Electricity larceny prevention method and system for multi-source data fusion of electric energy meter

Abstract

The invention relates to the technical field of electric energy metering and electric power system safety, and discloses an electricity larceny prevention method and system for multi-source data fusion of an electric energy meter, and the method comprises the steps: obtaining and preprocessing current and voltage waveform data and user electricity consumption behavior data; generating multi-dimensional power consumption characteristic parameters, performing screening to form a characteristic vector set, and constructing an abnormal power consumption detection reference set; dynamically verifying power utilization data based on the reference set, and analyzing a phase relation and a mismatch state of current and voltage signals; and generating an electricity consumption behavior deviation index to judge an electricity stealing behavior. The system comprises a multi-source data acquisition module, a feature modeling analysis module, an electrical relation verification module and an electricity stealing judgment module. According to the invention, through multi-source data fusion and multi-dimensional feature analysis, accurate detection of electricity larceny behaviors is realized, accuracy and reliability of electricity larceny prevention are improved, and the method is suitable for power utilization safety management of a smart power grid.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric energy metering and power system safety, in particular to an electricity stealing prevention method and system based on multi-source data fusion of electric energy meters. BACKGROUND

[0002] In the process of power system operation, electricity stealing has been an important problem affecting the economic benefits of power supply enterprises and disrupting the normal power supply and use order. Traditional electricity stealing prevention methods mainly rely on manual patrol, mechanical structure anti-tampering design of electric energy meters and simple electrical parameter monitoring, which have obvious limitations.

[0003] From the perspective of technical implementation, early mechanical electric energy meters mainly prevent external human damage through physical methods such as lead seals, but with the development of technology, illegal individuals use more covert technical means to steal electricity, such as bypassing metering devices, modifying circuit connections, and falsifying electricity data, making it difficult for physical electricity stealing prevention measures to effectively respond. Monitoring systems based on a single electrical parameter (such as current and voltage effective value) cannot fully capture changes in electricity characteristics caused by electricity stealing due to the single dimension of the data. For example, when electricity thieves steal electricity by changing the phase relationship of current and voltage, introducing harmonic interference or falsifying load curves, systems that only monitor effective values may not be able to detect abnormalities in time.

[0004] At the data processing level, traditional methods lack the ability to analyze multi-source data. There is an inherent correlation between the current waveform, voltage waveform and user electricity behavior data (such as device start and stop time, load power change) generated during the operation of the electric energy meter, and single-dimensional data analysis can easily overlook the characteristics of abnormal electricity behavior in different data domains. For example, under normal electricity conditions, the phase relationship of current and voltage, harmonic components and load power change have certain regularity, while electricity stealing behavior may cause abnormal correlation between these parameters, but traditional systems cannot effectively mine the correlation characteristics of multi-dimensional data.

[0005] From the actual application scenario, with the advancement of smart grid construction, user electricity scenarios are becoming increasingly complex, and the electricity usage patterns of industrial users, commercial users and residential users differ significantly, making it difficult for traditional uniform threshold detection methods to adapt to the individualized electricity characteristics of different users. For example, some industrial equipment may produce short-term high current and voltage fluctuations when starting, and if a fixed threshold is used for judgment, normal electricity behavior may be misjudged as electricity stealing, or electricity stealing behavior may be missed due to unreasonable threshold settings. In addition, manual patrol methods have low efficiency and poor timeliness, and cannot achieve real-time monitoring and early warning of electricity stealing behavior, and require a large amount of human and material resources.

[0006] In terms of technology development trends, the smart grid puts forward higher requirements for electricity theft prevention technology, and needs to realize the transformation from passive defense to active monitoring, from single parameter detection to multi-source data fusion analysis. The existing electricity theft prevention technology has deficiencies in the comprehensiveness of data collection, the depth of feature analysis, and the accuracy of anomaly detection, and cannot meet the demand of the smart grid for power consumption safety and metering reliability. Therefore, an electricity theft prevention method and system that can fuse multi-source data, deeply mine power consumption features, and realize dynamic and accurate detection are urgently needed to effectively deal with the increasingly complex electricity theft methods and protect the safe and stable operation of the power system and the legitimate rights and interests of power supply enterprises. SUMMARY

[0007] The purpose of the present application is to provide an electricity theft prevention method and system for multi-source data fusion of electric energy meters to solve the problems raised in the background art.

