Electric charge recovery data quality diagnosis and user behavior risk assessment method

Abstract

The invention discloses an electric charge recovery data quality diagnosis and user behavior risk assessment method, which comprises the following steps: acquiring multi-source measurement data in a first platform, and characterizing the multi-source measurement data; performing data performance improvement on the characterized multi-source measurement data, and performing quality diagnosis on the improved data; classifying the users by identifying interaction behaviors of the users and the first platform; and performing first element evaluation on the user in each classification result, and according to an evaluation result of the first element, evaluating a first feature risk. Accuracy and reliability of electricity consumption data are guaranteed, economic operation of a power grid and stability of power supply are guaranteed, and the method is of great importance to maintenance of overall healthy operation of the power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment. Background Art

[0002] With the advancement of global energy transformation, efficient, clean, and low-carbon energy forms represented by wind and solar energy have become the mainstream of the power industry. Distributed photovoltaic construction projects carried out by many power companies are gradually becoming an important part of power supply due to their high proportion, multi-regional, and multi-type characteristics. However, large-scale distributed photovoltaic power supply faces challenges in data identification, diagnosis, and user behavior risk assessment in terms of electricity consumption and electricity fee recovery. At present, there is a lack of an efficient, accurate, and unified method to deal with these problems. Summary of the invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment, comprising:

[0005] Acquire multi-source measurement data in the first platform, and characterize the multi-source measurement data;

[0006] Improve the data performance of multi-source measurement data after characterization, and perform quality diagnosis on the improved data;

[0007] Classifying users by identifying interaction behaviors between users and the first platform;

[0008] A first factor evaluation is performed on the user in each classification result, and a first characteristic risk is evaluated based on the evaluation result of the first factor.

[0009] As a preferred solution of the method for quality diagnosis of electricity fee recovery data and user behavior risk assessment of the present invention, wherein: the multi-source measurement data includes data information collected from different measurement points at the user in the first platform;

[0010] At the same time, bad data are detected on the multi-source measurement data, and good data are retained.

[0011] As a preferred solution of the method for quality diagnosis of electricity fee recovery data and user behavior risk assessment of the present invention, the characterization includes integrating and processing benign multi-source measurement data so that the data integrated from different measurement points reflects the characteristics of the user;

[0012] The integration processing includes, but is not limited to, integration based on time: synchronizing the data from different locations and different devices in time, and performing time delay correction and fusion on different measurement data.

[0013] As a preferred solution of the method for quality diagnosis of electricity fee recovery data and user behavior risk assessment of the present invention, the improvement of data performance includes optimizing the data step by step according to the results of the characteristic performance by identifying the characteristic performance of the data in different dimensions;

[0014] The quality diagnosis includes obtaining the user's data analysis results through a data analysis algorithm, and obtaining the quality diagnosis results by integrating the analysis results.

[0015] As a preferred solution of the method for quality diagnosis of electricity fee recovery data and user behavior risk assessment of the present invention, wherein: the first factor assessment includes a calculation index that can reflect the first characteristic risk;

[0016] The first characteristic risk is the risk that exists when the user interacts with the first platform.

[0017] As a preferred solution of the method for quality diagnosis of electricity fee recovery data and user behavior risk assessment of the present invention, wherein: the characteristic performance of the identified data in different dimensions, and the step-by-step optimization of the data according to the results of the characteristic performance include but are not limited to, in the dimension of data integrity, optimizing the integrity of the data through measurement integrity verification based on the scope of the object model;

[0018] In terms of data accuracy, the accuracy of data is optimized through measurement point calibration based on sliding windows;

[0019] In the dimension of the stability of the second system, the power imbalance problem in the second system is detected and corrected by long-time scale power balance verification;

[0020] The second system is the supply system of the first platform. The second system directly provides services to users and uploads data during the service process to the first platform, providing a data basis for the first platform's interactive behavior with users.

[0021] As a preferred solution of the method for quality diagnosis of electricity fee recovery data and user behavior risk assessment of the present invention, wherein: the data analysis results of the user are obtained by a data analysis algorithm, and the analysis results are integrated, including but not limited to, using PCA to perform data dimension reduction and pattern recognition to obtain the principal component matrix of the user, and the original data is multiplied by the principal component matrix to obtain the principal component score;

[0022] The principal component scores and corresponding weights are summed and integrated to obtain a comprehensive quality diagnosis result.

[0023] A quality diagnosis system for electricity fee recovery data and user behavior risk assessment using any method described in the present invention, characterized in that:

[0024] A collection unit, which obtains multi-source measurement data in the first platform and characterizes the multi-source measurement data;

[0025] The diagnostic unit improves the data performance of the multi-source measurement data after characterization and performs quality diagnosis on the improved data;

[0026] a classification unit, which classifies users by identifying interaction behaviors between users and the first platform;

[0027] The risk assessment unit performs a first factor assessment on the user in each classification result, and assesses the first characteristic risk according to the assessment result of the first factor.

[0028] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein: the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0029] A computer-readable storage medium stores a computer program, wherein: when the computer program is executed by a processor, the steps of any one of the methods of the present invention are implemented.

[0030] Beneficial effects of the present invention: The quality diagnosis of electricity bill recovery data and the user behavior risk assessment method provided by the present invention ensure the accuracy and reliability of electricity consumption data, and provide a solid data foundation for electricity bill recovery. It can monitor and warn of potential arrears risks in real time, helping power companies to manage risks more effectively and reduce losses. It strengthens the management of user credit, helps to improve users' awareness of voluntary payment, and promotes a virtuous cycle of electricity bill recovery. It provides an important reference basis for decision makers of power companies and helps to formulate more scientific and reasonable electricity bill recovery strategies. It helps to ensure the economic operation of the power grid and the stability of power supply, which is crucial to maintaining the overall healthy operation of the power grid. It has promoted innovation in data analysis and forecasting technology in the power industry and provided new impetus for technological progress in the industry. It has safeguarded the legitimate rights and interests of users and avoided unnecessary legal disputes through reasonable risk assessment and collection strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0032] Figure 1 An overall flow chart of a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment provided by the first embodiment of the present invention;

[0033] Figure 2 A flowchart of specific method steps in a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment provided by a second embodiment of the present invention;

[0034] Figure 3 A model diagram of D-PMU measurement range expansion scenario 1 in a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment provided by a second embodiment of the present invention;

[0035] Figure 4 A model diagram of D-PMU measurement range expansion scenario 2 in a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment provided by a second embodiment of the present invention;

[0036] Figure 5 A second embodiment of the present invention provides a method for diagnosing quality of electricity fee recovery data and assessing user behavior risk in a measurement integrity verification algorithm process based on an object model range. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0038] Example 1, reference Figure 1 , which is an embodiment of the present invention, provides a method for diagnosing the quality of electricity fee recovery data and assessing user behavior risks, including:

[0039] S1: Acquire multi-source measurement data in a first platform, and characterize the multi-source measurement data.

