Electricity larceny early warning method and system based on multi-modal characteristics of metering box, and medium
By combining expert rules based on the multimodal features of the metering box with machine learning models, the problems of inefficient manual investigation and false alarms in existing anti-theft technologies are solved, enabling real-time, rapid identification and accurate alarm of electricity theft.
Patent Information
- Application Number
- CN202511760974.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-27
- Publication Date
- 2026-03-31
AI Technical Summary
Existing anti-theft technologies rely on inefficient manual inspections, suffer from significant time delays and frequent false alarms, and cannot achieve efficient, real-time, and accurate anti-theft measures.
By combining the door status, vibration, magnetic field, and electrical data of the metering box, multimodal feature analysis is performed through expert rules and machine learning models to construct an electricity theft early warning method. A kernel logistic regression model based on multi-kernel learning is used for secondary judgment to eliminate false alarms caused by single factors, thereby achieving fast and accurate electricity theft identification.
It enables real-time and rapid initial screening and response to electricity theft, reduces false alarm rates, improves the efficiency and accuracy of electricity theft prevention, reduces resource waste, and meets timeliness requirements.
Smart Images
Figure CN121762919A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electricity theft detection technology, specifically to an electricity theft early warning method, system, and medium based on the multimodal characteristics of a metering box. Background Technology
[0002] In the operation and management of power systems, electricity theft is one of the core issues causing non-technical line losses in the power grid and resulting in huge economic losses. To address this challenge, various anti-theft methods and devices have been developed in existing technologies, but they generally suffer from three major drawbacks: reliance on inefficient manual labor, slow response, and high false alarm rates, which seriously restrict the efficiency and accuracy of anti-theft work.
[0003] Specifically, existing anti-electricity theft technologies have the following main shortcomings:
[0004] First, there is an over-reliance on and inefficiency of manual investigation. Currently, the mainstream method mainly adopts a "post-event analysis, on-site inspection" model. The typical process is as follows: First, an abnormally high line loss rate in a certain transformer area is detected through the electricity information collection system; then, data analysts, based on their experience, sift through a massive number of users to create a list of suspected users who may be involved in electricity theft; finally, inspectors are dispatched with detection equipment to conduct on-site inspections. This model heavily relies on the personal experience of the inspectors, and the entire investigation process is time-consuming and labor-intensive. Faced with a large user base, it's akin to "finding a needle in a haystack," resulting in extremely low efficiency. This manpower-driven investigation method has become a major bottleneck restricting the improvement of anti-electricity theft efficiency.
[0005] Secondly, there is a significant time delay between the occurrence of an anomaly and its detection. Due to the aforementioned manually-led model, a lengthy closed loop is formed from the moment a user commits electricity theft, to the system detecting the abnormal line loss rate, data analysts identifying and generating a work order, and finally, on-site personnel confirming and handling the matter. This process involves delays of several days or even weeks, making it impossible to provide early warnings and timely prevention of electricity theft. This significant time delay not only gives criminals ample time to commit crimes, causing continuous and irrecoverable electricity losses, but also makes evidence collection and investigation extremely difficult.
[0006] Third, the reliance on a single anomaly criterion leads to frequent false alarms. Existing automated monitoring methods largely depend on a single electrical quantity criterion (such as abnormal fluctuations in current or voltage, or imbalances in current between live and neutral wires). However, these abnormal signals are easily confused with normal user-side power load changes, equipment start-ups and shutdowns, or even non-theft factors such as aging lines or poor connections. The system frequently issues false alarms, generating a large number of false alarm work orders. This not only wastes valuable investigative resources, but also leads to staff fatigue with the alarm system, potentially causing them to overlook genuine electricity theft signals.