[0008] To achieve the above purpose, the present application provides the following technical solution: an electricity theft prevention method for multi-source data fusion of electric energy meters, the method comprising: S1: obtaining current waveform data, voltage waveform data and user terminal power consumption behavior data of an electric energy meter, and performing data preprocessing on the current waveform data, voltage waveform data and user terminal power consumption behavior data; S2: analyzing the preprocessed data to generate multi-dimensional power consumption feature parameters, performing feature screening to form a feature vector set according to the multi-dimensional power consumption feature parameters, and constructing an abnormal power consumption detection benchmark set according to the feature vector set; S3: dynamically verifying user power consumption data according to the abnormal power consumption detection benchmark set, identifying the phase relationship between the current signal and the voltage signal in the metering loop using multi-source data synchronization technology, and quantitatively analyzing the mismatch state of the current signal and the voltage signal; S4: generating a power consumption behavior deviation index according to the verification result of the abnormal power consumption detection benchmark set, and determining whether there is electricity theft behavior according to the power consumption behavior deviation index.

[0009] Preferably, in S1, the system presets a collection period, acquires multi-loop current waveform data, multi-loop voltage waveform data of the electric energy meter in the preset collection period, and user terminal uploaded power consumption behavior data; the current waveform data includes fundamental component and harmonic component; the voltage waveform data includes voltage effective value and instantaneous fluctuation data; the power consumption behavior data includes device start-stop time sequence and load power change data; the current waveform data, voltage waveform data and power consumption behavior data are subjected to data preprocessing, the data preprocessing includes signal filtering, data alignment, outlier rejection and normalization processing; the current waveform data is subjected to preprocessing to generate current feature sequence, and the voltage waveform data is subjected to preprocessing to generate voltage feature sequence; the current feature sequence and the voltage feature sequence of different loops are associated and matched to form a multi-loop electrical feature set.

[0010] Preferably, in S2, the following steps are included: S201: statistical feature parameters are extracted from the preprocessed current feature sequence, voltage feature sequence and power consumption behavior data, respectively, the statistical feature parameters include standard deviation, kurtosis coefficient and zero-crossing rate, and a multi-dimensional power consumption feature parameter is generated by combining each statistical feature parameter according to a preset weight; S202: a dynamic threshold is set for the multi-dimensional power consumption feature parameter in the multi-loop electrical feature set, a feature parameter meeting the threshold condition is selected to form an initial feature vector set, and a feature parameter not meeting the threshold condition is discarded; S203: each feature parameter in the initial feature vector set is subjected to cross-loop comparative analysis, a difference amount of the feature parameter in different loops in the same time window is extracted, an absolute value of the difference amount is calculated and marked as a loop deviation parameter; S204: the loop deviation parameter is compared with a preset deviation threshold, a feature parameter exceeding the deviation threshold is selected to join an abnormal power consumption detection reference set, and a load power mutation feature is supplemented to the abnormal power consumption detection reference set according to the user terminal power consumption behavior data.

[0011] Preferably, in S3, the mismatch state of the current signal and the voltage signal is subjected to quantitative analysis, including the following steps: S301: current instantaneous value and voltage instantaneous value of the metering loop are collected in real time, a phase angle difference value of the current and the voltage is calculated, and if the phase angle difference value exceeds a preset tolerance range, an abnormal phase event is determined; S302: the number of abnormal phase events in a preset period is counted as , and the load power change amplitude in the user terminal power consumption behavior data is acquired synchronously as ; S303: according to the number of abnormal phase events and the load power change amplitude The phase mismatch index is calculated, and the specific calculation formula is as follows: ; If the phase mismatch index exceeds the preset alarm threshold, the electricity stealing suspicion mark is triggered.

[0012] Preferably, in S4, the verification results of the abnormal power consumption detection reference set in the same user historical power consumption data are extracted, the deviation degrees of each detection reference are weighted and summed to generate a power consumption behavior deviation degree index, and if the power consumption behavior deviation degree index exceeds the preset threshold for a continuous number of times, it is determined that there is electricity stealing behavior.

