Method, system and equipment for identifying abnormal electricity utilization behaviors of power distribution network courts based on multi-dimensional perception and storage medium

Abstract

The invention relates to the technical field of intelligent monitoring of a power distribution network, in particular to a method, a system and equipment for identifying abnormal power utilization behaviors of a power distribution network district based on multi-dimensional perception and a storage medium. Intelligent sensing terminals are deployed on a transformer side, an outgoing line side and a user side of a transformer area to acquire electrical quantity parameters, electric energy quality parameters and environmental parameters, and data acquisition frequency is dynamically adjusted according to a load rate of the transformer area; performing quality verification on the multi-dimensional perception data, extracting time sequence features and statistical features, constructing a power consumption behavior feature vector, performing clustering classification on users based on the feature vector, and establishing and periodically updating a user behavior baseline; constructing a deep neural network to output an abnormal probability, calculating a characteristic deviation degree based on a behavior baseline, and carrying out weighted fusion to obtain a comprehensive abnormal index; and when the comprehensive anomaly index exceeds a judgment threshold value, judging that the behavior is abnormal, calculating a branch power difference value by the associated transformer area topological structure, and analyzing an anomaly time characteristic to complete anomaly positioning and traceability.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for power distribution networks, and in particular to a method, system, device, and storage medium for identifying abnormal electricity consumption behavior in power distribution network areas based on multi-dimensional perception. Background Technology

[0002] As the final link in the power system, distribution network areas directly provide power supply services to users. Traditional power management in distribution areas mainly relies on transformer monitoring terminals and smart meters for data collection. However, the collection points are sparse, the collection frequency is fixed, and the types of data obtained are limited, including only basic electrical quantities such as voltage and current. However, this single sensing method is difficult to fully reflect the actual operating status of the distribution area and cannot collect in-depth power quality parameters such as power factor, harmonic content, and three-phase imbalance. This results in insufficient precision in the perception of the operating status of the distribution area and affects the accuracy of distribution area management.

[0003] Currently available methods for identifying abnormal electricity consumption mainly employ fixed threshold alarms, such as triggering alarms when the line loss rate exceeds 15% or the voltage exceeds a limit. However, this method relies on manual experience to set thresholds, lacking flexibility. On one hand, fixed thresholds cannot adapt to the different electricity consumption characteristics of different types of users. Residential, commercial, and industrial users have different normal electricity consumption patterns, and most use a uniform standard for judgment, easily leading to a large number of false alarms or missed alarms. On the other hand, simple judgment based on thresholds is difficult to identify highly concealed abnormal behaviors, such as electricity theft using intelligent means or gradual equipment failures. These abnormal behaviors are often not characterized by a single parameter exceeding a limit, but rather by an overall deviation in the electricity consumption pattern. This results in existing methods lacking the ability to deeply analyze and learn electricity consumption behavior patterns, failing to establish refined electricity consumption behavior models, leading to weak anomaly identification capabilities and heavy reliance on manual experience. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by this invention is how to address the issues in the prior art, such as the limited sensing methods, incomplete data collection, weak anomaly identification capabilities, and inability to establish electricity consumption behavior models.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception. The method includes collecting electrical quantity parameters, power quality parameters, and environmental parameters from the transformer side, outgoing line side, and user side of the distribution network area to form multi-dimensional perception data. The data collection frequency is dynamically adjusted according to the load rate of the distribution area. When the load rate of the distribution area is lower than a first threshold, the collection frequency is reduced, and when the load rate of the distribution area is higher than a second threshold, the collection frequency is increased. The quality of multidimensional sensing data is verified and the temporal and statistical features of electricity consumption behavior are extracted. Electricity consumption behavior feature vectors are constructed, and users in the transformer area are clustered and classified based on the electricity consumption behavior feature vectors. The normal range of each feature parameter is calculated for each type of user, a user behavior baseline is established, and the user behavior baseline is updated regularly. An anomaly detection model is constructed, which combines the anomaly probability output by the deep learning network with the feature deviation calculated based on the user behavior baseline to obtain a comprehensive anomaly index. When the comprehensive anomaly index exceeds the judgment threshold, it is judged as an abnormal behavior. The branch power difference is calculated based on the topology of the associated transformer area, the abnormal time characteristics are analyzed, and the anomaly location and source are completed.

[0007] As a preferred embodiment of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in this invention, the method of collecting electrical quantity parameters, power quality parameters, and environmental parameters on the transformer side, outgoing line side, and user side of the distribution network area includes collecting the three-phase voltage, three-phase current, active power, reactive power, and power factor of the total load of the transformer on the low-voltage side of the transformer. Load distribution data for each outgoing line is collected at each outgoing point in the transformer area; detailed electricity consumption data is collected after the electricity meters of users whose electricity consumption capacity exceeds the set value; The collected data includes basic electrical quantities, power parameters, total harmonic distortion of voltage, total harmonic distortion of current, three-phase imbalance, temperature, and humidity.

[0008] As a preferred embodiment of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multidimensional perception as described in this invention, the method involves: performing quality verification on the multidimensional perception data and extracting the temporal and statistical features of the electricity consumption behavior to construct an electricity consumption behavior feature vector, including calculating the daily electricity consumption, average load rate, standard deviation of load rate, average power factor, and three-phase imbalance degree within the time window as statistical features. The load curve within the time window is divided into multiple time periods, and the average power of each time period constitutes the load curve feature vector as a time series feature. Combining statistical features and time-series features forms a complete feature vector of electricity consumption behavior.

[0009] As a preferred embodiment of the multi-dimensional perception-based method for identifying abnormal electricity consumption behavior in distribution network areas according to the present invention, the establishment of a user behavior baseline includes: using a clustering algorithm to cluster the electricity consumption behavior feature vectors of all users in the distribution area, classifying users into residential users, commercial users, and industrial and commercial users; for each type of user, calculating the mean and standard deviation of daily electricity consumption, average load factor, power factor, and three-phase imbalance; determining the normal range of each feature parameter based on the mean and standard deviation; and recalculating and updating the user behavior baseline monthly using the latest data.

[0010] The beneficial effects of this preferred technical solution are as follows: It achieves automatic classification of users in the transformer area through clustering algorithms, avoiding the one-size-fits-all management approach of traditional methods. The initial extraction of electricity consumption behavior feature vectors provides multi-dimensional criteria for user classification. In the intermediate stage, the mean and standard deviation of feature parameters are calculated for each type of user, establishing a normal range that conforms to the actual electricity consumption patterns of each type of user. Subsequently, the behavior baseline is updated monthly to ensure that the baseline always matches the current electricity consumption pattern, adapting to seasonal changes, climate fluctuations, and evolving electricity consumption habits. Overall, it achieves refined management of electricity consumption behavior, reducing false alarm and missed alarm rates caused by differences in user types.

