor, the program including an abnormal power usage data detection method.

A computer-readable storage medium on which a computer program is stored, the computer program being executed by a processor includes a method for detecting abnormal power consumption data.

An abnormal power consumption data detection system, as shown in Figure 7, includes a data acquisition module, a data cleaning module, a data dimensionality reduction module, a model building module, and a screening abnormal user module; wherein,

The acquiring data module is used for acquiring electricity consumption data by means of electricity consumption information collection;

The data cleaning module is used to clean the collected electricity consumption data, and detect the type of dirty data in the electricity consumption data to obtain valid electricity consumption data; the types of dirty data include: missing values, duplicate values, extremely large Minimum value, load burr, negative impact value;

The data dimensionality reduction module is used to perform feature dimensionality reduction on the effective power consumption data by using a daily load characteristic index, wherein the daily load characteristic index includes a load rate, a peak-valley difference rate, a maximum utilization hour rate, a peak load rate, Load rate during flat period and load rate during valley period;

Described model establishment module is used for several isolated trees to construct into isolated forest, utilizes the isolated forest algorithm to establish the first analysis model, utilizes the evaluation curve to carry out model evaluation;

The abnormal user screening module is used to screen the target data by using the first analysis model, and perform data mining on the screened data to screen out abnormal electricity users.

Further, it also includes establishing a management scheme module, and the establishment management scheme module is used to establish a line loss management index in the station area, and the state identification of the line loss management index in the station area includes a coverage class, a household change class, a collectible class, and a data Class, line loss class; state identification of the collected power consumption data of multiple stations, and take corresponding control measures for different states to form a station area line loss management plan;

The data acquisition module includes a cloud storage unit, and the cloud storage unit is configured to store the power consumption data in a distributed manner on multiple independent storage servers, and the types of the storage servers include metadata management services, volume management services, and block data management services;

The data cleaning module includes a missing value filling unit, and the calculation formula of the missing value filling unit for filling empty missing values according to the fluctuation period characteristic of the electricity load is as follows:

Among them, X _i represents the electricity load at the current moment, i is the moment when the load data is missing, ranging from 1 to 24, and a1 and a2 are the weighting coefficients of the load at the corresponding moment of the two days before and after the current moment and the two time points before and after the current moment.

In a specific embodiment, in the development and design of the system, a service-oriented architecture system is used as the overall design ideological structure, and the data acquisition module adopts a special-purpose terminal of the 05 version of the statute, which can collect the user's electric energy meter every 15 minutes, that is, a total of 24 hours. The voltage, current and power data of 96 points, that is, the data set S is an n*24-order initial load curve matrix composed of n daily load curves. This module realizes distributed storage of the collected massive data through cloud storage technology. After the data is processed, it is obtained: From September 2018 to March 2019, a county power company has a total of 701 stations, with a total capacity of 349,000 KVA, an average single-unit capacity of 497.8KVA, and a cumulative power loss of 46,000 Kwh. The average station area The line loss rate is 2.69%.

Further, as shown in Figure 6, the collection cluster periodically collects information from the user terminal, and stores the data in the cloud storage and query environment by calling the storage interface; Concurrent storage, and provide upward power data indexing and efficient query functions. The parallel ETL (Extraction-Transformation-Loading) environment is responsible for the data exchange between the archive information in the original relational database and the cloud computing environment; the ETL management tool is used to establish the data table mapping relationship and the task execution strategy, and the system uses the parallel ETL tool for the associated system. The data in the real-time tracking, acquisition and consistency check. The parallel analysis and computing environment is responsible for running the isolated forest algorithm to mine abnormal data. Front-end interfaces include SQL-like (Structured Query Language) interfaces, Web services, client packages, etc., and provide query and analytical computing services for external systems. The mapping tool adopts the optimization technology of SQL to Map/Reduce based on query rewriting, converts the original SQL into a query graph, and uses the rewriting rules to evolve into various forms to realize the application of the original stored procedure form to cloud computing The assisted migration, correctness verification and performance optimization of the environment can greatly reduce the migration cost of relational database applications to cloud computing, improve development efficiency, and improve the overall performance of parallel computing.