[0013] Preferably, the application also includes an electricity meter multi-source data fusion anti-electricity stealing system, which is applied to the above-mentioned electricity meter multi-source data fusion anti-electricity stealing method and includes a multi-source data acquisition module, a feature modeling analysis module, an electrical relationship verification module, and an electricity stealing judgment module. The multi-source data acquisition module is used to acquire current waveform data, voltage waveform data, and user terminal power consumption behavior data of the electricity meter and perform data preprocessing. The feature modeling analysis module is used to analyze the preprocessed data to generate multi-dimensional power consumption feature parameters and construct an abnormal power consumption detection reference set. The electrical relationship verification module is used to dynamically verify user power consumption data according to the abnormal power consumption detection reference set and analyze the mismatch state of the current signal and the voltage signal. The electricity stealing judgment module is used to generate a power consumption behavior deviation degree index and judge electricity stealing behavior.

[0014] Preferably, the multi-source data acquisition module includes an electrical quantity acquisition unit and a user behavior acquisition unit. The electrical quantity acquisition unit is used to acquire multi-loop current waveform data and voltage waveform data of the electricity meter. The user behavior acquisition unit is used to acquire device start-stop time series and load power change data of the user terminal.

[0015] Preferably, the feature modeling analysis module includes a feature extraction unit and a reference construction unit. The feature extraction unit is used to extract standard deviation, kurtosis coefficient, and zero-crossing rate feature parameters from the preprocessed data. The reference construction unit is used to filter feature parameters that meet threshold conditions and supplement load power mutation features to the abnormal power consumption detection reference set.

[0016] Preferably, the electrical relationship verification module includes a phase monitoring unit and a mismatch calculation unit. The phase monitoring unit is used to detect the phase angle difference between the current and the voltage and identify abnormal phase events. The mismatch calculation unit is used to calculate a phase mismatch index in combination with the load power variation amplitude.

[0017] Preferably, the electricity stealing judgment module comprises a deviation calculation unit and a judgment unit. The deviation calculation unit is used to weight and sum the deviations of the abnormal electricity consumption detection criteria. The judgment unit is used to judge the electricity stealing behavior according to the continuous triggering number of the deviation index.

[0018] Compared with the prior art, the present application has the following advantages: In the data acquisition and preprocessing link, the system acquires multi-loop current waveform (including fundamental and harmonic components), voltage waveform (including effective value and instantaneous fluctuation data), user equipment start-stop time sequence, load power variation and other multi-source data through preset acquisition cycle, and ensures the accuracy and consistency of the data through signal filtering, data alignment, abnormal value elimination and normalization and other preprocessing operations. This process not only realizes comprehensive acquisition of electricity consumption information, but also provides a rich data basis for subsequent feature analysis through the construction of multi-loop electrical feature set, solving the problem that traditional single parameter acquisition cannot capture complex electricity stealing features.

[0019] The feature modeling analysis module can effectively identify the feature parameters with significant abnormalities by extracting statistical feature parameters such as standard deviation, kurtosis coefficient and zero-crossing rate, and combining dynamic threshold screening and cross-loop comparison analysis, forming an abnormal electricity consumption detection criterion set. At the same time, the load power mutation feature is supplemented, so that the criterion set more comprehensively reflects the feature range of normal electricity consumption behavior. This multi-dimensional feature extraction and dynamic criterion construction method overcomes the limitations of traditional fixed threshold detection, can adapt to the individualized electricity consumption mode of different users, and reduces the probability of false judgment and omission. For example, through the analysis of cross-loop deviation parameters, the abnormal differences of different loop feature parameters caused by electricity stealing can be found in time, and the introduction of load power mutation feature can effectively capture the abnormal power changes that may be accompanied by electricity stealing behavior.

[0020] The electrical relationship verification module utilizes multi-source data synchronization technology to monitor the phase angle difference of current and voltage in the metering loop in real time, calculates the phase mismatch index by counting the number of abnormal phase events and the amplitude of load power change. This method dynamically combines the phase relationship of electrical parameters with load changes, and can accurately quantify the mismatch state of current and voltage signals. When the phase relationship is abnormal due to electricity stealing behavior (such as changing the wiring method to make the current and voltage phase deviate from the normal range) or the load power appears abnormal fluctuation, the phase mismatch index can timely reflect this abnormality, trigger the electricity stealing suspicion mark, and realize real-time early warning of electricity stealing behavior. Compared with the traditional method of monitoring only a single parameter of phase or power, this module significantly improves the sensitivity and accuracy of abnormal detection through multi-parameter fusion analysis.