[0011] As a preferred embodiment of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in this invention, the construction of the abnormal identification model includes: constructing a multi-layer deep neural network, where the network input is an electricity consumption behavior feature vector and the output is an abnormal probability; A deep neural network was trained using known abnormal samples as positive samples and normal electricity consumption data as negative samples. The user's current feature parameters are compared with the normal range of the user behavior baseline of the category, and the distance between the current value and the boundary of the normal range is calculated as the feature deviation. The anomaly probability and the feature deviation are weighted and fused to obtain a comprehensive anomaly index, wherein the weight of the anomaly probability is higher than the weight of the feature deviation.

[0012] The beneficial effects of this preferred technical solution are as follows: The construction of a deep neural network enables automatic learning of complex nonlinear electricity consumption patterns. In the early stage, the feature vector of electricity consumption behavior is used as the network input, enabling the model to capture the correlation between high-dimensional features. In the intermediate stage, positive and negative samples are used to train the deep neural network, enabling the model to identify hidden abnormal behaviors. At the same time, the feature deviation is calculated to provide rule-based interpretability. Subsequently, by weighted fusion of anomaly probability and feature deviation, the advantages of deep learning in pattern recognition and the stability of rule baseline are combined. The anomaly probability is given a higher weight to ensure sensitivity to complex patterns and improve the accuracy and reliability of anomaly identification.

[0013] As a preferred embodiment of the multi-dimensional perception-based method for identifying abnormal electricity consumption behavior in a distribution network area according to the present invention, the step of analyzing abnormal time characteristics and completing abnormal location and source tracing includes, when an abnormal user is identified, obtaining the total power of the branch where the abnormal user is located and the power of each user; and calculating the difference between the total power of the branch and the sum of the power of each user to obtain the branch power difference. The branch power difference is compared with the theoretical line loss value. When the branch power difference is significantly greater than the theoretical line loss value, it is determined that there is unmetered electricity consumption in the branch. The distribution of anomalies across time periods is statistically analyzed, and the duration between the first and last time an anomaly is detected is calculated as the duration of the anomaly.

[0014] As a preferred embodiment of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in this invention, the method further includes classifying the abnormal behavior according to the comprehensive abnormality index, location confidence, and duration of abnormality. For different levels of abnormal behavior, generate an anomaly tracing report that includes abnormal user information, anomaly feature description, related topology information, suspected anomaly type, and handling suggestions.

[0015] Secondly, embodiments of the present invention provide a power consumption anomaly identification system for distribution network areas based on multi-dimensional perception, which includes a data acquisition module for acquiring multi-dimensional perception data from the transformer side, outgoing line side and user side of the distribution network area, and dynamically adjusting the data acquisition frequency according to the load rate of the area. The data processing module is used to perform quality verification and normalization on multidimensional sensing data, and extract the temporal and statistical features of electricity consumption behavior. The user classification module is used to cluster and classify users in the transformer area, and to establish and regularly update user behavior baselines. The anomaly identification module is used to input electricity consumption data into a deep neural network to calculate the anomaly probability, calculate the feature deviation, and weightedly fuse them to obtain a comprehensive anomaly index. The location and source tracing module is used to calculate the branch power difference based on the topology of the associated transformer area, analyze the abnormal time characteristics, and complete the anomaly location and source tracing.

[0016] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in the first aspect of the present invention.

[0017] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in the first aspect of the present invention.

[0018] The beneficial effects of this invention are as follows: By deploying multi-level sensing terminals on the transformer side, outgoing line side, and user side of the distribution area, this invention achieves the coordinated collection of electrical quantity parameters, power quality parameters, and environmental parameters, solving the monitoring blind spot problem caused by traditional single sensing methods; combined with an adaptive collection strategy, it dynamically adjusts the data collection frequency according to the distribution area load rate, reduces the collection frequency during low load periods to save communication resources, and increases the collection frequency during high load periods to capture abnormal signs, avoiding resource waste or data loss caused by fixed frequency collection.

[0019] In the data processing stage, electricity consumption behavior feature vectors are constructed by extracting time-series and statistical features. Clustering algorithms are used to classify users into different types such as residential, commercial, and industrial and commercial. For each type of user, the normal range of feature parameters is calculated and a dynamically updated behavior baseline is established to solve the problem of false alarms and missed alarms caused by the traditional one-size-fits-all management method ignoring user differences.

[0020] In the anomaly identification phase, a deep neural network is constructed to output the anomaly probability. Simultaneously, the feature deviation based on the behavioral baseline is calculated. These two factors are weighted and fused to obtain a comprehensive anomaly index, combining the complex pattern recognition capabilities of deep learning with the interpretability of rule-based baselines. This overcomes the limitation of simple threshold judgment in identifying concealed anomalies. In the location and tracing phase, the branch power difference is calculated based on the associated transformer area topology, and the anomaly time characteristics are analyzed. This achieves closed-loop management from anomaly detection to location, shortening anomaly response time and reducing transformer area line loss rate and maintenance costs. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used 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 those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 The flowchart shows a method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception. Figure 2 A computer equipment diagram for a method to identify abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception; Figure 3 This is another flowchart of a method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0026] Example 1 Reference Figure 1 - Figure 2 This is the first embodiment of the present invention, which provides a method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception, including: S100: Collects electrical quantity parameters, power quality parameters, and environmental parameters from the transformer side, outgoing line side, and user side of the distribution network to form multi-dimensional sensing data. The data collection frequency is dynamically adjusted according to the load rate of the distribution area. When the load rate of the distribution area is lower than the first threshold, the collection frequency is reduced, and when the load rate of the distribution area is higher than the second threshold, the collection frequency is increased.

[0027] S200: Perform quality verification on multi-dimensional sensing data and extract time-series and statistical features of electricity consumption behavior, construct electricity consumption behavior feature vectors, cluster and classify users in the transformer area based on electricity consumption behavior feature vectors, calculate the normal range of each feature parameter for each type of user, establish user behavior baselines, and update user behavior baselines regularly.

[0028] S300: Construct an anomaly detection model. The anomaly detection model combines the anomaly probability output by the deep learning network with the feature deviation calculated based on the user behavior baseline to obtain a comprehensive anomaly index.

[0029] S400: When the comprehensive anomaly index exceeds the judgment threshold, it is judged as an abnormal behavior. The branch power difference is calculated based on the topology of the associated transformer area, the abnormal time characteristics are analyzed, and the anomaly location and source are completed.