[0007] In summary, the core shortcomings of existing anti-theft technologies lie in their passivity, lag, and low intelligence. This stems from their failure to fully utilize multi-dimensional data fusion analysis and intelligent algorithms, hindering the fundamental shift from manual surveys to intelligent, precise judgment, and from post-incident handling to pre-emptive warnings and in-process intervention. Therefore, there is an urgent need for a new technological solution to overcome these shortcomings and achieve efficient, real-time, and accurate anti-theft measures. Summary of the Invention
[0008] Addressing the core shortcomings of existing anti-electricity theft technologies, such as reliance on inefficient manual inspections, significant time delays from anomaly detection to discovery, and frequent false alarms, this invention aims to provide an electricity theft early warning method, system, and medium based on the multimodal characteristics of metering boxes. This solution combines physical quantity monitoring signals (access control, vibration, and magnetic fields) with electrical quantity monitoring signals to form a multimodal characteristic judgment system. It utilizes prior knowledge to set expert rules for initial judgment, eliminating false alarms caused by single factors. For complex and concealed electricity theft methods, machine learning models are used to deeply mine and correlate the pre-processed multidimensional data, identifying abnormal patterns that are difficult to summarize using expert experience. This further filters out false alarms from a probabilistic perspective, ensuring high confidence in reported alarm information.
[0009] The above-mentioned technical objective of the present invention is achieved through the following technical solution:
[0010] This solution provides a method for early warning of electricity theft based on the multimodal characteristics of meter boxes. The method includes:
[0011] Real-time monitoring of basic data of the metering box; the basic data includes metering box door status data, metering box vibration data, metering box magnetic field data and electrical data;
[0012] The basic data undergoes a first preprocessing step, and based on the preprocessed basic data, expert rule judgment is performed to determine the type of electricity theft risk.
[0013] For cases with a high risk of electricity theft or above, the basic data undergoes a second preprocessing, and the preprocessed basic data is then input into the pre-built classification model for a second determination.
[0014] An electricity theft alarm will be triggered based on the results of the second assessment.
[0015] A further optimized solution is that the first preprocessing includes:
[0016] Extract door switch features based on meter box door status: use binary variables {0, 1} to represent door status, where 0 represents closed and 1 represents open; and combine the door opening and closing time to determine whether the meter box door opening time is normal or not.
[0017] Vibration characteristics are extracted from the vibration data in the metering box: a preset vibration threshold is used to compare the vibration data with the vibration threshold to determine whether the vibration data is abnormal.
[0018] The magnetic field characteristics are extracted from the magnetic field data in the measuring box: a preset magnetic field threshold is set, and the magnetic field data is compared with the magnetic field threshold, as well as the magnetic field change is compared with the magnetic field threshold to determine whether the vibration data is abnormal.
[0019] Based on electrical data, electrical features are extracted: preset electrical thresholds are used to compare the Euclidean norm of the electrical data with the corresponding preset electrical thresholds to determine whether the electrical data is abnormal.
[0020] A further optimized solution involves using expert rules to determine the type of electricity theft risk based on the aforementioned basic data; including:
[0021] An abnormal meter box opening time, abnormal vibration data, abnormal magnetic field data, and abnormal electrical data are designated as events A, B, C, and D, respectively. A multimodal event list is constructed: an event list `event_list` is initialized to store events occurring within time T; when a new event occurs: the new event is added to `event_list`, and the current time T is recorded. d ; Clear events from the event list that occurred before time T d - The event of T; the new event is any one of event A, event B, event C, and event D;
[0022] Based on the multimodal event list and expert rules, the event judgment is performed to determine the electricity theft risk level.
[0023] A further optimized solution involves determining the electricity theft risk level based on the multimodal event list and expert rules; including the following methods:
[0024] Inspect the event combinations in the multimodal event list:
[0025] If no incidents occur, then there is no risk of electricity theft.
[0026] If only one of the events A, B, and C occurs, the risk of electricity theft is considered low.
[0027] If only event D occurs, it is considered a risk of electricity theft.
[0028] If any one of events A, B, and C occurs, and event D also occurs, then the risk level is determined to be high.
[0029] If at least two of events A, B, and C occur, and event D occurs, then the risk is classified as extremely high.