[0040] The multi-source measurement data includes data information collected from different measurement points at the user in the first platform; at the same time, the multi-source measurement data is subjected to bad detection, and good data is retained.

[0041] It should be noted that bad data and normal data show a weak correlation in time series. When the power system is operating normally, the measured values ​​at adjacent time points change slightly, and it can be considered that the difference in the measured values ​​is mainly composed of noise signals. When bad data exists, this difference in measured values ​​will contain mutation signals and noise signals.

[0042] In the embodiment of the present application, the first platform refers to the electricity fee recovery platform, and the actual distribution network measurement data set is composed of multiple sources, including a data acquisition and monitoring system, an advanced measurement system, and a micro-synchronous phasor measurement device.

[0043] In an optional embodiment, the first platform may be an electric vehicle charging management platform, which is used to monitor and manage the operation of multiple charging stations and the status of charging piles. Under this platform, the collection of multi-source measurement data may include the following data sources and data types: Charging pile operation data, including electrical parameters such as power output, voltage, current, and charging status (charging, standby, fault, etc.) of each charging pile, to help the platform monitor the status of each charging pile in real time. User charging behavior data, collected from the user's charging records, including the start and end time of each charging, charging amount, charging time, user account information, payment method, etc. These data can be used to analyze the user's charging habits and optimize the efficiency of charging piles. Charging station environmental data, including environmental data such as temperature, humidity, and air pressure around the charging station.

[0044] In the present application, characterization includes integrating and processing benign multi-source measurement data so that the data integrated from different measurement points reflects the characteristics of the user. The integration process includes, but is not limited to, integration based on time: synchronizing the data from different locations and different devices in time, and performing time delay correction and fusion on different measurement data. In the power system measurement of the present application, it specifically includes: first, the D-PMU measurement range is expanded, and after the expansion is completed, multiple measurement data are aligned. The D-PMU measurement has a high-precision time tag, and its sampling frequency can be as low as milliseconds, so that accurate remote synchronous measurement can be performed. In contrast, SCADA measurement data only carries the time stamp of the master station receiving the data. In order to ensure the consistency of measurement data from different sources in the time dimension, this paper uses D-PMU measurement data as a benchmark, and adjusts the alignment of SCADA and AMI data according to the time tag of D-PMU. In addition to D-PMU measurement, the remaining measurements still have the problem of being out of sync in different places after alignment. Therefore, it is necessary to fuse the multi-source measurement data at the state estimation moment.

[0045] In other optional embodiments, the characterization can also be spatial characterization of data from different geographical locations, so that the data can reflect the impact of spatial distribution and geographical location on power load or energy consumption. For example, the power load data is integrated by region to generate load characteristics of each region, which are further used to analyze the load distribution trend and identify abnormal areas. With the assistance of geographic information system (GIS) data, environmental information (such as temperature, humidity, weather conditions, etc.) of different regions can be associated to obtain detailed energy consumption characteristics of each region. Or event-driven characterization, in the power system, multi-source measurement data is characterized according to specific events (such as peak power consumption, equipment failure, extreme weather, etc.). By marking the data characteristics of specific events and establishing an event response model, it is possible to respond more quickly when similar events occur in the future. For example, the voltage, frequency and other data during the accident are recorded and compared with the data in the normal state to generate a data model with event characteristics.

[0046] S2: Improve the data performance of the characterized multi-source measurement data and perform quality diagnosis on the improved data.

[0047] The improvement of data performance includes optimizing the data step by step according to the results of the characteristic performance by identifying the characteristic performance of the data in different dimensions. Further, in the embodiments of the present application, it specifically includes optimizing the integrity of the data by measuring the integrity check based on the object model range in the dimension of data integrity; optimizing the accuracy of the data by measuring the point check based on the sliding window in the dimension of data accuracy; and detecting and correcting the power imbalance problem in the second system by long-time scale power balance check in the dimension of the stability of the second system.

[0048] The second system is the supply system of the first platform. The second system directly provides services to users and uploads data during the service process to the first platform, providing a data basis for the first platform's interactive behavior with users. Under the premise that the embodiment of the present application is an electricity fee recovery platform, this second system is the power supply system in the power grid.

[0049] The quality diagnosis includes obtaining the user's data analysis results through a data analysis algorithm, and obtaining the quality diagnosis results by integrating the analysis results.

[0050] Data quality assessment is a key step and basic link to ensure the effectiveness of data quality management. Based on the assessment results, the root causes of data quality problems are analyzed in depth, and corresponding improvement measures are proposed accordingly. These measures involve necessary corrections and optimizations to the measurement data reported by the source to improve data quality, which is also the core goal of data quality diagnosis method research.

[0051] In the embodiments of the present application, principal component analysis (PCA) is used, which is a statistical technique for data dimension reduction and pattern recognition. In multi-source measurement data quality diagnosis, PCA can help identify the main components in the data, thereby evaluating the data quality. The user's principal component matrix is ​​obtained, and the original data is multiplied by the principal component matrix to obtain the principal component score; the principal component score and the corresponding weight are summed and integrated to obtain a comprehensive quality diagnosis result.