[0030] It should be noted that, as the final link in the power supply service, the electricity consumption behavior of distribution network areas is highly dynamic and diverse. Different types of users have fundamentally different electricity consumption patterns. Traditional fixed threshold monitoring methods cannot meet these diverse needs. Furthermore, when the load of the distribution area changes drastically, it is easy to cause insufficient data collection or waste of resources. At the same time, the anomaly judgment rules that rely solely on human experience lack flexibility and are difficult to identify highly concealed abnormal electricity consumption behaviors, such as gradual electricity theft and changes in the nature of electricity consumption.

[0031] Therefore, through the steps from S100 to S400, firstly, by combining multi-level and multi-dimensional data collection with an adaptive collection strategy, a panoramic perception of the transformer area's operating status and optimized allocation of collection resources are achieved; secondly, by extracting electricity consumption behavior features and classifying users through clustering, personalized behavioral baselines for different types of users are established, addressing the problem of traditional methods neglecting user differences; thirdly, by fusing deep learning networks with rule baselines, an anomaly identification model with both intelligent recognition capabilities and interpretability is constructed; finally, by associating the transformer area's topology structure for anomaly localization and source tracing analysis, closed-loop management from anomaly discovery to problem localization is achieved, improving the accuracy of identifying abnormal electricity consumption behavior in the transformer area, reducing false alarm and false negative rates, and shortening anomaly response time.

[0032] Example 2 Reference Figure 1 - Figure 3 This is the second embodiment of the present invention.

[0033] In this embodiment, step S100 involves collecting electrical quantity parameters, power quality parameters, and environmental parameters from the transformer side, outgoing line side, and user side of the distribution network to form multi-dimensional sensing data. The data acquisition frequency is dynamically adjusted according to the distribution area load rate. When the distribution area load rate is lower than a first threshold, the acquisition frequency is reduced; when the distribution area load rate is higher than a second threshold, the acquisition frequency is increased. This includes the following steps A1-A2: A1: Collect electrical quantity parameters, power quality parameters, and environmental parameters from the transformer side, outgoing line side, and user side of the distribution network area, including collecting the three-phase voltage, three-phase current, active power, reactive power, and power factor of the total load of the transformer on the low-voltage side of the transformer. Load distribution data for each outgoing line is collected at each outgoing point in the transformer area; detailed electricity consumption data is collected after the electricity meters of users whose electricity consumption capacity exceeds the set value; The collected data includes basic electrical quantities, power parameters, total harmonic distortion of voltage, total harmonic distortion of current, three-phase imbalance, temperature, and humidity.

[0034] Specifically, a main monitoring terminal is installed on the low-voltage side of the distribution transformer in the transformer substation. The main monitoring terminal is equipped with a high-precision current transformer with a measurement accuracy of 0.2 class, ensuring accurate measurement of the total load current of the transformer. The parameters collected include three-phase voltages Ua, Ub, and Uc, three-phase currents Ia, Ib, and Ic, active power P, reactive power Q, and power factor cosφ. The main monitoring terminal uses a high-speed analog-to-digital converter chip with a sampling rate of 128 points per cycle, which can accurately capture instantaneous changes in electrical quantities. The collected three-phase voltage data is used to monitor the power supply quality of the transformer substation and determine whether there are voltage overruns or three-phase imbalance problems. The three-phase current data reflects the overall load level of the transformer substation and the load distribution of each phase. The active and reactive power data are used to calculate the total power consumption and power factor of the transformer substation and evaluate the operating efficiency of the transformer substation.

[0035] Branch monitoring terminals are installed at each outgoing line in the transformer substation area. The number of branch monitoring terminals is determined by the number of outgoing lines in the area, generally one branch monitoring terminal per outgoing line. The branch monitoring terminals collect load distribution data for each outgoing line, including parameters such as outgoing line current and power. Through the data from the branch monitoring terminals, maintenance personnel can understand the load sharing of each outgoing line within the substation area and identify whether there are overloads or uneven load distribution issues on any particular outgoing line. The branch monitoring terminals also collect voltage data for the outgoing lines to determine the voltage quality at the power supply end of that outgoing line.

[0036] For key users with a power capacity greater than 50 kilowatts, a user-side monitoring terminal is installed after their electricity meter. The user-side monitoring terminal collects detailed electricity consumption data, including the user's real-time power, cumulative power consumption, power factor, voltage, current, and other parameters. The deployment of the user-side monitoring terminal enables refined monitoring of the electricity consumption behavior of key users and timely detection of abnormal electricity consumption behavior. The user-side monitoring terminal adopts a non-intrusive installation method, collecting current signals through current transformers and voltage signals through voltage sampling lines. The installation process does not require power outages and does not affect the user's normal electricity use.

[0037] The collected data types cover three dimensions. The first dimension is basic electrical quantities, including three-phase voltages Ua, Ub, and Uc, and three-phase currents Ia, Ib, and Ic. The second dimension is power parameters, including active power P, reactive power Q, and power factor cosφ.

[0038] The third dimension consists of power quality parameters and environmental parameters, including total harmonic distortion of voltage (THDu), total harmonic distortion of current (THDi), three-phase imbalance (Uimb), and ambient temperature (T) and humidity (H).

[0039] The total harmonic distortion (THDu) of voltage and the total harmonic distortion (THDi) of current reflect the degree of influence of nonlinear loads in the transformer substation. When THDu or THDi exceeds the limit, it indicates that there are many harmonic pollution sources in the transformer substation, such as rectifiers and frequency converters.

[0040] Three-phase imbalance reflects the balance of three-phase load distribution. When the three-phase imbalance is too large, it will lead to excessive neutral current in the transformer, increase line loss, and even affect equipment safety.

[0041] A2: The data acquisition frequency is dynamically adjusted according to the load factor of the transformer area. When the load factor of the transformer area is lower than the first threshold, the acquisition frequency is reduced, and when the load factor of the transformer area is higher than the second threshold, the acquisition frequency is increased.

[0042] Specifically, the transformer load factor is defined as the ratio of the actual total active power of the transformer area to the rated capacity of the transformer. The calculation formula is: (1) In the formula, For the load factor of the transformer area, This represents the actual total active power of the transformer substation. This refers to the rated capacity of the transformer.

[0043] The load factor of a transformer substation reflects the utilization level of the transformer. When the load factor is low, the substation operates smoothly and the electrical parameters change slowly. When the load factor is high, the substation operates close to full load, the electrical parameters fluctuate more, and the risk of abnormality increases.

[0044] Based on real-time changes in load rate, an adaptive acquisition strategy is used to dynamically adjust the data acquisition frequency. The acquisition frequency is set in segments according to the load rate: (2) When the load factor is below 0.3, the transformer area is in a low-load operation state. At this time, the data acquisition frequency is set to collect data once every 15 minutes. Low-frequency acquisition can meet the basic monitoring needs of the transformer area's operating status, while reducing communication bandwidth usage and data storage pressure, thus lowering operating costs.