[0021] The electricity stealing judgment module analyzes the verification results of the abnormal electricity detection reference set in the historical electricity data of the same user, generates an electricity behavior deviation index using weighted summation, and determines electricity stealing behavior in combination with the number of continuous triggers. This dynamic analysis method based on historical data fully considers the individual characteristics of user electricity behavior and the regularity in time series, and can effectively distinguish between normal electricity fluctuations and abnormal deviations caused by electricity stealing. For example, for some users who have short-term electricity mode changes due to production process adjustment, the system can avoid misjudgment through learning from historical data and adjusting the dynamic reference; for continuous abnormal deviations, the system can timely determine electricity stealing behavior through the continuous trigger mechanism, improving the reliability and stability of the determination result.

[0022] In addition, the electricity stealing prevention system constructed by the present application realizes the full-process automatic processing from data collection, feature analysis, abnormal detection to electricity stealing judgment through the organic cooperation of the multi-source data collection module, the feature modeling analysis module, the electrical relationship verification module and the electricity stealing judgment module. The system not only can monitor the electricity state in real time, but also can deeply mine the feature performance of electricity stealing behavior in different data domains through multi-dimensional data fusion and dynamic modeling, providing an efficient and accurate technical means for the electricity safety management of smart grid. Compared with traditional electricity stealing prevention technology, the present application has significant improvement in data utilization efficiency, feature analysis depth and abnormal detection ability, and can effectively cope with the increasingly complex electricity stealing means, and guarantee the safe and stable operation of the power system and the economic benefit of the power supply enterprise. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The working principle diagram of the electricity stealing prevention method and system of the electric energy meter multi-source data fusion described in the present application; Figure 2 The design diagram of multi-source data collection preprocessing; Figure 3 The design diagram of abnormal electricity detection reference construction; Figure 4 Design diagram for use behavior deviation degree determination. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0025] Please refer to Figures 1-4 The present application relates to a power stealing prevention method based on power meter multi-source data fusion. The method realizes accurate identification of power stealing behavior through the whole process design of multi-dimensional data acquisition, feature modeling, electrical relationship verification and power stealing behavior determination. The specific implementation steps are as follows: S1: Obtain current waveform data, voltage waveform data and user terminal power consumption behavior data of the power meter, and perform data preprocessing on the current waveform data, voltage waveform data and user terminal power consumption behavior data; S2: Analyze the preprocessed data to generate multi-dimensional power consumption feature parameters, perform feature screening to form a feature vector set according to the multi-dimensional power consumption feature parameters, and construct an abnormal power consumption detection benchmark set according to the feature vector set; S3: Dynamically verify the user power consumption data according to the abnormal power consumption detection benchmark set, identify the phase relationship between the current signal and the voltage signal in the metering loop by using multi-source data synchronization technology, and quantitatively analyze the mismatch state of the current signal and the voltage signal; S4: Generate a power consumption behavior deviation degree index according to the verification result of the abnormal power consumption detection benchmark set, and determine whether there is power stealing behavior according to the power consumption behavior deviation degree index.

[0026] The present application will be further described below in combination with Examples 1 to 5: Example 1: The multi-source data acquisition module includes an electrical quantity acquisition unit and a user behavior acquisition unit, which are used to obtain current waveform data, voltage waveform data and user terminal power consumption behavior data of the power meter and perform data preprocessing.