[0052] In an optional implementation, a quality diagnosis method based on time series analysis and aggregate scoring is adopted to obtain comprehensive quality diagnosis results by analyzing the user's historical data and multidimensional data. For the user's historical data (such as the user's electricity consumption behavior data, power consumption trend, abnormal conditions, etc.), time series decomposition is first applied to decompose the data into trend, seasonality and residual components to help identify the long-term change pattern and periodic fluctuations of the data. A trend model is constructed using methods such as ARIMA (autoregressive integrated moving average model) and LSTM (long short-term memory neural network). ARIMA is suitable for simpler time series analysis, while LSTM is suitable for complex and nonlinear historical data. In this way, possible quality problems in the future can be predicted and the typical behavior patterns of users can be identified. Abnormal behavior is identified through residual analysis. If a certain section of data deviates greatly from the historical trend, it may represent an abnormality in user behavior or system data, which is helpful for subsequent quality diagnosis. Based on the different historical behavior data of users and system operation data, the features of each dimension are extracted. All feature scores are weighted and summed according to the weights to obtain the user's comprehensive quality score.

[0053] S3: Classify users by identifying interaction behaviors between users and the first platform.

[0054] Furthermore, the AP algorithm can classify users into different categories by analyzing their payment behavior and time characteristics, which can help power companies better understand users' payment habits and formulate more effective electricity fee recovery strategies.

[0055] S4: Perform a first factor evaluation on the user in each classification result, and evaluate the first characteristic risk based on the evaluation result of the first factor.

[0056] Furthermore, the first factor assessment includes a calculation indicator that can reflect the first characteristic risk; the first characteristic risk is the risk that exists when the user interacts with the first platform.

[0057] In the embodiment of the present application, the evaluation result of the factor of electricity fee recovery behavior is the first characteristic risk of "risk at the time of payment", which will reflect the credit problem. Then the user's credit is the indicator that needs to be calculated, that is, the first factor evaluation result.

[0058] In an optional implementation, on the electric vehicle charging management platform, users use the platform to make charging reservations, real-time charging monitoring, and payment settlements. One of the main interactive behaviors of users is to make reservations for charging piles. In this interactive behavior, the main first characteristic risk is the user's reservation default risk, that is, the risk that the user does not arrive at the charging pile on time or completes the charging after making a reservation. This risk directly affects the utilization efficiency of the charging pile and increases the management cost of the platform. In order to evaluate the user's reservation default risk, the reservation fulfillment rate can be used as the first factor. The level of the fulfillment rate can reflect whether the user fulfills the reservation behavior on time, and further reflects the user's usage habits and the reliability of the platform interaction.

[0059] On the other hand, this embodiment also provides a quality diagnosis system for electricity fee recovery data and a user behavior risk assessment system, which includes:

[0060] The acquisition unit acquires multi-source measurement data in the first platform and characterizes the multi-source measurement data. The diagnosis unit improves the data performance of the characterized multi-source measurement data and performs quality diagnosis on the improved data. The classification unit classifies users by identifying the interaction behavior between the users and the first platform. The risk assessment unit performs a first factor assessment on the users in each classification result, and assesses the first feature risk according to the assessment result of the first factor.

[0061] Example 2, reference Figure 2-5 , which is an embodiment of the present invention, provides a specific method for data quality diagnosis of electricity fee recovery based on multi-source measurement information in a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment.

[0062] like Figure 2 The following is the flow chart of the entire algorithm. ①-⑧ Specific implementation process:

[0063] S1: Multi-source measurement data fusion.

[0064] The actual distribution network measurement data set is composed of multiple sources, including data acquisition and monitoring systems, advanced measurement systems, and micro-synchronous phasor measurement units (D-PMUs). A multi-source measurement data fusion strategy is proposed, which focuses on four main aspects: the detection of bad data solves the problem of preliminary preprocessing of multi-source measurement data and removes bad data outside the threshold; the expansion of the D-PMU measurement range solves the problem of limited number of current measurement channels, covers a wider area, and improves data diversity; the alignment of measurement data enables accurate remote synchronous measurement; and time delay fusion solves the data at the time when state estimation is required in multi-source measurement. These strategies are aimed at improving the accuracy and reliability of distribution network measurement data, thereby optimizing the monitoring and management of the power grid.

[0065] S1.1: Bad data detection: An adaptive detection mechanism is proposed, which can set an adaptive threshold to identify bad data while eliminating measurement errors. The calculation formula of this threshold is as follows:

[0066]

[0067] Among them, α k is the threshold at time k, σ Δ,k is the standard deviation of the equivalent measurement noise, Z avg,k is the mean value of the measurement in the sliding window at time k, σ L,k is the standard deviation of the absolute value of the difference between the measurements in the sliding window at time k. If the absolute value of the difference between the measurements at time k and k+1 is less than α k+1 , the measurement data at time k+1 is considered good; if this condition is not met, the measurement data at time k+1 is identified as bad data, and the mean of the remaining measurements in the sliding window is assigned to the measurement value at time k+1. This method can effectively identify and correct bad data and improve the accuracy and reliability of data.

[0068] S1.2: D-PMU measurement range expansion: PMU (distribution network micro-synchronous phasor measurement unit) is an advanced power system measurement technology. It is expanded and optimized based on the traditional PMU to adapt to the characteristics and needs of the distribution network.

[0069] In the power system, considering the limited number of current measurement channels of the D-PMU device, this paper discusses two different D-PMU measurement configuration scenarios. The scenarios are as follows. The voltage phasor of node w is marked in the figure. D-PMU voltage phasor measurements at node v and node w and And the actual current phasor of branch vw and the current phasor measured by the D-PMU There is also the impedance Z of the branch vw vw,imp .

[0070] In the first scenario, if Figure 3 As shown in Figure 1, a node w is equipped with a D-PMU at one end for voltage measurement, while the branch vw is also equipped with a D-PMU for current measurement. Based on these device configurations, the voltage phasor of the other end node w can be calculated.

[0071]

[0072] like Figure 4 As shown in Figure 2, in scenario 2, both nodes v and w are configured with D-PMU voltage measurement, but branch vw is not configured with D-PMU current measurement. The current phasor of branch vw needs to be calculated.

[0073] According to Ohm's law and Kirchhoff's voltage law in circuit theory, the current can be calculated by dividing the voltage difference between two nodes by the impedance. The specific formula is:

[0074]

[0075] It should be noted that in actual applications, the configuration and use of D-PMU is not limited to the above two scenarios. The main reason for selecting the above two scenarios for analysis is to demonstrate the application of D-PMU data in power grid state estimation, especially how to optimize the use of these data when resources are limited. These two cases represent two extreme cases of D-PMU data use: one is that both current and voltage are directly measured, and the other is that only voltage is measured without direct current measurement. Through the analysis of these two cases, we can better understand how to deal with different monitoring configurations in actual power grid monitoring, and how to use limited monitoring resources to maximize the observability and controllability of the power grid.