[0045] When the load factor is between 0.3 and 0.7, the transformer substation enters a medium load operation state, and the data collection frequency is increased to once every 5 minutes to ensure timely tracking of changes in the substation's operating status. When the load factor exceeds 0.7, the substation enters a high load operation state, at which point the transformer is close to full load, and the data collection frequency is increased to once every 1 minute to promptly detect abnormal situations such as voltage fluctuations and overloads that may occur under high load conditions.

[0046] In this embodiment, step S200 involves quality verification of the multi-dimensional sensing data and extraction of temporal and statistical features of electricity consumption behavior to construct an electricity consumption behavior feature vector. Based on the electricity consumption behavior feature vector, users in the transformer area are clustered and classified. For each user category, the normal range of each feature parameter is calculated, a user behavior baseline is established, and the user behavior baseline is updated periodically. This includes the following steps B1-B2: B1: Perform quality verification on multidimensional sensing data and extract the temporal and statistical features of electricity consumption behavior to construct an electricity consumption behavior feature vector, including calculating the daily electricity consumption, average load rate, standard deviation of load rate, average power factor, and three-phase imbalance as statistical features within the time window; The load curve within the time window is divided into multiple time periods, and the average power of each time period constitutes the load curve feature vector as a time series feature. Combining statistical features and time-series features forms a complete feature vector of electricity consumption behavior.

[0047] Specifically, after receiving the data uploaded by the monitoring terminal, the main station first executes a data quality verification process. This process includes three steps: integrity verification, timestamp continuity verification, and rationality verification. Integrity verification checks whether the data packet contains all necessary fields, such as whether parameters like voltage, current, and power are complete. Timestamp continuity verification verifies whether the time interval between two consecutive data acquisitions conforms to the preset acquisition frequency, determining if data loss exists. For detected data gaps, linear interpolation is used to complete the missing data. (3) In the formula, The interpolated data is for time t. and These are the measured data at the preceding and following times, respectively.

[0048] Linear interpolation assumes that the data changes linearly over a short period of time, and extrapolates the data value at the missing time by using known data from previous and subsequent time points.

[0049] The rationality check examines whether electrical quantities are within the physically possible range. Taking voltage data as an example, it determines whether the measured voltage meets the following requirements: (4) In the formula, The rated voltage is 220 volts. The measurement is for voltage. If the voltage exceeds the range, it is marked as an abnormal data point. According to national standards, the allowable deviation of the supply voltage is ±7% of the rated voltage; here, ±10% is used as a lenient boundary for judging the reasonableness of the data. If the measured voltage exceeds this range, it is marked as an abnormal data point and processed using the median filtering method. (5) In the formula, This is the filtered voltage value. This is the median function.

[0050] Median filtering uses the median of data from adjacent time points to replace outlier data, which can remove noise interference from sudden changes while preserving the overall trend of data change.

[0051] The data that passed quality verification were normalized to eliminate the influence of different units on subsequent analysis. The maximum-minimum normalization method was used. (6) In the formula, Here, x represents the normalized data, and x represents the original data. and These are the historical minimum and maximum values ​​for this parameter, respectively.

[0052] The normalized data is stored in a time-series database and indexed according to timestamps for easy subsequent querying and analysis.

[0053] Feature extraction employs a sliding time window method, with a window length of 24 hours and a sliding step of 1 hour. For each time window, the following statistical features are calculated: Daily electricity consumption equals the integral of power at each moment within the window: (7) In the formula, P(t) is the power-time function. Let be the power of the i-th sampling point. is the sampling interval, and N is the total number of sampling points within the window.

[0054] Daily electricity consumption reflects a user's total electricity consumption level in a day and is a basic indicator for measuring the scale of a user's electricity consumption.

[0055] The average load factor reflects the average electricity consumption level within the window: (8) In the formula, The average load factor, The load rate of the i-th sampling point is calculated from Formula 1 in the first step.

[0056] The average load factor reflects the average intensity of electricity consumption by users. Users with a high load factor consume large amounts of electricity that are also more continuous.

[0057] The standard deviation of load factor reflects the degree of fluctuation in electricity consumption: (9) In the formula, This represents the standard deviation of the load factor; the greater the fluctuation, the larger the standard deviation.

[0058] Residential users typically have a larger standard deviation in load factor because their electricity consumption is influenced by daily routines, resulting in significant peak-to-valley differences. Industrial users, on the other hand, have a smaller standard deviation in load factor because industrial production is relatively stable, leading to smaller fluctuations in electricity load.

[0059] The average power factor reflects the characteristics of electrical equipment: (10) In the formula, The average power factor, Let be the power factor of the i-th sampling point.

[0060] The power factor reflects the ratio of active power to apparent power. A low power factor indicates that the user uses more inductive load equipment, such as motors and transformers.

[0061] Three-phase unbalance reflects the distribution of three-phase loads: (11) In the formula, For three-phase imbalance, This represents the effective value of the three-phase current.

[0062] Three-phase unbalance reflects the degree of balance in the distribution of three-phase loads. Ideally, the three-phase loads should be evenly distributed so that the three-phase currents are basically equal.

[0063] In addition to statistical features, time-series features were also extracted. The 24-hour load curve was divided into 96 time periods, each lasting 15 minutes. The average power of each period constituted the load curve feature vector. (12) In the formula, The characteristic vector of the load curve, Let be the average power in the k-th time period.

[0064] The load curve eigenvector retains the temporal information of electricity consumption behavior, which can reflect the electricity consumption patterns of users at different times of the day.

[0065] Residential users' load curves exhibit a double-peak characteristic, with higher electricity consumption between 6-8 am and 6-10 pm; commercial users' load curves remain at a consistently high level during business hours; and industrial users' load curves typically remain stable during working hours.

[0066] The extracted multidimensional feature vectors are combined as follows: (13) In the formula, The complete feature vector has 101 dimensions (5 statistical features plus 96 temporal features).

[0067] This feature vector comprehensively characterizes the user's electricity consumption behavior, including both statistical quantities reflecting the scale and intensity of electricity consumption and time-series information reflecting the time pattern of electricity consumption.

[0068] B2: Establish user behavior baselines, including using clustering algorithms to cluster the electricity consumption behavior feature vectors of all users in the transformer area, and classifying users into residential users, commercial users and industrial and commercial users; For each type of user, calculate the mean and standard deviation of daily electricity consumption, average load rate, power factor, and three-phase imbalance. The normal range of each characteristic parameter is determined based on the mean and standard deviation; the user behavior baseline is recalculated and updated monthly using the latest data.