The cloud storage unit adopts a parallel ETL environment to atomically decompose the original computationally intensive and complex tasks and assign them to different task processing nodes for concurrent and synchronous processing, thereby improving data processing efficiency and data processing capacity and ensuring data processing performance.

The system also includes a loss reduction auxiliary decision-making module, which mainly includes two parts: loss reduction decision support function and loss reduction scheme database management. This module checks users with abnormal power consumption data, focusing on the following contents: a. Whether there is theft in the station area Electricity behavior; b. The load operation changes in the platform area, and whether there is a switch; c. Whether the transformer in the platform area is light-loaded or heavy-loaded; d. The operation of the reactive power compensation equipment; e. Whether the three-phase load is balanced; f. The voltage quality; g. Whether the transformer, circuit, and metering equipment are reasonable and normal; the main reasons for the abnormality of the electric energy metering device are meter failure, transformer failure, junction box failure and terminal failure, etc.; h. Whether the low-voltage power supply radius is too long; i. Abnormal line loss caused by other reasons.

Table 1 is the analysis and statistics table of the line loss type station area of the abnormal power consumption data detection system in March 2019:

As can be seen from Table 1, there are 694 stations in the current line loss rate compliance, accounting for about 99% of the total number of stations within the management scope. Through further detailed analysis of the station area, the main reason for the low station collection coverage rate is the low installation rate of non-resident collection equipment in most station areas. After the cause is found out, the installation plan of the acquisition equipment should be adjusted. In addition, the problem of inaccurate household change relationship in Taiwan area ranks second in affecting the control effect of Taiwan area. Among the 456 stations with 100% collection coverage, 312 household change relationships are accurate, with an accuracy rate of 68%. A survey of 144 stations with inaccurate household change relationships found that the main reasons were: firstly, some old station areas had lost data; While reasonably arranging the installation of acquisition equipment, attention should be paid to the verification of the relationship between households in the station area, and the two-way customer identification device in the station area can also be used to assist in the verification of the relationship between the household changes in the station area.

The above are only preferred embodiments of the present invention, and do not limit the present invention in any form; any person of ordinary skill in the industry can smoothly implement the present invention as shown in the accompanying drawings and above; however, any Those skilled in the art, without departing from the scope of the technical solution of the present invention, make use of the above-disclosed technical content to make some changes, modifications and equivalent changes of evolution are equivalent embodiments of the present invention; at the same time, Any alteration, modification and evolution of any equivalent changes made to the above embodiments according to the essential technology of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims (10)