[0076] Other possible scenarios: 1. Single-ended configuration with voltage measurement only. 2. Configuration with current measurement only. 3. Mixed configuration. 4. No D-PMU.

[0077] The reason for choosing these two cases for analysis is mainly to show how to use the data of D-PMU to calculate or estimate the voltage and current phasors in the power grid under different configurations.

[0078] Scenario 1: Single-ended D-PMU current measurement, in this scenario, one end node of the branch is configured with D-PMU voltage measurement and current measurement. This means that accurate measurements of the voltage and current at that end can be obtained simultaneously. Based on Ohm's law and circuit theory, the voltage phasor of the other end node can be calculated.

[0079] Scenario 2: Two-terminal D-PMU voltage measurement, no current measurement, in this case, both ends of the branch are configured with D-PMU voltage measurement, but no current measurement. This means that only the voltage measurement values ​​at both ends can be obtained, and the current cannot be measured directly. The current phasor can still be calculated from the voltage difference and the branch impedance.

[0080] S1.3: Alignment of multiple measurement data.

[0081] S1.3.1: SCADA measurement alignment.

[0082] SCADA measurement equipment mainly collects data types such as branch power, voltage and current amplitude. Among them, there is a linear relationship between the branch current amplitude collected by the SCADA system and the current amplitude collected by the D-PMU system. Therefore, this paper uses the Pearson correlation coefficient to identify and filter out bad data and determine the best corresponding time tags between SCADA and D-PMU measurement data to achieve accurate alignment of the two data.

[0083] Consider a specific SCADA device whose master receives data at a time stamp of t r ; The delay range and sampling period of the device are τ W and T S In the time period [t r , t r +T S n s ], the master station receives n s The measurement data can be expressed as a vector Z S =[z s,1 ,z s,2 ,...,z s,n ].

[0084] The alignment steps are as follows:

[0085] Step 1: Screening and reorganization of D-PMU measurement data. r -τ W , t r ], the master station receives n d The D-PMU measurement is recorded as For each z s，1 , find the D-PMU measurement closest to its sampling time as the synchronous measurement. Then,P,1,1 、z P,2,1 , is the starting item of the sequence, according to the period T S Extract n from the total D-PMU measurement sequence s -1 measurement data, forming the D-PMU measurement matrix, the expression is as follows:

[0086]

[0087] Where: Z P is the D-PMU measurement matrix. p,nd,n Represents an element in the D-PMU measurement matrix. p It represents the D-PMU measurement matrix, which is related to the synchronous measurement (Phase). d Indicates that in the time period [t r -τ W ,t r ], the total number of D-PMU measurements received by the master station. n represents the number of D-PMU measurements received by the master station after each synchronous measurement according to the period T S The number of measurement data extracted, i.e., ns-1. In the D-PMU measurement matrix, Z p,nd,n The element at the ndth row and the nth column represents the period T in the ndth synchronous measurement sequence. S The values ​​of n measurement data extracted.

[0088] Step 2: Calculate the Pearson correlation coefficient. Calculate Z S With Z P The Pearson correlation coefficient of each row vector is expressed as follows.

[0089]

[0090] Where: Z P,j Z P The j-th row vector of d ρ s (j) and Cov(Z S ,Z P,j ) are Z S With Z P,j Pearson correlation coefficient and covariance of s and σ P,j Z S and Z P,j The standard deviation of .

[0091] Step 3: Get the synchronized D-PMU measurements for SCADA, as shown in the following equation.

[0092]

[0093] Where: J is Z P Zhong and Z S The number of the row vector with the largest Pearson correlation coefficient.

[0094] Step 4: Get the sampling time stamp of SCADA.

[0095] Calculate the SCADA measurement delay as:

[0096] τ S =τ W -T P J (7)

[0097] Where: τ S Z S The measurement delay of P is the sampling period of D-PMU measurement.

[0098] Z P The time scale is given to Z S , that is, t 0 =t r -τ S (8)

[0099] Where: t 0 Z S Middle Z s，1 The sampling time stamp.

[0100] S1.3.2: AMI measurement alignment.

[0101] The communication network of AMI (Advanced Metering Infrastructure) adopts a two-way communication standard, which enables the energy meter to send power consumption information to the power company in real time, and the power company can also send control signals to the energy meter. This two-way communication capability greatly improves the intelligence and automation level of the power grid. The measuring device in the AMI system has a built-in remote calibration clock, which can ensure the time accuracy of the power consumption information, thereby ensuring the synchronization of the data. When these time-stamped data are uploaded to the master station, the master station can use this time stamp to align the data and achieve time synchronization between different data sources, which is crucial for the operation monitoring and analysis of the power grid.

[0102] In smart distribution networks, the measurement accuracy and sampling frequency of D-PMU are higher than traditional measurement methods. Through the time synchronization signal provided by GPS or Beidou system, D-PMU can realize real-time synchronous monitoring of distribution network operation information, including the collection of phasor information such as current and voltage, as well as the calculation and acquisition of information such as power, phase, and power angle. This makes D-PMU applicable in many fields such as state estimation, fault diagnosis, and coordinated control of distribution networks.

[0103] After receiving the power consumption information from AMI, the master station can align the data based on the time stamp of AMI data and the time stamp of D-PMU data. This data alignment is particularly important for the operation status estimation of the distribution network, because it can ensure the consistency of data from different sources in time, thereby more accurately reflecting the actual operation of the distribution network. Through this alignment, the accuracy of state estimation can be improved, and precise control and optimized management of the distribution network can be achieved.

[0104] S1.4: Delay fusion of multi-source measurement data.

[0105] Except for D-PMU measurement, the remaining measurements still have the problem of being out of sync in different locations after alignment. Therefore, it is necessary to fuse the multi-source measurement data at the state estimation time.