[0030] A further optimization scheme is that the electrical data anomaly includes at least one of the following: current imbalance exceeding a preset balance threshold, current change exceeding a current threshold, and power factor change exceeding a power factor threshold.
[0031] The expert rule-based judgment has low computational overhead and fast response speed, enabling real-time and rapid initial screening and response to typical, high-risk electricity theft behaviors, thus meeting the timeliness requirements. For example, a single door opening may be normal maintenance, but if it is accompanied by abnormal current and magnetic field, it can be basically identified as electricity theft, greatly eliminating false alarms caused by a single factor.
[0032] A further optimized approach is that the second preprocessing includes the following methods:
[0033] Acquire basic data from time t−Δt to the current time t, and extract a second feature from the basic data: door opening / closing frequency feature; electrical data statistical feature; vibration signal energy feature and magnetic field change rate feature; the electrical data statistical feature includes mean and variance;
[0034] The second feature is then standardized.
[0035] A further optimization is that the method for constructing the classification model includes:
[0036] Construct a kernel logistic regression model based on multi-kernel learning, in which each second feature is nonlinearly mapped to a high-dimensional space using different kernel functions for linear classification;
[0037] The kernel logistic regression model is as follows:
[0038] The predicted probability P of the electricity theft for the i-th sample i for:
[0039] ;
[0040] Where, α i K(z) represents the nonlinear mapping coefficient of the i-th sample; i (z) represents the kernel function of the i-th sample; b represents the bias term; N represents the sample; z represents the second feature vector of the input; z i Let σ represent the second feature vector of the i-th sample; σ represents the sigmoid function. , where e is the natural base;
[0041] Training and validation samples are constructed by combining historical data determined by expert rules. The parameters α and b of the kernel logistic regression model are trained based on the training samples, and the parameters α and b are optimized based on the validation samples. During the training process, the loss function is minimized by minimizing the regularized negative log-likelihood loss.
[0042] A further optimization scheme is proposed, wherein the loss function is:
[0043] ;
[0044] in, K represents the loss function; K represents the kernel matrix, whose elements are K. ij ;K ij =K(z i , z j )=ϕ(z i ) T ϕ(z j ); ϕ(*) denotes the nonlinear mapping function; T denotes the transpose; λ denotes the regularization parameter; log denotes the logarithmic function; y i This represents the true label of the i-th training sample, with a value of 0 or 1, such as y. i =1 indicates that the i-th training sample is stealing electricity, y i =0 indicates that the i-th training sample is normal; α represents the nonlinear mapping coefficient vector; that is, α=[α1, α2, ..., α...]. N ] T ;α T Kα represents the regularization term.
[0045] This solution also provides an electricity theft early warning system based on the multimodal characteristics of meter boxes, used to implement the aforementioned electricity theft early warning method based on the multimodal characteristics of meter boxes. The system includes:
[0046] The data acquisition module is used to monitor the basic data of the metering box in real time; the basic data includes metering box door status data, metering box vibration data, metering box magnetic field data, and electrical data.
[0047] The first determination module is used to perform a first preprocessing on the basic data, and perform expert rule determination based on the first preprocessed basic data to determine the type of electricity theft risk.
[0048] The second determination module is used to perform a second preprocessing on the basic data for cases with a high risk of electricity theft or above, and then input the preprocessed basic data into the constructed classification model for a second determination.
[0049] The alarm module is used to trigger an electricity theft alarm based on the results of the secondary judgment.
[0050] This solution also provides a computer-readable medium on which a computer program is stored, the computer program being executed by a processor to implement the electricity theft early warning method based on the multimodal characteristics of the metering box as described above.
[0051] Compared with the prior art, the present invention has the following beneficial effects:
[0052] 1. This invention provides a method, system, and medium for electricity theft early warning based on the multimodal characteristics of metering boxes. This solution combines physical quantity monitoring signals (door switch status, vibration, and magnetic field) with electrical quantity monitoring signals to form a multimodal characteristic judgment system. It uses prior knowledge to set expert rules for initial judgment, eliminating false alarms caused by single factors. For complex and concealed electricity theft methods, it uses machine learning models to deeply mine and analyze the multidimensional preprocessed data, which can identify abnormal patterns that are difficult to summarize with expert experience. It further filters out false alarms from a probabilistic level, ensuring high confidence of the reported alarm information.