[0069] Specifically, 90 consecutive days of electricity consumption behavior feature vector data from all users within the transformer area were collected to form a training dataset. The K-means clustering algorithm was used to classify users into different types. The core idea of ​​the K-means algorithm is to divide the data into K clusters, minimizing the distance between data points within a cluster and maximizing the distance between clusters. First, the number of clusters K was determined, and the elbow method was used to select the optimal K value. The sum of squares (SSE) within each cluster under different K values ​​was calculated. (14) In the formula, For the k-th cluster, The center of the k-th cluster, Let be the feature vector of the i-th user.

[0070] Plot the K-SSE curve and select the K value where the curve shows a clear inflection point as the optimal number of clusters.

[0071] The K-means algorithm iteratively updates the cluster centers until the cluster centers no longer change or the maximum number of iterations is reached. The cluster center update formula is: (15) In the formula, The cluster center in the (t+1)th iteration, Let be the number of samples in cluster k at the t-th iteration. The iteration process continues until the change in cluster centers is less than a threshold. Or it can reach the maximum number of iterations of 200.

[0072] The iterative process consists of two steps: an assignment step and an update step. In the assignment step, each user is assigned to the cluster containing the nearest cluster center; in the update step, the cluster center is recalculated based on the mean of all samples in the current cluster.

[0073] After clustering, the 186 users in the transformer area were divided into three categories. There were 152 residential users. The electricity consumption characteristics of this category are: low daily electricity consumption, generally between 3,000 and 20 kWh; low and fluctuating load factor, ranging from 0.15 to 0.42; high power factor, ranging from 0.88 to 0.96; and a load curve showing a double-peak characteristic in the morning and evening.

[0074] There are 24 commercial users. The electricity consumption characteristics of this type of user are: moderate daily electricity consumption, generally between 15,000 and 55,000 kWh, high load rate during business hours, ranging from 0.35 to 0.68, moderate power factor, ranging from 0.82 to 0.92, and the load curve remains at a high level during business hours.

[0075] There are 10 industrial and commercial users. The electricity consumption characteristics of this type of user are large daily electricity consumption, generally between 48,000 and 160 kWh, high and stable load rate, with a load rate range of 0.58 to 0.85, low power factor, with a range of 0.75 to 0.88, and the load curve remains stable during working hours.

[0076] For each user category, calculate its normal behavior baseline. The baseline includes the expected range of each characteristic parameter. Taking daily electricity consumption as an example, calculate the average daily electricity consumption for this user category: (16) The average daily electricity consumption represents the typical electricity consumption level of this type of user.

[0077] Calculate the standard deviation of daily electricity consumption: (17) In the formula, This represents the average daily electricity consumption. Let M be the standard deviation of daily electricity consumption, and M be the sample size of this type of user. The daily electricity consumption of the j-th sample is calculated using Formula 7.

[0078] The standard deviation reflects the dispersion of daily electricity consumption for this type of user. The larger the standard deviation, the greater the internal variation among this type of user.

[0079] The normal range for daily electricity consumption for this type of user is defined as follows: (18) In the formula, and The lower and upper limits of the normal range for daily electricity consumption are defined using the principle of three times the standard deviation, and this range covers 99.7% of normal samples.

[0080] Using the 3-standard-deviation principle, based on the normal distribution theory, this range covers 99.7% of normal samples.

[0081] Similarly, normal ranges are established for all characteristic parameters such as average load factor, standard deviation of load factor, power factor, and three-phase imbalance, forming a complete behavioral baseline for this type of user. The behavioral baseline is not fixed and is recalculated and updated monthly using the latest 30 days of data, so that the baseline can adapt to seasonal changes and changes in electricity consumption habits.

[0082] For example, during the high temperatures of summer, the air conditioning load of residential users increases, leading to an increase in the average daily electricity consumption. By updating the baseline monthly, the normal range can be adjusted in a timely manner to avoid misjudging normal seasonal electricity consumption growth as abnormal.

[0083] In this embodiment, step S300 involves constructing an anomaly detection model. This model combines the anomaly probability output by the deep learning network with the feature deviation calculated based on the user behavior baseline to obtain a comprehensive anomaly index, including the following C1 step: C1: Construct an anomaly detection model, including constructing a multi-layer deep neural network, with the network input being the electricity consumption behavior feature vector and the output being the anomaly probability; A deep neural network was trained using known abnormal samples as positive samples and normal electricity consumption data as negative samples. The user's current feature parameters are compared with the normal range of the user behavior baseline of the category, and the distance between the current value and the boundary of the normal range is calculated as the feature deviation. The anomaly probability and feature deviation are weighted and fused to obtain a comprehensive anomaly index, in which the weight of the anomaly probability is higher than the weight of the feature deviation.

[0084] Specifically, a multi-layer deep neural network model is constructed to identify abnormal electricity consumption behavior: the network input is the 101-dimensional electricity consumption behavior feature vector extracted in step B1, and the output is the abnormal probability. The network adopts a fully connected feedforward structure, including an input layer, three hidden layers, and an output layer. The input layer receives the 101-dimensional feature vector. The first hidden layer contains 64 neurons, the second hidden layer contains 32 neurons, the third hidden layer contains 16 neurons, and the output layer contains 1 neuron. The neurons in the hidden layers are fully connected, and each connection corresponds to a weight parameter. The total number of network parameters is approximately 101×64 + 64×32 + 32×16 + 16×1 = 9072 parameters.

[0085] The hidden layer activation function uses the ReLU function, defined as ReLU(x) = max(0,x). The advantages of the ReLU function are its simplicity of computation, which accelerates network training and alleviates the vanishing gradient problem. The forward propagation calculation for the first hidden layer is as follows: (19) In the formula, This is the output vector of the first hidden layer. This is the weight matrix. For bias vectors, This is the activation function. Subsequent hidden layer calculations are similar.

[0086] The weight matrix and bias vector are initialized using the Xavier initialization method before training to ensure that the output variance of each layer remains consistent in the early stage of training. The calculation process of the second and third hidden layers is similar. The output of the second hidden layer is h(2)=ReLU(W(2)h(1)+b(2)) and the output of the third hidden layer is h(3)=ReLU(W(3)h(2)+b(3)).

[0087] The output layer uses the Sigmoid activation function, so that the output value between 0 and 1 represents the probability of an anomaly. The output layer is calculated as follows: (20) In the formula, This represents the probability of an anomaly. Let b be the output layer weight vector, and b be the bias scalar. Output for the third hidden layer. .

[0088] The anomaly probability ranges from 0 to 1. The closer the value is to 1, the more likely the user is to have abnormal electricity usage behavior.

[0089] The binary cross-entropy loss function is used to measure the difference between the model's predicted values ​​and the true labels: (twenty one) In the formula, Let be the true label of the i-th sample (1 for abnormal, 0 for normal). The anomaly probability predicted by the model. This represents the number of training samples.