[0106] S1.4.1: The measurement accuracy of the SCADA measurement system is relatively low. According to the theory of error propagation, if extrapolation or other prediction techniques are used to integrate data, the error of the integrated data may increase. Therefore, it is chosen to use the SCADA sampling data closest to the state estimation time point for direct calculation to avoid error propagation. In this way, the fusion value of SCADA data can be directly calculated using the following formula:

[0107] Z SE,S (t) = Z M,S (t S,pre ) (9)

[0108] Where: Z SE,S (t) is the measurement fusion value of SCADA at time t; Z M,S (t S,pre ) is SCADA in t S,pre The measurement sequence at time t S,pre is the last collection time of SCADA before time t.

[0109] S1.4.2: The sampling period of AMI measurement may be as long as several minutes, which may lead to an increase in measurement errors. In order to reduce this error, a specific time window is set, the length of which is equal to the pseudo-measurement prediction period. If the AMI measurement data can arrive within this time window before the critical time point of state estimation, these actual measurement data are used to calculate the state estimation. If the AMI measurement data does not arrive within the time window, the pseudo-measurement data generated based on the pulse neural network method will be used for state estimation. This method is used to estimate the pseudo-measurement value of the node injected power, thereby ensuring the accuracy and reliability of the state estimation. The calculation formula of the AMI fusion value is as follows:

[0110]

[0111] Where: Z SE,A (t) is the measurement fusion value of AMI at time t; Z M,A (t A,pre ) is AMI at t A,pre The measurement sequence at time t A,pre is the last collection time of AMI before time t; Z M,P (t) is the pseudo-measurement prediction value at time t; t SE,pre is the time of last state estimation; T W is the time window width.

[0112] S2: Multi-source measurement data quality diagnosis method.

[0113] The data quality diagnosis method proposes a measurement integrity verification algorithm based on the object model range to verify whether the measurement data is within a reasonable range; the measurement point verification algorithm based on the sliding window is used to quickly screen missing points and jump points in the data; the long-time scale power balance verification algorithm is used to identify abnormal problems and ensure the stable operation of the power system; finally, the principal component analysis method is used to calculate the comprehensive score of the data measurement points.

[0114] S2.1: Use a measurement integrity verification algorithm based on the scope of the object model.

[0115] The measurement integrity verification algorithm based on the object model range is a method for verifying whether the measurement data is within a reasonable range. This algorithm is usually used in power systems, industrial automation and other fields that require precise measurement. Figure 5Used to illustrate how to design and implement such a verification algorithm. (1) Data object modeling. By using a structured design method to model the general data objects of power dispatching, including the relationship, ID coding, metadata definition and data dictionary of data objects, the structure and relationship of data objects in the power system are defined. (2) Analysis of the basic information model of the object. Use the object model to perform association analysis and obtain the scope of the object model to determine which measurement point data should be included and the expected scope of the measurement data. (3) Management relationship analysis. Analyze the dispatching management relationship, operation and maintenance management relationship, and the subordinate relationship between equipment containers and equipment to divide the scope of measurement points for reporting data of power dispatching centers in various regions and clarify the relationship between dispatching organizations, operation and maintenance organizations and equipment. (4) Determination of measurement point scope. According to the results of management relationship analysis, determine the equipment and measurement points that each dispatching center is responsible for, and obtain the scope of measurement point data that should be reported within the jurisdiction of each power dispatching center. (5) Measurement point quantity statistics. According to the voltage level dimension, the expected number of measurement points and the actual number of measurement points received are counted respectively. (6) Calculation of data source completeness rate. By comparing the number of measurement points that should be uploaded and the number that are actually uploaded, the integrity rate of each data source is calculated, and the measurement integrity of each data source is quantitatively evaluated. (7) Verification result analysis. Analyze the verification results, identify cases where data is missing or incomplete, evaluate the calculated integrity rate, and determine whether measures need to be taken.

[0116] S2.2: The sliding window-based measurement point verification algorithm is an efficient detection method designed for large-scale time series data. It implements incremental processing by gradually moving a fixed-size window on the data. The algorithm focuses on quickly screening missing points and jump points to adapt to the high time correlation and strong computational timeliness requirements of the data. In each window, the algorithm only processes new data points, avoiding repeated calculations and significantly improving efficiency. At the same time, it can detect and process outliers in the data in a timely manner, such as identifying missing points and jump points through statistical methods or machine learning models, and performing data repair or anomaly marking. In addition, the algorithm has real-time update capabilities to ensure that the verification process continues to reflect the latest data status as new data continues to flow in. In order to optimize performance, parallel computing, hardware acceleration and other technologies can be used, and feedback mechanisms can be introduced to adjust parameters to adapt to changes in data characteristics. This algorithm is particularly suitable for application scenarios that require real-time data processing, such as financial transaction monitoring and industrial production process control, and effectively guarantees data quality and the accuracy of analysis results.

[0117] Assume that the time series D is an ordered set of elements consisting of measurement time and measurement value, D = {(t 0 , d 0 ), (t 1 , d 1 ),…,(t n , d n)}, define a sliding window S with a time interval of w w [t, t+w]. Under normal circumstances, the measured values ​​in w are continuous and the number of measured values ​​is fixed. When w is very small, the change of D will not be large, and it is basically evenly distributed near the mean of all D values ​​in the current window. When the measured value D of a certain measurement point m deviates significantly from the mean of the current window, m is likely to be a jump point. Therefore, this paper sets thresholds for the daily measurement sequence of the measurement point in different time periods based on the sliding window to determine the jump point.

[0118] The specific process of the measurement point calibration algorithm based on sliding window is as follows.

[0119] (1) Select the sliding window S w The time interval w size is used to place the sliding window Make its spatial center coincide with the point to be checked.

[0120] (2) Determine the validity of the point to be measured. If the data of the point to be measured is not null or an illegal value, the point is considered to be a valid measurement point and can continue to participate in the calculation. Otherwise, the point is considered to be a missing point and is included in the missing point sequence.

[0121] (3) Calculate the mean of the measurement sequence values ​​within the sliding window The formula is:

[0122]

[0123] (4) Calculate the spatial distance of the measurement point to be verified from the sliding window measurement mean Right now

[0124]

[0125] (5) Divide into different time intervals, extract the measurement feature values ​​statistically, and set the time interval threshold sequence τ = {τ 1 , τ 2 ,…,τ n}.