[0053] 2. This invention provides a method, system, and medium for early warning of electricity theft based on the multimodal characteristics of metering boxes. Expert rule-based judgment has low computational overhead and fast response speed, enabling real-time and rapid initial screening and response to typical, high-risk electricity theft behaviors, meeting the timeliness requirement. For example, a single door opening may be normal maintenance, but if it is accompanied by abnormal current and magnetic field, it can be basically identified as electricity theft, greatly eliminating false alarms caused by a single factor. The machine learning model is only activated for situations with high electricity theft risk, avoiding resource waste and focusing on solving complex and hidden difficult cases, thus providing accuracy assurance.
[0054] 3. This invention provides a method, system, and medium for early warning of electricity theft based on the multimodal features of a metering box; its classification model combines multiple kernel functions, processing the door switch status features, electrical data features, vibration signal features, and magnetic field features with different kernel functions, and then combining them; it improves the logistic regression model from the perspective of kernel logistic regression, mapping the original features to a high-dimensional space through kernel functions, thereby enabling the handling of nonlinear classification problems. Attached Figure Description
[0055] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:
[0056] Figure 1 This is a schematic diagram of a method for early warning of electricity theft based on the multimodal characteristics of a metering box.
[0057] Figure 2 This is a schematic diagram of an electricity theft early warning system based on the multimodal characteristics of a metering box.
[0058] Figure 3 This is a schematic diagram of the principle of a door magnetic induction device;
[0059] Figure 4 This is a picture of a door magnetic sensor device.
[0060] Figure 5 This is a schematic diagram of the power supply wiring for the data acquisition module;
[0061] Figure 6 A schematic diagram of the alarm interface in the control center;
[0062] Figure 7 This is a schematic diagram of the alarm interface in the control center. Detailed Implementation
[0063] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.
[0064] To address the core shortcomings of existing anti-electricity theft technologies, such as reliance on inefficient manual inspections, significant time delays from anomaly occurrence to detection, and frequent false alarms, this solution provides the following embodiments to resolve the aforementioned technical problems:
[0065] Example 1
[0066] This embodiment provides a method for early warning of electricity theft based on the multimodal characteristics of a metering box, such as... Figure 1 As shown, the method includes:
[0067] Step 1: Real-time monitoring of the basic data of the metering box; the basic data includes metering box door status data, metering box vibration data, metering box magnetic field data, and electrical data;
[0068] Step 2: Perform a first preprocessing on the basic data, and then perform expert rule judgment based on the preprocessed basic data to determine the type of electricity theft risk;
[0069] The first preprocessing in step two includes:
[0070] Extract door switch features based on meter box door status: use binary variables {0, 1} to represent door status, where 0 represents closed and 1 represents open; and combine the door opening and closing time to determine whether the meter box door opening time is normal or not.
[0071] Vibration characteristics are extracted from the vibration data in the metering box: a preset vibration threshold is used to compare the vibration data with the vibration threshold to determine whether the vibration data is abnormal.
[0072] The magnetic field characteristics are extracted from the magnetic field data in the measuring box: a preset magnetic field threshold is set, and the magnetic field data is compared with the magnetic field threshold, as well as the magnetic field change is compared with the magnetic field threshold to determine whether the vibration data is abnormal.
[0073] Based on electrical data, electrical features are extracted: preset electrical thresholds are used to compare the Euclidean norm of the electrical data with the corresponding preset electrical thresholds to determine whether the electrical data is abnormal.