[0090] In practical applications, after extracting features from real-time collected electricity consumption data using the method in step B1, the data is input into a trained deep neural network model. The model then performs forward propagation to calculate the output anomaly probability (panomaly). Simultaneously, the user's current feature parameters are compared with the user behavior baseline established in step B2 to calculate the feature deviation. Taking daily electricity consumption as an example, the daily electricity consumption deviation is calculated as follows: (twenty two) In the formula, This refers to the deviation of daily electricity consumption. This represents the current daily electricity consumption. and This is the normal range determined by formula 18 in step four.

[0091] When a user's daily electricity consumption is within the normal range, the deviation is 0; when a user's daily electricity consumption exceeds the normal range, the deviation is equal to the distance between the actual value and the boundary of the normal range. The greater the deviation, the greater the difference between the user's current electricity consumption behavior and the historical normal pattern.

[0092] The combined anomaly probability output by the deep learning model and the rule-based feature deviation are used to calculate the comprehensive anomaly index: (twenty three) In the formula, To establish a comprehensive anomaly index, weighting coefficients of 0.7 and 0.3 were determined based on historical case statistics. Deep learning models, with their strong ability to recognize complex patterns, are given higher weights, providing interpretable supplementary information for baseline deviations. The standard deviation of daily electricity consumption calculated using formula 17 in step four is used to normalize the deviation.

[0093] Deep learning models have a strong ability to identify complex nonlinear patterns, therefore they are given a higher weight of 0.7. Baseline deviation provides an interpretability supplement, which can intuitively reflect the degree of deviation between user electricity consumption behavior and historical normal patterns, and is given a weight of 0.3.

[0094] In this embodiment, when the comprehensive anomaly index exceeds the judgment threshold in step S400, it is judged as an abnormal behavior. The branch power difference is calculated based on the topology of the associated transformer area, the abnormal time characteristics are analyzed, and the anomaly location and source tracing are completed. This includes the following steps D1-D2: D1: Analyze the characteristics of abnormal time, complete the anomaly location and source tracing, including, when an abnormal user is identified, obtain the total power of the branch where the abnormal user is located and the power of each user; calculate the difference between the total power of the branch and the sum of the power of each user to obtain the branch power difference. The branch power difference is compared with the theoretical line loss value. When the branch power difference is significantly greater than the theoretical line loss value, it is determined that there is unmetered electricity consumption in the branch. The distribution of anomalies across time periods is statistically analyzed, and the duration between the first and last time an anomaly is detected is calculated as the duration of the anomaly.

[0095] Specifically, retrieve the transformer area topology data. The transformer area topology can be represented as a tree structure, with the transformer as the root node, each level of distribution line as a branch node, and users as leaf nodes. Obtain the monitoring data of the branch where the abnormal user is located, including the total branch power Pbranch measured by the branch monitoring terminal and the power consumption of each user on that branch. Calculate the difference between the total branch power and the sum of the power consumption of each user. (twenty four) In the formula, This is the branch power difference. The total power measured by the branch monitoring terminal. Let be the power consumption of the i-th user on this branch.

[0096] If the branch power difference ΔP is greater than the theoretical line loss value, it indicates that there is unmetered electricity use in that branch, that is, there is unmetered electricity use, which may be electricity theft or unauthorized power connection.

[0097] Theoretical line loss is calculated based on line parameters and load current: (25) In the formula, Here, I represents the line loss power, I represents the line current, and R represents the line resistance.

[0098] For example, if a branch line is 500 meters long and uses LGJ-50 aluminum stranded wire with a resistivity of 0.64 Ω / km, then the line resistance R = 0.64 × 0.5 = 0.32 Ω. When the branch current is 50 amperes, the theoretical line loss power Ploss = 50² × 0.32 = 800 watts.

[0099] Define the location reliability of anomalies: (26) In the formula, To determine location reliability, a larger difference indicates more accurate localization. At that time, it was determined that there was clear unmetered electricity consumption on that branch.

[0100] For example, if the theoretical line loss of a branch is 800 watts and the actual branch power difference ΔP is 2000 watts, then the location confidence level Cloc = 2000 / 800 × 100% = 250%, which far exceeds the 200% threshold, confirming that there is unmetered electricity consumption in this branch.

[0101] It is also necessary to analyze the temporal characteristics of abnormal behavior and statistically analyze the time period distribution of abnormal occurrences: by statistically analyzing the hourly abnormality index within 24 hours, an abnormal time period distribution map is drawn; if the abnormality mainly occurs between 11 PM and 5 AM the next day or on weekends, it may be electricity theft, because electricity thieves tend to choose times when electricity use supervision is lax to commit theft. If the abnormality persists and gradually worsens, it may be equipment failure or illegal electricity use, requiring further on-site verification. Calculate the duration of the abnormality: (27) In the formula, The duration of the anomaly. The moment when the anomaly was first detected. This is the last time an anomaly was detected.

[0102] The duration of an anomaly reflects the persistence and severity of the abnormal behavior. Short-term, occasional anomalies may be measurement errors or normal fluctuations in user electricity consumption habits, while long-term, persistent anomalies are more likely to be genuine abnormal electricity consumption behavior.

[0103] D2: This also includes classifying abnormal behavior based on the comprehensive anomaly index, location reliability, and duration of anomaly. For different levels of abnormal behavior, generate an anomaly tracing report that includes abnormal user information, anomaly feature description, related topology information, suspected anomaly type, and handling suggestions.

[0104] Specifically, according to the comprehensive anomaly index Location reliability and duration of abnormality Abnormal behavior is classified into categories. The classification rules are as follows: (28) An emergency-level anomaly that meets both the high anomaly index and high location reliability indicates that the evidence of the anomaly is sufficient and the situation is serious. An alarm SMS and telephone notification should be sent to the operation and maintenance personnel immediately, requiring on-site verification and handling within 24 hours.

[0105] Important anomalies that meet either a high anomaly index or a prolonged duration will be highlighted in red on the interface and require handling within 3 days. General anomalies with a moderate anomaly index and a short duration will be included in the routine inspection plan and handled during the next scheduled inspection.

[0106] A complete anomaly tracing report is automatically generated, consisting of six parts. The first part is the basic information of the abnormal user, including account number, user name, address, electricity usage type, and contracted capacity.

[0107] The second part describes the abnormal features, including the comprehensive abnormality index, the abnormal probability output by the deep learning model, the deviation of each feature parameter, the distribution map of abnormal time periods, and the duration of the abnormality.

[0108] The third part is the associated topology information, including the name of the branch where the user is located, the branch monitoring terminal number, the branch power difference, the theoretical line loss value, and the location reliability.

[0109] The fourth section lists suspected anomaly types, which are automatically inferred based on their characteristics. If the anomaly index is high and mainly occurs at night, it is suspected electricity theft; if the power factor is abnormally low and the three-phase imbalance is increased, it is suspected illegal use of electricity or unauthorized connection of high-power equipment; if the load suddenly drops or the daily electricity consumption decreases abnormally, it is suspected metering failure.