[0126] (6) Compare the spatial deviation distance with the corresponding time interval threshold. When , the measurement point to be calibrated is determined to be a normal value, otherwise it is considered that the measurement point to be calibrated is a jump point.

[0127] S2.3: Assume that the equipment sequence includes components such as AC line terminals, transformer windings, and converters. According to the topological structure of the equipment in the plant, these equipment are classified by voltage level. In the same plant, all equipment with the same voltage level are grouped together, D vol = {V ac ,V tf ,Vct}, where the AC line voltage is set to V ac = {V ac,1 ,V ac,2 ,...,V ac,n}、The transformer winding voltage is set to V tf = {V tf,1 ,V tf,2 ,...,V tf,n}、The converter voltage is set to V ct = {V ct,1 ,V ct,2 ,...,V ct,n}, the time series is set to V = {(t 0 ,v 0 ),(t 1 ,v 1 ),...,(t n ,v n )}.

[0128] Find D according to the measurement point calibration algorithm based on sliding window vol The non-empty valid time point sequence of all devices in is denoted as T = {T ac ,T tf ,T ct}, where T ac,i ={t 0 ,t 1 ,...,t n} is the AC line end time series, similarly T tf,i ={t 0 ,t 1 ,...,t n} is the transformer winding time series, T ct,i ={t 0 ,t 1 ,...,t n} is the converter time series, and the effective time point sequence T′ of all power balance checks in the plant is calculated. The number of effective points is the length len(T′) of the effective time sequence T′.

[0129]

[0130] For each type of equipment in the power system (including AC line terminals, transformer windings, and converters), the deviation between their inflow and outflow active power is calculated at a given effective time point. Transformer windings Inverter Then add the active power deviations of all devices in the same effective time section to get The calculation formula is:

[0131]

[0132] In the power system, in order to analyze the characteristics of active power, it is necessary to extract the key characteristic values ​​related to it. To this end, a threshold sequence τ is set, which contains the power deviation thresholds allowed under different voltage levels. Specifically, τ is composed of different elements, each element corresponds to a specific voltage level, such as τ 3 Represents the threshold of 3kV voltage level, τ 66 represents the threshold value of 66kV voltage level, and so on, until τ 1000 , which represents the threshold value of 1000kV voltage level. These threshold values ​​help to quantitatively analyze whether the power deviation at each voltage level is within the acceptable range.

[0133] In power system analysis, for each valid time section, the active power deviation of the plant needs to be calculated. Then, this deviation is compared with the preset power deviation threshold τ vol For comparison. Less than τ vol , it is considered that there is no active power imbalance problem in this time section. On the contrary, if Greater than or equal to τ vol , then the time point is identified as the active power unbalance point and recorded in a file called P s ' t This sequence contains all the time points determined as power imbalance and their corresponding deviation values, in the form of Finally, the sequence P s ' t The length of the power imbalance point is the number of points, expressed in mathematical terms as len(P s ' t ). This process is crucial for detecting and analyzing active power balance problems in power systems.

[0134] Calculate the total active power deviation of the plant and deviation quantification index The formula is:

[0135]

[0136] In the analysis of active power in power system, it is feasible to use different time scales to calculate the quantitative index of active power deviation. Considering the characteristics of measurement data collection, daily data at the minute level is usually selected for verification. This method can not only effectively reveal the daily periodic changes of active power in power system, but also ensure the speed and efficiency of calculation. In the implementation process, the ratio of power imbalance points can be obtained through the calculated deviation quantitative index, which can characterize whether there is a quality abnormality problem in the target measurement data. At the same time, by analyzing the total active power deviation and the value of the target measurement data point, the type of abnormal problem in the target measurement data can be identified. For the power balance verification of primary equipment such as lines and transformers, the deviation quantitative index is obtained by calculating the inflow and outflow active power at each end of the equipment in the same time section, which is also an effective power balance verification method. By comparing the measurement point information with the basic information model of the object, the model number of the line end and the transformer winding and the number of measurement points of the actual reported data can be analyzed, so as to quickly locate data quality problems such as model maintenance errors, data omissions or underreporting. In general, choosing the right time scale to calculate deviation quantification indicators is crucial for active power analysis of power systems. It can help us evaluate data quality, identify abnormal problems, and ensure the stable operation of power systems.

[0137] S2.4: Data quality assessment is a key step and basic link to ensure the effectiveness of data quality management. Based on the assessment results, the root causes of data quality problems are analyzed in depth, and corresponding improvement measures are proposed accordingly. These measures involve necessary corrections and optimizations to the measurement data reported by the source to improve data quality, which is also the core goal of data quality diagnosis method research.

[0138] Principal component analysis (PCA) is a statistical technique used for data dimensionality reduction and pattern recognition. In multi-source measurement data quality diagnosis, PCA can help identify the main components in the data and thus assess the data quality. The following are the specific steps and formulas for using PCA to assess data quality:

[0139] Step 1: Data preprocessing

[0140] Data standardization: Standardize the raw data to eliminate dimensional differences.

[0141]

[0142] Among them, x ij is the jth index value of the ith sample, and j are the mean and standard deviation of the j-th indicator respectively.

[0143] Step 2: Calculate the covariance matrix

[0144] Calculate the covariance matrix: Calculate the covariance matrix for the standardized data.

[0145]

[0146] Among them, z i is the standardized data vector of the ith sample, is the sample mean vector.

[0147] Step 3: Solve for eigenvalues ​​and eigenvectors

[0148] Solve the eigenvalue and eigenvector: Perform eigendecomposition on the covariance matrix C to obtain the eigenvalue λ 1 ,λ 2 ,...,λ n and the corresponding eigenvector e 1 ,e 2 ,...,e n .

[0149] Step 4: Select the principal components

[0150] Select the principal component: According to the size of the eigenvalue, select the eigenvectors corresponding to the first k eigenvalues ​​as the principal component {e 1 ,e 2 ,...,e k}.

[0151] Step 5: Calculate principal component scores

[0152] Calculate the principal component scores: Multiply the original data by the principal component matrix to obtain the principal component scores.

[0153] Y=Z·E (19)

[0154] Among them, Z is the standardized data matrix, E is the principal component matrix composed of the first k eigenvectors, and Y is the principal component score matrix.