[0074] Based on fundamental data, expert rule-based judgment is used to determine the types of electricity theft risks, including:
[0075] An abnormal meter box opening time, abnormal vibration data, abnormal magnetic field data, and abnormal electrical data are designated as events A, B, C, and D, respectively. A multimodal event list is constructed: an event list `event_list` is initialized to store events occurring within time T; when a new event occurs: the new event is added to `event_list`, and the current time T is recorded. d ; Clear events from the event list that occurred before time T d - The event of T; the new event is any one of event A, event B, event C, and event D;
[0076] Electrical data anomalies include at least one of the following: current imbalance exceeding a preset balance threshold, current change exceeding a current threshold, and power factor change exceeding a power factor threshold.
[0077] The risk level of electricity theft is determined by analyzing a multimodal event list and expert rules. This step specifically includes the following methods:
[0078] Inspect the event combinations in the multimodal event list:
[0079] If no incidents occur, then there is no risk of electricity theft.
[0080] If only one of the events A, B, and C occurs, the risk of electricity theft is considered low.
[0081] If only event D occurs, it is considered a risk of electricity theft.
[0082] If any one of events A, B, and C occurs, and event D also occurs, then the risk level is determined to be high.
[0083] If at least two of events A, B, and C occur, and event D occurs, then the risk is classified as extremely high.
[0084] Step 3: For cases with a high risk of electricity theft or above, perform a second preprocessing on the basic data, and input the preprocessed basic data into the constructed classification model for a second determination;
[0085] The second preprocessing includes the following methods:
[0086] Acquire basic data from time t−Δt to the current time t, and extract a second feature from the basic data: door opening / closing frequency feature; electrical data statistical feature; vibration signal energy feature and magnetic field change rate feature; the electrical data statistical feature includes mean and variance;
[0087] The second feature is then standardized.
[0088] The method for constructing the classification model includes:
[0089] S31, Construct a kernel logistic regression model based on multi-kernel learning, in which each second feature is nonlinearly mapped to a high-dimensional space using different kernel functions for linear classification;
[0090] The kernel logistic regression model is as follows:
[0091] The predicted probability P of the electricity theft for the i-th sample i for:
[0092] ;
[0093] Where, α i K(z) represents the nonlinear mapping coefficient of the i-th sample; i (z) represents the kernel function of the i-th sample; b represents the bias term; N represents the sample; z represents the second feature vector of the input; z i Let σ represent the second feature vector of the i-th sample; σ represents the sigmoid function. , where e is the natural base;
[0094] S32, combine historical data determined by expert rules to construct training and validation samples, train the parameters α and b of the kernel logistic regression model based on the training samples, and optimize the parameters α and b based on the validation samples; during the training process, the loss function is minimized by the regularized negative log-likelihood loss.
[0095] Specifically, the loss function in this embodiment is:
[0096] ;
[0097] in, K represents the loss function; K represents the kernel matrix, whose elements are K. ij ;K ij =K(z i , z j )=ϕ(z i ) T ϕ(z j ); ϕ(*) denotes the nonlinear mapping function; T denotes the transpose; λ denotes the regularization parameter; log denotes the logarithmic function; y iThis represents the true label of the i-th training sample, with a value of 0 or 1, such as y. i =1 indicates that the i-th training sample is stealing electricity, y i =0 indicates that the i-th training sample is normal; α represents the nonlinear mapping coefficient vector; that is, α=[α1, α2, ..., α...]. N ] T ;α T Kα represents the regularization term.
[0098] Step 4: Set up an electricity theft alarm based on the results of the second determination.
[0099] Example 2
[0100] This embodiment provides an electricity theft early warning system based on the multimodal characteristics of a metering box, such as... Figure 2 As shown, the system used to implement the electricity theft early warning method based on the multimodal characteristics of the metering box described in Embodiment 1 includes:
[0101] The acquisition module is used to monitor the basic data of the metering box in real time. The basic data includes metering box door status data, metering box vibration data, metering box magnetic field data, and electrical data. Specifically, the acquisition module installed on the metering box includes: a door magnetic induction device for acquiring metering box door status data, an electrical metering device for acquiring electrical data, an electromagnetic induction device for acquiring metering box magnetic field data, and a vibration induction device for acquiring metering box vibration data.