[0110] The fifth section lists the user's historical anomaly records, including the time of occurrence, type, and outcome of each anomaly, helping operations personnel determine if the user is a repeat offender. The sixth section provides handling recommendations based on the anomaly type and severity.

[0111] For suspected cases of electricity theft, it is recommended to immediately inspect the metering device and the wiring after the meter on-site to check for any illegal connections, meter reversals, or meter replacements. For suspected cases of illegal electricity use, it is recommended to check whether the user's actual electricity usage is consistent with the contract and whether there has been any unauthorized change of the electricity usage nature. For suspected cases of metering malfunction, it is recommended to replace the electricity meter or calibrate the accuracy of the metering device.

[0112] The anomaly tracing report is pushed to the operation and maintenance management. Operation and maintenance personnel can view the report details through computer or mobile terminal, arrange on-site verification work according to the report content, and after the on-site verification results are fed back, the verification results are linked with the anomaly tracing report for archiving, forming a closed-loop management of anomaly cases.

[0113] In summary, by deploying high-precision main monitoring terminals on the low-voltage side of the distribution transformer in the transformer area, branch monitoring terminals at each outgoing line, and user-side monitoring terminals after the meters of key users, a three-dimensional sensing network covering the entire transformer area is constructed. The types of data collected have expanded from basic electrical quantities to multi-dimensional parameters such as harmonic distortion rate, three-phase imbalance, temperature, and humidity. Regarding data quality assurance, a triple mechanism of integrity verification, timestamp continuity verification, and rationality verification is used to complete missing data using linear interpolation, ensuring data reliability and consistency. In the feature extraction stage, a sliding time window method is used to calculate daily electricity consumption, average load factor, standard deviation of load factor, and average power factor. The system analyzes statistical characteristics such as three-phase imbalance and divides the load curve into multiple time periods to extract time-series features, forming a complete electricity consumption behavior feature vector that reflects both the scale of electricity consumption and captures temporal patterns. Using the K-means clustering algorithm, users are automatically classified into three categories: residential, commercial, and industrial / commercial. A behavioral baseline containing the normal range of each feature parameter is established for each category and dynamically updated monthly to ensure the baseline always aligns with the current electricity consumption pattern. The constructed deep neural network employs a three-layer hidden layer structure, training the model to learn complex electricity consumption patterns using known abnormal samples. A comprehensive anomaly index is obtained by weighted fusion of feature deviations, improving the accuracy and reliability of anomaly identification.

[0114] Example 3 This example is based on a 10 kV distribution transformer substation in actual operation in a certain region. The substation has a power supply radius of 1.2 km, a transformer capacity of 500 kVA, and serves 186 users, including 152 residential users, 24 small shop users, and 10 general industrial and commercial users. A three-dimensional sensing network was constructed by deploying one main monitoring terminal on the transformer side, four outgoing branch monitoring terminals, and ten key user-side monitoring terminals in the substation. Data collection was conducted continuously for six months, accumulating approximately 8.5 million records. A simulation environment was built on the MATLAB platform to reproduce the actual operating scenario of the substation, and eight typical anomaly cases (three cases of electricity theft, two cases of illegal electricity use, two cases of metering failure, and one case of equipment aging) were incorporated to verify the algorithm's performance. The effectiveness was evaluated by comparing traditional anomaly detection methods based on fixed thresholds, methods using only statistical analysis, and the method proposed in this patent, based on four dimensions: accuracy, false negative rate, false alarm rate, and response time.

[0115] 2. Validity Verification Form Table 1: Comparison of Anomaly Detection Performance

[0116] In eight preset anomaly cases, this method correctly identified all cases, with only one false alarm (a resident experienced a sudden increase in electricity load, which was later found to be normal behavior due to the addition of air conditioning equipment). The fixed threshold method, due to its unreasonable threshold setting, has poor ability to identify concealed electricity theft and gradual faults, resulting in a false alarm rate as high as 50%. While statistical analysis is superior to the fixed threshold method, it lacks the ability to learn complex nonlinear patterns, achieving an accuracy rate of only 75%. This method combines the pattern recognition capabilities of deep learning with the interpretability of the rule baseline, achieving an accuracy rate of 98.8%. The response time is reduced from weeks in traditional methods to hours, enabling rapid detection of anomalies.

[0117] Table 2: Results of User Segmentation and Baseline Establishment

[0118] The K-means clustering algorithm successfully divided 186 users into three categories (there were no large industrial users or mixed users in this area), achieving a classification accuracy of over 96.7%. The electricity consumption behavior characteristics of each category differed significantly: residential users had lower daily electricity consumption, lower and more volatile load rates, and higher power factors; commercial users had moderate electricity consumption and higher load rates during business hours; and industrial and commercial users had higher electricity consumption and higher, more stable load rates. The behavioral baseline established for each user category accurately reflects their normal electricity consumption patterns, providing a reliable criterion for anomaly identification.

[0119] Table 3: Anomaly Localization and Source Tracing Results

[0120] The system achieved 100% accuracy in locating the eight anomaly cases, with six cases pinpointing specific users and two cases pinpointing branch lines (confirmed through subsequent door-to-door investigation). The anomaly index was positively correlated with the severity of the anomaly, with electricity theft generally exhibiting a higher anomaly index than other types. The location reliability reflected the sufficiency of the anomaly evidence; higher confidence levels indicated more accurate location. The system-generated source tracing reports were comprehensive, including descriptions of anomaly characteristics, topological correlation analysis, and handling recommendations, effectively guiding on-site verification. All eight cases were confirmed after on-site verification, with no misjudgments.

[0121] Table 4: Economic Benefit Analysis

[0122] After applying this method, the detection rate of abnormal electricity use in the transformer substation significantly improved, increasing from an average of 2.3 cases per year to 8.7 cases, effectively curbing electricity theft and illegal electricity use. The recovered electricity loss increased from an average of 4,800 kWh per year to 18,500 kWh, which translates to an average annual increase in revenue of approximately 8,200 yuan (based on a price of 0.6 yuan per kWh). Due to the implementation of intelligent online monitoring, the number of on-site inspections decreased by 62.5%, significantly reducing labor costs. The line loss rate in the transformer substation decreased from 6.8% to 4.2%, a reduction of 2.6 percentage points, reaching an excellent level. The overall operation and maintenance cost saved an average of 14,000 yuan per year, a decrease of 43.8%. The economic benefits are significant.

[0123] Example 4 The above is a schematic scheme for a method of identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception. It should be noted that the technical solution of this system for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception belongs to the same concept as the technical solution of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception described above. Details not described in detail in the technical solution of the system for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception in this embodiment can be found in the description of the technical solution of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception described above.