[0027] The electrical quantity acquisition unit realizes real-time acquisition of multi-loop electrical data through the sensors and acquisition circuits built in the electric energy meter. Taking a three-phase four-wire electric energy meter as an example, its built-in three-way Hall current sensor and three-way voltage acquisition channel correspond to A, B, and C three-phase incoming line loops, and can be extended to connect to the current transformers (CT) and voltage transformers (PT) of up to four branch circuits. The range of the Hall current sensor is 0-100A, the resolution is 0.1A, and the linear measurement of alternating current is realized by using the magnetic balance principle. The instantaneous value data of the fundamental wave current (50Hz) and the 3rd, 5th, 7th, etc. harmonic components can be synchronously acquired, and the sampling frequency is set to 1kHz. Each sampling point contains the current amplitude, phase, and harmonic number identifier. The voltage acquisition channel obtains the voltage instantaneous value of each phase through a resistance voltage division network, the range is 0-400V (phase voltage), the resolution is 0.1V, and the voltage effective value and instantaneous fluctuation data (such as the amplitude and duration of temporary rise, temporary drop, and interruption event) can be acquired. The sampling frequency is kept synchronous with the current acquisition. Each 15-minute is a data acquisition cycle, and each cycle generates an original data file containing the current waveform (fundamental wave and harmonic components) and voltage waveform (effective value and instantaneous value) of each loop. The data format is a binary file, and the data volume of a single cycle is about 50MB (calculated at 4 loops and 1kHz sampling rate).

[0028] The user behavior acquisition unit receives the power consumption behavior data uploaded by the user terminal through the wireless communication module. The user terminal includes smart sockets, smart home appliance controllers and other devices, each terminal is configured with a Zigbee communication module, complies with the IEEE802.15.4 protocol, the communication frequency band is 2.4GHz, the transmission rate is 250kbps, and the communication distance can reach 30 meters in an indoor environment. The user behavior data includes two categories: one is the time sequence of device start and stop, such as the specific timestamp (accurate to seconds) of each start and stop event of air conditioners, water heaters, refrigerators and other home appliances, and the device type identifier (such as the device ID code corresponding to the air conditioner, water heater and other categories); the other is the load power change data, which is acquired in real time by the power metering chip (such as ADE7758) built in the smart socket, with a sampling interval of 1 minute, recording the active power, reactive power and power factor change curve of each device. In each acquisition cycle (synchronized with the electrical quantity acquisition cycle, i.e. 15 minutes), the user behavior acquisition unit polls each user terminal through the Zigbee network to collect all start and stop event records and power change data in the cycle, forming a structured data file (such as JSON format) containing device ID, timestamp, power value and other fields.

[0029] The data preprocessing stage performs a standardization process on three types of raw data. First, signal filtering is performed on current waveform data and voltage waveform data: a Butterworth band-pass filter is used, with a passband of 45Hz-55Hz (fundamental component) and 135Hz-165Hz, 225Hz-275Hz, 315Hz-385Hz (corresponding to 3rd, 5th, 7th harmonic components) for current signals, and a decay slope of -40dB / octave to eliminate high-frequency noise and DC offset; for voltage signals, the passband is set to 45Hz-55Hz (fundamental) and the noise suppression band to retain the effective frequency band of voltage transient events. The filtered data is time-aligned by a synchronous clock, and the system uses a high-precision RTC clock module (error ±5ppm) and periodically synchronizes with the cloud server through NTP protocol to ensure that the timestamp error of multi-source data is less than 1ms. For current effective value, voltage fluctuation amplitude and other parameters, the IQR (interquartile range) algorithm is used to remove outliers: the 25th percentile (Q1) and 75th percentile (Q3) of the data are calculated, and the outlier range is determined as less than Q1-1.5×IQR or greater than Q3+1.5×IQR (where IQR=Q3-Q1), and outliers outside the range are replaced by interpolation (such as linear interpolation using the adjacent two points).

[0030] The normalization processing link maps the feature parameters of different dimensions to the standardization. For current harmonic components (such as 3rd harmonic current amplitude), voltage instantaneous value and other data, the Min-Max normalization method is used, the formula is where and are the minimum and maximum values of the parameter in the historical data, and the data is mapped to the [0,1] interval. For device start-stop time series, a time window encoding method is used, dividing a 15-minute period into 900 1-second time units, each unit corresponding to a binary value (0 represents no start-stop event, 1 represents a start-stop event), forming a one-hot encoding vector of start-stop events. For load power change data, a moving average method is used for smoothing, calculating the average value of the first 3 sampling points as the power feature value of the current period to reduce the influence of random fluctuations.