[0102] The door magnetic sensor device includes an opening / closing signal triggering section and a wireless signal transmitting section. The overall circuit is as follows: Figure 3 As shown in the image. Among the reed switch models, the SH-12 type, with its short action and return times, is selected and combined with a bar permanent magnet to determine the opening and closing status of the metering box. A physical image of the sensing module is shown below. Figure 4 As shown; the wireless signal transmitter is powered by a DC3V button battery; during installation, the permanent magnet is fixed to the metering box door frame, and the reed switch is installed on the movable metering box door side, while ensuring that the two are on the same horizontal plane and the distance between them is within 2cm.
[0103] For the power supply of the acquisition module, considering the stability of the power supply, a 100V AC power supply method was selected, drawing power from the high-voltage acquisition terminal. A new circuit interface was fabricated from an AC100V to DC12V power adapter to directly power the signal generator of the acquisition module. This power supply and communication circuitry is as follows: Figure 5 As shown: The AC / DC power conversion module is connected to the main terminal of the acquisition module, and the wireless signal receiver is connected between the auxiliary terminal of the acquisition module and the AC / DC power conversion module.
[0104] The first determination module is used to perform a first preprocessing on the basic data, and perform expert rule determination based on the first preprocessed basic data to determine the type of electricity theft risk.
[0105] The second determination module is used to perform a second preprocessing on the basic data for cases with a high risk of electricity theft or above, and then input the preprocessed basic data into the constructed classification model for a second determination.
[0106] The alarm module is used to trigger an electricity theft alarm based on the secondary judgment result, and it can be used.
[0107] Information transmission in this system consists of two parts. The first part is the transmission of basic data from the metering box to the control center via the on-site acquisition module. The second part is the transmission of electricity theft alarm signals from the control center to the corresponding user's mobile WeChat official account. In this embodiment, the transmission device for transmitting basic data to the control center is a ZigBee wireless transmission module. The basic data includes metering box door status data, metering box vibration data, metering box magnetic field data, and electrical data; it also includes user ID information (e.g., user ID information is determined based on the metering box corresponding to the acquisition module). The control center searches its database for information fields such as transformer area, user name, and meter number based on the user ID. This information is then categorized and classified according to the relevant power supply unit to form a historical ledger. Simultaneously, the real-time electricity theft alarm information is pushed to the user's WeChat official account. The control center's display interface is as follows: Figure 6 and Figure 7 As shown.
[0108] In this embodiment, the monitoring and acquisition module is installed in the metering boxes of dedicated transformer users in 200 transformer substations, covering areas with high risk of electricity theft. The module pushes electricity theft alarm information in real time through a public WeChat account, with an average response time of 6.4 seconds. The alarm information includes: user number, transformer substation affiliation, meter number, and abnormal start time (accurate to the second).
[0109] Results of the investigation: A total of 28 valid alarms were triggered within 3 months;
[0110] On-site verification confirmed 25 cases of electricity theft. Typical methods included: short-circuiting the metering current circuit (9 cases), replacing the internal sampling resistor of the meter (7 cases), adding magnets to interfere with the operation of the meter (6 cases), and damaging the voltage connection (3 cases). A total of 658,000 yuan in electricity costs were recovered. During the same period, the line loss qualification rate of the distribution area increased from 94.1% before installation to 98.6%, and the number of high-loss distribution areas decreased by 32%. Among them, the line loss of 8 distribution areas with long-term line loss of more than 7% was reduced to less than 3%.
[0111] The real-time monitoring system provided in this embodiment, based on door magnetic induction and ZigBee communication technology, combines a reed switch (door magnetic induction) with ZigBee wireless communication technology to achieve real-time, remote, and low-power monitoring of the opening and closing status of the metering box. Through a wireless, low-power, self-organizing ZigBee network, it solves the challenges of long-distance, multi-node, and easy deployment. By combining multi-dimensional feature data with expert rule-based judgment and machine learning-based secondary judgment, false alarms are avoided. This ensures real-time monitoring under normal conditions while preventing alarms from being triggered during legitimate operations by staff, thus improving the system's practicality and user-friendliness.