[0124] This embodiment also provides a multi-dimensional perception-based power distribution network area abnormal behavior identification system, including: The data acquisition module is used to collect multi-dimensional sensing data from the transformer side, outgoing line side, and user side of the distribution network, and dynamically adjust the data acquisition frequency according to the load rate of the distribution area. The data processing module is used to perform quality verification and normalization on multidimensional sensing data, and extract the temporal and statistical features of electricity consumption behavior. The user classification module is used to cluster and classify users in the transformer area, and to establish and regularly update user behavior baselines. The anomaly identification module is used to input electricity consumption data into a deep neural network to calculate the anomaly probability, calculate the feature deviation, and weightedly fuse them to obtain a comprehensive anomaly index. The location and source tracing module is used to calculate the branch power difference based on the topology of the associated transformer area, analyze the abnormal time characteristics, and complete the anomaly location and source tracing.

[0125] This embodiment also provides an electronic device suitable for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception as proposed in the above embodiment.

[0126] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception, as proposed in the above embodiments.

[0127] The storage medium proposed in this embodiment and the method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0128] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception, characterized in that: include, The system collects electrical quantity parameters, power quality parameters, and environmental parameters from the transformer side, outgoing line side, and user side of the distribution network to form multi-dimensional sensing data. The data collection frequency is dynamically adjusted according to the load rate of the distribution area. When the load rate of the distribution area is lower than the first threshold, the collection frequency is reduced, and when the load rate of the distribution area is higher than the second threshold, the collection frequency is increased. The quality of multidimensional sensing data is verified and the temporal and statistical features of electricity consumption behavior are extracted. Electricity consumption behavior feature vectors are constructed, and users in the transformer area are clustered and classified based on the electricity consumption behavior feature vectors. The normal range of each feature parameter is calculated for each type of user, a user behavior baseline is established, and the user behavior baseline is updated regularly. An anomaly detection model is constructed, which combines the anomaly probability output by the deep learning network with the feature deviation calculated based on the user behavior baseline to obtain a comprehensive anomaly index. When the comprehensive anomaly index exceeds the judgment threshold, it is judged as an abnormal behavior. The branch power difference is calculated based on the topology of the associated transformer area, the abnormal time characteristics are analyzed, and the anomaly location and source are completed.

2. The method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in claim 1, characterized in that: The collection of electrical parameters, power quality parameters, and environmental parameters from the transformer side, outgoing line side, and user side of the distribution network includes collecting the three-phase voltage, three-phase current, active power, reactive power, and power factor of the total load of the transformer on the low-voltage side of the transformer. Load distribution data for each outgoing line is collected at each outgoing point in the transformer area; detailed electricity consumption data is collected after the electricity meters of users whose electricity consumption capacity exceeds the set value; The collected data includes basic electrical quantities, power parameters, total harmonic distortion of voltage, total harmonic distortion of current, three-phase imbalance, temperature, and humidity.

3. The method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in claim 2, characterized in that: The process involves quality verification of multidimensional sensing data and extraction of temporal and statistical features of electricity consumption behavior to construct an electricity consumption behavior feature vector. This vector includes the calculation of daily electricity consumption, average load rate, standard deviation of load rate, average power factor, and three-phase imbalance within a time window as statistical features. The load curve within the time window is divided into multiple time periods, and the average power of each time period constitutes the load curve feature vector as a time series feature. Combining statistical features and time-series features forms a complete feature vector of electricity consumption behavior.

4. The method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in claim 3, characterized in that: The establishment of the user behavior baseline includes: using a clustering algorithm to cluster the electricity consumption behavior feature vectors of all users in the transformer area, classifying users into residential users, commercial users, and industrial and commercial users; for each type of user, calculating the mean and standard deviation of daily electricity consumption, average load rate, power factor, and three-phase imbalance; determining the normal range of each feature parameter based on the mean and standard deviation; and recalculating and updating the user behavior baseline monthly using the latest data.

5. The method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in claim 4, characterized in that: The construction of the anomaly detection model includes, Construct a multi-layer deep neural network with electricity consumption behavior feature vectors as input and anomaly probability as output. A deep neural network was trained using known abnormal samples as positive samples and normal electricity consumption data as negative samples. The user's current feature parameters are compared with the normal range of the user behavior baseline of the category, and the distance between the current value and the boundary of the normal range is calculated as the feature deviation. The anomaly probability and the feature deviation are weighted and fused to obtain a comprehensive anomaly index, wherein the weight of the anomaly probability is higher than the weight of the feature deviation.

6. The method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in claim 5, characterized in that: The analysis of abnormal time characteristics to complete abnormal location and source tracing includes, when an abnormal user is identified, obtaining the total power of the branch where the abnormal user is located and the power of each user; calculating the difference between the total power of the branch and the sum of the power of each user to obtain the branch power difference. The branch power difference is compared with the theoretical line loss value. When the branch power difference is significantly greater than the theoretical line loss value, it is determined that there is unmetered electricity consumption in the branch. The distribution of anomalies across time periods is statistically analyzed, and the duration between the first and last time an anomaly is detected is calculated as the duration of the anomaly.

7. The method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in claim 6, characterized in that: It also includes classifying abnormal behavior based on the comprehensive anomaly index, location reliability, and duration of anomaly; For different levels of abnormal behavior, generate an anomaly tracing report that includes abnormal user information, anomaly feature description, related topology information, suspected anomaly type, and handling suggestions.

8. A distribution network transformer area abnormal power consumption behavior identification system based on multi-dimensional perception, based on the distribution network transformer area abnormal power consumption behavior identification method based on multi-dimensional perception as described in any one of claims 1 to 7, characterized in that: It also includes a data acquisition module, which is used to collect multi-dimensional sensing data from the transformer side, outgoing line side and user side of the distribution network, and dynamically adjust the data acquisition frequency according to the load rate of the distribution area; The data processing module is used to perform quality verification and normalization on multidimensional sensing data, and extract the temporal and statistical features of electricity consumption behavior. The user classification module is used to cluster and classify users in the transformer area, and to establish and regularly update user behavior baselines. The anomaly identification module is used to input electricity consumption data into a deep neural network to calculate the anomaly probability, calculate the feature deviation, and weightedly fuse them to obtain a comprehensive anomaly index. The location and source tracing module is used to calculate the branch power difference based on the topology of the associated transformer area, analyze the abnormal time characteristics, and complete the anomaly location and source tracing.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for identifying abnormal electricity consumption behavior in a distribution network area based on multi-dimensional perception as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for identifying abnormal electricity consumption behavior in distribution network areas based on multi-dimensional perception as described in any one of claims 1 to 7.

Images