[0155] Step 6: Comprehensive Rating

[0156] Comprehensive score: Calculate the comprehensive score R based on the principal component scores and corresponding weights (usually the variance contribution rate of the principal components).

[0157]

[0158] Among them, ω i is the weight of the i-th principal component, y i is the score of the i-th principal component.

[0159] The key to PCA is dimensionality reduction, which is to approximate most of the variability in the original data set by selecting fewer principal components. This not only helps to simplify the data structure, but also removes noise and redundant information, thereby improving the efficiency and accuracy of data analysis.

[0160] Example 3 is an embodiment of the present invention, which provides a specific implementation method of electricity fee recovery risk assessment in a method for quality diagnosis of electricity fee recovery data and user behavior risk assessment.

[0161] It should be noted that the application of long short-term memory network (LSTM) in electricity bill recovery risk analysis is an advanced method that integrates the characteristics of data with different time spans. It uses the deep learning ability of LSTM to accurately assess the risks in the electricity bill recovery process and provide effective electricity bill collection strategies based on this. This method can significantly improve the accuracy of risk prediction, thereby optimizing the electricity bill recovery process.

[0162] The risk assessment of electricity bill recovery first uses the AP algorithm to classify electricity users. After classification, it uses various credit-related data of users, such as credit records, judicial decision information, etc., to conduct a comprehensive assessment and uses the entropy weight method to construct a credit assessment matrix. The application of LSTM in electricity bill recovery risk analysis can help the model learn the behavior patterns and credit history of electricity users, thereby predicting future payment behavior.

[0163] As in Example 2, Figure 2 ⑨- shown Specific implementation steps for risk assessment of electricity bill recovery.

[0164] Electricity customers classification:

[0165] Affected by economic and corporate operating habits, electricity bill payment behavior shows significant individual differences. In order to more accurately understand the payment behavior of different users, it is possible to classify and categorize electricity users in detail. This classification method can reveal the inherent patterns of user payment behavior.

[0166] Affinity Propagation Clustering (AP) is a clustering algorithm based on message passing, which determines the cluster center and the ownership of data points by passing information between data points. This method does not require the number of clusters to be specified in advance and is applicable to multi-dimensional and multi-category data.

[0167] First, the electricity bill payment similarity matrix K(a) s,bIt is obtained by weighting the Euclidean distance from the discrete characteristics of the electricity bill payment time of two electricity users as and bs to the index center cluster point z. ω represents the weight of the discrete characteristic value. The following formula defines the similarity matrix:

[0168] K(a) s,b =ωL as +(1-ω)L bs (twenty one)

[0169] Where: L as , L bs are the Europeanized distances from the discrete characteristics of the electricity bill payment time of electricity customers as and bs to the indicator center cluster point z; ω is the discrete feature value weight.

[0170] It should be noted that the electricity bill payment similarity matrix is ​​calculated based on the Euclidean distance from the discrete characteristics of the electricity bill payment time between electricity customers to the indicator center cluster point and the discrete characteristic value weight.

[0171] Specifically, the calculation of this matrix involves the following key steps:

[0172] Discrete feature extraction: First, it is necessary to extract time-discrete features from the payment data of electricity users. These features include payment frequency, time point, payment amount, etc.

[0173] Euclidean distance calculation: For two electricity users as and bs, calculate the Euclidean distance from their time discrete features to a certain index center cluster point z. This distance L as and L bs The similarity of the payment characteristics of two customers can be quantified.

[0174] Weight adjustment: The influence of the Euclidean distance is adjusted by the discrete eigenvalue weight ω. This weight is set according to the importance of the payment feature or the reliability of the payment data.

[0175] Similarity matrix construction: Use the formula K(a)s,b=ωL as +(1-ωL bs ) to construct the initial similarity matrix. This formula combines the payment characteristics of the two customers and takes into account the influence of weights.

[0176] Next, we use iterative optimization: optimize the clustering process through multiple iterations of the F index and update the similarity matrix.

[0177] In the clustering process, the F index of multiple iterations is used as the bias function and convergence basis. After clustering, the similarity matrix K(a) s,b Updated to:

[0178] K(a) s,b =Ca +Δf×F min (twenty two)

[0179] Among them, C a is the diagonal median of the user electricity bill payment similarity matrix, Δf is the iterative retrieval threshold, and F min is the minimum value of multiple iterations. Through the above formula, the similarity matrix K(a) can be updated s,b , thereby obtaining the electricity fee recovery classification of electricity users.

[0180] User credit assessment:

[0181] Credit assessment plays a key role in predicting electricity bill recovery. The credit rating of a customer is closely related to the timeliness of his or her payment: customers with high credit ratings usually pay more promptly; conversely, customers with low credit ratings may be at risk of defaulting on payments. Due to the different characteristics and credit environments of users in different regions, the weights of credit assessment indicators will also vary.

[0182] After classifying the electricity users, the entropy weight method is used to construct the credit evaluation matrix. This method uses various credit-related data of users, such as credit records, judicial ruling information, etc., to conduct a comprehensive evaluation. In order to ensure the comprehensiveness and objectivity of the evaluation, multiple key indicators are selected, and the specific indicators are shown in Table 1. Through this multi-dimensional evaluation, the credit status of users can be judged more accurately, thereby improving the prediction accuracy of electricity fee recovery.

[0183] Table 1 Credit assessment weights for electricity customers

[0184] Weight Feature indicators Value range <![CDATA[W 1 ]]> User credit Number of breaches <![CDATA[W 2 ]]> judicial adjudication Number of judgments <![CDATA[W 3 ]]> Bank credit grade <![CDATA[W 4 ]]> Arrears mark Number of breaches <![CDATA[W 5 ]]> Urban and rural signs Urban / rural <![CDATA[W 6 ]]> Voltage level 220V~110kV

[0185] When constructing a credit evaluation matrix for electricity users, the entropy weight method is an effective tool that can determine the weight of each evaluation indicator, thereby more accurately reflecting the importance of each indicator in the credit evaluation. This paper selects six indicators to construct the evaluation matrix and performs standardization and weight calculation through the following steps:

[0186] Data normalization: First, the original data U is normalized to V using the following formula:

[0187]

[0188] Among them, n is the total number of indicators, and min(U) is the minimum value of indicator U.

[0189] Calculate information entropy: For each indicator, calculate its information entropy λ, the formula is:

[0190]

[0191] where p iis the probability distribution of the standard center value p.