[0112] This embodiment deeply integrates the data acquisition module, the user's official WeChat account terminal, and the user database to automatically identify and push user information; it enables real-time push and historical query of alarm information, replacing the traditional SMS method, reducing costs and improving user experience; and it achieves unified monitoring and operation and maintenance debugging of multiple distribution areas and multiple metering boxes, relying on a combination of wired and wireless connections to complete data interaction between modules.
[0113] This solution can be used in scenarios that require remote monitoring of the opening and closing status of cabinet doors and low-power wireless transmission, such as monitoring the illegal opening of various power equipment cabinets (e.g., distribution boxes, switch cabinets, charging piles), security monitoring of industrial equipment cabinets, communication base stations, etc. It forms a closed-loop management system of "monitoring-analysis-early warning-response", enabling power supply companies to proactively detect and curb electricity theft.
[0114] Example 3
[0115] This embodiment provides a computer-readable medium storing a computer program. The computer program, when executed by a processor, can implement the electricity theft early warning method based on multimodal characteristics of a metering box as described in Embodiment 1; specifically, it performs the following steps:
[0116] Step 1: Monitor the basic data of the metering box in real time; the basic data includes metering box door status data, metering box vibration data, metering box magnetic field data, and electrical data.
[0117] Step 2: Perform a first preprocessing on the basic data, and then perform expert rule judgment based on the preprocessed basic data to determine the type of electricity theft risk.
[0118] Step 3: For cases with a high risk of electricity theft or above, perform a second preprocessing on the basic data, and input the preprocessed basic data into the constructed classification model for a second determination.
[0119] Step 4: Activate the electricity theft alarm based on the results of the second determination.
[0120] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for early warning of electricity theft based on multimodal features of a metering box, characterized in that the method... include: Real-time monitoring of basic data of the metering box; the basic data includes metering box door status data, metering box vibration data, metering box magnetic field data and electrical data; The basic data undergoes a first preprocessing step, and based on the preprocessed basic data, expert rule judgment is performed to determine the type of electricity theft risk. For cases with a high risk of electricity theft or above, the basic data undergoes a second preprocessing, and the preprocessed basic data is then input into the pre-built classification model for a second determination. An electricity theft alarm will be triggered based on the results of the second assessment.
2. The electricity theft early warning method based on multimodal features of a metering box according to claim 1, characterized in that, The first preprocessing includes: Extract door switch features based on meter box door status: use binary variables {0, 1} to represent door status, where 0 represents closed and 1 represents open; and combine the door opening and closing time to determine whether the meter box door opening time is normal or not. Vibration characteristics are extracted from the vibration data in the metering box: a preset vibration threshold is used to compare the vibration data with the vibration threshold to determine whether the vibration data is abnormal. The magnetic field characteristics are extracted from the magnetic field data in the measuring box: a preset magnetic field threshold is set, and the magnetic field data is compared with the magnetic field threshold, as well as the magnetic field change is compared with the magnetic field threshold to determine whether the vibration data is abnormal. Based on electrical data, electrical features are extracted: preset electrical thresholds are used to compare the Euclidean norm of the electrical data with the corresponding preset electrical thresholds to determine whether the electrical data is abnormal.
3. The electricity theft early warning method based on multimodal features of a metering box according to claim 2, characterized in that, Based on the aforementioned basic data, expert rule-based judgment is performed to determine the type of electricity theft risk, including: An abnormal meter box opening time, abnormal vibration data, abnormal magnetic field data, and abnormal electrical data are designated as events A, B, C, and D, respectively. A multimodal event list is constructed: an event list `event_list` is initialized to store events occurring within time T; when a new event occurs: the new event is added to `event_list`, and the current time T is recorded. d ; Clear events from the event list that occurred before time T d - The event of T; the new event is any one of event A, event B, event C, and event D; Based on the multimodal event list and expert rules, the event judgment is performed to determine the electricity theft risk level.