1. an abnormal electricity consumption data detection method, is characterized in that, comprises the steps: Obtain data, and obtain electricity consumption data through electricity consumption information collection; Data cleaning, cleaning the collected electricity consumption data, and detecting the type of dirty data in the electricity consumption data to obtain valid electricity consumption data; the types of dirty data include: missing values, duplicate values, maximum and minimum values, Load burr, negative impact value; Data dimensionality reduction, using daily load characteristic indicators to perform feature dimension reduction on the effective power consumption data, the daily load characteristic indicators include load rate, peak-to-valley difference rate, maximum utilization hour rate, peak load rate, and average load rate , valley load rate; Model establishment, several isolated trees are constructed into an isolated forest, the first analysis model is established by using the isolated forest algorithm, and the model is evaluated by using the evaluation curve; Screen abnormal users, screen target data by using the first analysis model, perform data mining on the screened data, and screen out users with abnormal electricity consumption. 2 . The method for detecting abnormal electricity consumption data according to claim 1 , wherein the electricity consumption information collection method comprises cloud storage, and the cloud storage is used for scattered storage of the electricity consumption data in a plurality of devices. 3 . On an independent storage server, the types of the storage server include a metadata management service, a volume management service, and a block data management service. 3. A kind of abnormal electricity consumption data detection method as claimed in claim 1 or 2, it is characterized in that, in step data cleaning, also comprises: the calculation formula of filling empty missing value according to the fluctuation period characteristic of electricity load is as follows: Among them, X _i represents the electricity load at the current moment, i is the moment when the load data is missing, ranging from 1 to 24, and a1 and a2 are the weighting coefficients of the load at the corresponding moment of the two days before and after the current moment and the two time points before and after the current moment. 4. a kind of abnormal electricity consumption data detection method as claimed in claim 3, is characterized in that, also comprises the step before the step of acquiring data: Establish a management plan and establish line loss management indicators in the station area. The status identifiers of the station area line loss management indicators include coverage type, household change type, recoverable type, data type, and line loss type; The electrical data is used to identify the state, and corresponding control measures are taken for different states to form a line loss management plan in the station area. 5. a kind of abnormal electricity data detection method as claimed in claim 1 is characterized in that, in step model establishment, also comprises, utilizes receiver operating characteristic ROC curve, area under the curve AUC, cumulative recall curve and For the P-R curve, the precision is the ordinate and the recall is the abscissa for model evaluation. 6. A method for detecting abnormal electricity consumption data according to claim 1 or 5, wherein the isolated forest algorithm includes a first-stage algorithm and a second-stage algorithm, and the first-stage algorithm includes constructing a plurality of The isolated trees form an isolated forest; the second-stage algorithm includes using the generated isolated forest to evaluate the test data, and to calculate an anomaly score for the detected data. 7. An electronic device, comprising: a processor; a memory; and a program, wherein the program is stored in the memory and configured to be executed by a processor, the program comprising for performing the method of claim 1 . 8. A computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to execute the method according to claim 1. 9. An abnormal power consumption data detection system, characterized in that it comprises a data acquisition module, a data cleaning module, a data dimensionality reduction module, a model building module, and a screening abnormal user module; wherein, The acquiring data module is used for acquiring electricity consumption data by means of electricity consumption information collection; The data cleaning module is used to clean the collected electricity consumption data, and detect the type of dirty data in the electricity consumption data to obtain valid electricity consumption data; the types of dirty data include: missing values, duplicate values, extremely large Minimum value, load burr, negative impact value; The data dimensionality reduction module is used to perform feature dimensionality reduction on the effective power consumption data by using a daily load characteristic index, wherein the daily load characteristic index includes a load rate, a peak-valley difference rate, a maximum utilization hour rate, a peak load rate, Load rate during flat period and load rate during valley period; The model establishment module is used for constructing an isolated forest from several isolated trees, using an isolated forest algorithm to establish a first analysis model, and using an evaluation curve to evaluate the model; The abnormal user screening module is used for screening the target data by using the first analysis model, and performing data mining on the screened data to screen out abnormal electricity users. 10. A system for detecting abnormal power consumption data as claimed in claim 1, further comprising establishing a management scheme module, the establishment management scheme module is used to establish a station area line loss management index, the station area line The status identification of the loss management indicators includes coverage type, household change type, recoverable type, data type, and line loss type; status identification is performed on the collected electricity consumption data of multiple stations, and corresponding control measures are taken for different states. , to form a line loss management plan in the Taiwan area; The data acquisition module includes a cloud storage unit, and the cloud storage unit is configured to store the power consumption data in a distributed manner on multiple independent storage servers, and the types of the storage servers include metadata management services, volume management services, and block data management services; The data cleaning module includes a missing value filling unit, and the calculation formula of the missing value filling unit for filling empty missing values according to the fluctuation period characteristic of the electricity load is as follows: Among them, X _i represents the electricity load at the current moment, i is the moment when the load data is missing, ranging from 1 to 24, and a1 and a2 are the weighting coefficients of the load at the corresponding moment of the two days before and after the current moment and the two time points before and after the current moment.