[0192] Calculate indicator weights: Calculate the weight W of each indicator based on information entropy i , the formula is:

[0193]

[0194] Here, W i Represents the weight of the i-th indicator.

[0195] The weights calculated through the above steps can be used to comprehensively evaluate the credit status of electricity users. This entropy-based evaluation method can objectively reflect the importance of each indicator in credit evaluation and help improve the accuracy and reliability of credit evaluation.

[0196] Electricity fee recovery risk analysis:

[0197] Long short-term memory (LSTM) is a special type of recurrent neural network designed to process sequence data and is particularly good at capturing long-term dependencies. Applying LSTM in electricity bill recovery risk analysis can help the model learn the behavior patterns and credit history of electricity customers, thereby predicting future payment behavior.

[0198] Input data and implicit information:

[0199] Input data set Z a = {G a ,S a ,Q a}, where G a S is the credit evaluation data of the input electricity customer; a Historical payment data of electricity users; Q a Behavioral characteristic data of electricity users.

[0200] The total implicit information of the input is L m =L g +L q +L s ,

[0201] Where L g is the implicit information of the input credit rating; L s It contains the historical payment data of electricity users; L q It is the implicit data of the behavioral characteristics of electricity users.

[0202] Risk analysis data calculation:

[0203] The calculation formula for the risk analysis data To of electricity cost recovery of electricity users is:

[0204] T 0 =Δa([Za ,L m ])×tanh(Z a ) (26)

[0205] Where: Δa is the input activation function of the LSTM network; tanh is the forget gate storage activation function.

[0206] As can be seen from the above formula, the risk analysis data of electricity bill recovery can be obtained through the LSTM network. On this basis, according to the electricity bill recovery risk assessment score level of the power supply company, a collection suggestion list and collection suggestion are generated. The electricity bill collection suggestion is divided according to the risk level of the user. Level 1 risk users belong to the high-risk group. Such users may be unable to pay the electricity bill due to the break of the capital chain. Therefore, it is recommended to take emergency collection measures immediately. Level 2 risk users face medium risks. Although their capital chain operates normally, they have a history of arrears in electricity bills, so timely collection is required to avoid the deterioration of arrears. Level 3 risk users are in a state of potential risk in the capital chain. In this regard, the power supply company should issue a reminder to encourage users to pay on time. Level 4 risk users are low-risk or no-risk users. Their payment behavior is normal and there is no arrears in electricity bills. Therefore, no special attention or collection is required. For different levels of risk users, the marketing department of the power supply company should adopt differentiated management strategies to improve the efficiency of electricity bill recovery and ensure the stability of power supply.

[0207] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for quality diagnosis of electricity fee recovery data and user behavior risk assessment, characterized in that: include: Acquire multi-source measurement data in the first platform, and characterize the multi-source measurement data; Improve the data performance of multi-source measurement data after characterization, and perform quality diagnosis on the improved data; Classifying users by identifying interaction behaviors between users and the first platform; A first factor evaluation is performed on the user in each classification result, and a first characteristic risk is evaluated based on the evaluation result of the first factor.

2. The method for quality diagnosis of electricity fee recovery data and user behavior risk assessment according to claim 1, characterized in that: The multi-source measurement data includes data information collected from different measurement points at the user in the first platform; At the same time, the multi-source measurement data is subjected to bad detection, and good data is retained.

3. The method for quality diagnosis of electricity fee recovery data and user behavior risk assessment according to claim 2, characterized in that: The characterization includes integrating and processing benign multi-source measurement data so that the data integrated from different measurement points reflects the characteristics of the user; The integration processing includes, but is not limited to, integration based on time: synchronizing the data from different locations and different devices in time, and performing time delay correction and fusion on different measurement data.

4. The method for quality diagnosis of electricity fee recovery data and user behavior risk assessment according to claim 3, characterized in that: The improvement of data performance includes identifying characteristic performance of data in different dimensions and optimizing the data step by step according to the results of the characteristic performance; The quality diagnosis includes obtaining the user's data analysis results through a data analysis algorithm, and obtaining the quality diagnosis results by integrating the analysis results.

5. The method for quality diagnosis of electricity fee recovery data and user behavior risk assessment according to claim 4, characterized in that: The first factor assessment includes an accounting indicator that can reflect the first characteristic risk; The first characteristic risk is the risk that exists when the user interacts with the first platform.

6. The method for quality diagnosis of electricity fee recovery data and user behavior risk assessment according to claim 4, characterized in that: The identification data has characteristic performances in different dimensions, and the data is optimized step by step according to the results of the characteristic performances, including but not limited to, in the dimension of data integrity, optimizing the integrity of the data through measurement integrity verification based on the scope of the object model; In terms of data accuracy, the accuracy of data is optimized through measurement point calibration based on sliding windows; In the dimension of the stability of the second system, the power imbalance problem in the second system is detected and corrected by long-time scale power balance verification; The second system is the supply system of the first platform. The second system directly provides services to users and uploads data during the service process to the first platform, providing a data basis for the first platform's interactive behavior with users.

7. The method for quality diagnosis of electricity fee recovery data and user behavior risk assessment according to claim 6, characterized in that: The data analysis results of the user are obtained by the data analysis algorithm, and the analysis results are integrated, including but not limited to using PCA to perform data dimension reduction and pattern recognition to obtain the user's principal component matrix, and the original data is multiplied by the principal component matrix to obtain the principal component score; The principal component scores and corresponding weights are summed and integrated to obtain a comprehensive quality diagnosis result.

8. A system for diagnosing quality of electricity fee recovery data and assessing user behavior risk using the method according to any one of claims 1 to 7, characterized in that: A collection unit, which obtains multi-source measurement data in the first platform and characterizes the multi-source measurement data; The diagnostic unit improves the data performance of the characterized multi-source measurement data and performs quality diagnosis on the improved data; a classification unit, which classifies users by identifying interaction behaviors between users and the first platform; The risk assessment unit performs a first factor assessment on the user in each classification result, and assesses the first characteristic risk according to the assessment result of the first factor.

9. A computer device comprising: A memory and a processor; the memory stores a computer program, wherein the processor implements the steps of any method as claimed in claim 1 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images