4. The electricity theft early warning method based on multimodal features of a metering box according to claim 3, characterized in that, Based on the multimodal event list and expert rules, the event judgment is performed to determine the electricity theft risk level; Including methods: Check the event combinations in the multimodal event list: If no incidents occur, then there is no risk of electricity theft. If only one of the events A, B, and C occurs, the risk of electricity theft is considered low. If only event D occurs, it is considered a risk of electricity theft. If any one of events A, B, and C occurs, and event D also occurs, then the risk level is determined to be high. If at least two of events A, B, and C occur, and event D occurs, then the risk is classified as extremely high.
5. The electricity theft early warning method based on multimodal features of a metering box according to claim 3, characterized in that, The electrical data anomalies include at least one of the following: current imbalance exceeding a preset balance threshold, current change exceeding a current threshold, and power factor change exceeding a power factor threshold.
6. The electricity theft early warning method based on multimodal features of a metering box according to claim 1, characterized in that, The second preprocessing includes the following methods: Obtain the basic data from time t−Δt to the current time t, and extract the second feature from the basic data: the door opening and closing frequency feature; Statistical characteristics of electrical data; Vibration signal energy characteristics and magnetic field change rate characteristics; the electrical data statistical characteristics include mean and variance; The second feature is then standardized.
7. The electricity theft early warning method based on multimodal features of a metering box according to claim 2, characterized in that, The method for constructing the classification model includes: Construct a kernel logistic regression model based on multi-kernel learning, in which each second feature is nonlinearly mapped to a high-dimensional space using different kernel functions for linear classification; The kernel logistic regression model is as follows: The predicted probability P of the electricity theft for the i-th sample i for: ； Where, α i K(z) represents the nonlinear mapping coefficient of the i-th sample; i (z) represents the kernel function of the i-th sample; b represents the bias term; N represents the sample; z represents the second feature vector of the input; z i Let σ represent the second feature vector of the i-th sample; σ represents the sigmoid function. , where e is the natural base; Training and validation samples are constructed by combining historical data determined by expert rules. The parameters α and b of the kernel logistic regression model are trained based on the training samples, and the parameters α and b are optimized based on the validation samples. During the training process, the loss function is minimized by minimizing the regularized negative log-likelihood loss.
8. The electricity theft early warning method based on multimodal features of a metering box according to claim 7, characterized in that, The loss function is: ； in, K represents the loss function; K represents the kernel matrix, whose elements are K. ij ;K ij =K(z i , z j )=ϕ(z i ) T ϕ(z j ); ϕ(*) denotes the nonlinear mapping function; T denotes the transpose; λ denotes the regularization parameter; log denotes the logarithmic function; y i This represents the true label of the i-th training sample, with a value of 0 or 1, such as y. i =1 indicates that the i-th training sample is stealing electricity, y i =0 indicates that the i-th training sample is normal; α represents the nonlinear mapping coefficient vector; that is, α=[α1, α2, ..., α...]. N ] T ;α T Kα represents the regularization term.
9. A power theft early warning system based on the multimodal characteristics of a metering box, characterized in that, The system is used to implement the electricity theft early warning method based on the multimodal characteristics of a metering box as described in any one of claims 1-8, the system comprising: The data acquisition module is used to monitor the basic data of the metering box in real time; the basic data includes metering box door status data, metering box vibration data, metering box magnetic field data, and electrical data. The first determination module is used to perform a first preprocessing on the basic data, and perform expert rule determination based on the first preprocessed basic data to determine the type of electricity theft risk. The second determination module is used to perform a second preprocessing on the basic data for cases with a high risk of electricity theft or above, and then input the preprocessed basic data into the constructed classification model for a second determination. The alarm module is used to trigger an electricity theft alarm based on the results of the secondary judgment.
10. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, when executed by a processor, can implement the electricity theft early warning method based on the multimodal characteristics of a metering box as described in any one of claims 1-8.