[0031] After the pre-processing is completed, the current characteristic sequence and the voltage characteristic sequence of different loops are associated and matched. Taking the incoming line loop L1 as an example, the current characteristic sequence includes the fundamental current amplitude sequence, the 3rd harmonic current phase sequence, etc., and the voltage characteristic sequence includes the phase voltage effective value sequence, the voltage sag duration sequence, etc. A one-to-one correspondence is established through the loop number (such as L1, L2, L3 and branch loop ID) to form a multi-loop electrical characteristic set. The set is stored in the form of a structured data table, and the fields include loop ID, timestamp, current fundamental amplitude, current 3rd harmonic phase, voltage effective value, voltage sag number, etc., providing standardized input data for subsequent feature modeling analysis.

[0032] The hardware selection of the entire multi-source data acquisition module meets the industrial-level reliability requirements. The Hall current sensor and the voltage acquisition circuit adopt electromagnetic shielding design, and the anti-interference capability meets the GB / T17626.2 (electrostatic discharge immunity) and GB / T17626.4 (electrical fast transient burst immunity) standards. The wireless communication module supports AES-128 encrypted transmission to ensure the security of user behavior data during transmission. The data preprocessing algorithm is realized by FPGA acceleration. For 4-loop data with a sampling rate of 1 kHz, the filtering, alignment, denoising and normalization processing can be completed within 200 ms, meeting the real-time requirements.

[0033] Embodiment 2: The feature modeling analysis module includes a feature extraction unit and a reference construction unit, which are used to analyze the pre-processed data to generate multi-dimensional power consumption feature parameters and construct an abnormal power consumption detection reference set.

[0034] The feature extraction unit extracts statistical feature parameters from the pre-processed current characteristic sequence, voltage characteristic sequence and power consumption behavior data. Taking the current characteristic sequence as an example, for the fundamental current effective value sequence (unit: A) of a certain loop, the standard deviation is calculated to reflect the load fluctuation degree, and the formula is wherein is the current effective value at the th time in the sequence, is the mean value of the sequence, is the sequence length (such as 9000 sampling points in each collection period, ). For the voltage waveform data in the voltage characteristic sequence, the kurtosis coefficient is calculated to represent the degree of deviation of the waveform from the normal distribution, and the formula is wherein is the voltage instantaneous value, is the voltage mean value, is the voltage standard deviation. The calculation of the current zero-crossing rate is realized by a threshold detection algorithm, which sets the interval of the current absolute value less than 0.1 A as the zero-crossing area, and counts the number of times that the current crosses the zero axis from the positive half cycle to the negative half cycle or vice versa within a unit time (e.g., 1 second), forming the zero-crossing rate characteristic parameter (unit: times / second).

[0035] For the device start-stop time sequence in the user terminal electricity consumption behavior data, the feature extraction unit converts it into a start-stop event frequency feature, that is, the number of start-stops of a specific device within a collection period (15 minutes) (e.g., the number of start-stops of the air conditioner within the period). The load power change data is generated by calculating the absolute value of the power difference between adjacent two sampling points (1 minute interval), forming the power change rate characteristic parameter (unit: kW / min), for example, the power of a certain device at time t is 2 kW, and the power at time t+1 is 5 kW, then the power change rate is 3 kW / min.

[0036] The generation of multi-dimensional electricity consumption characteristic parameters needs to be combined with preset weights. Taking the current standard deviation, the voltage kurtosis coefficient, and the current zero-crossing rate as examples, the preset weights are 40%, 30%, and 30%, respectively. The weighted sum formula is Generate comprehensive characteristic parameters , wherein is the normalized value of the zero-crossing rate. This weight system can be determined by manual setting or machine learning algorithm (such as linear regression) optimization, to ensure that the representation ability of each characteristic parameter to electricity stealing behavior is reasonably reflected.

[0037] The reference construction unit performs dynamic threshold screening on the multi-dimensional electricity consumption characteristic parameters in the multi-loop electrical characteristic set. The setting of the dynamic threshold is based on the statistical characteristics of the historical data. Taking the line current effective value standard deviation as an example, first collect all the current standard deviation data of this loop in the past 7 days, calculate the mean value and the standard deviation , and set the dynamic threshold to (corresponding to the 95% confidence interval in statistics). When the current period calculation of the current standard deviation exceeds the threshold, it is determined that the parameter meets the threshold condition and is included in the initial feature vector set; parameters that do not meet the threshold condition (such as standard deviation less than ) are temporarily discarded.

[0038] After forming the initial feature vector set, the reference construction unit performs cross-loop comparison analysis on each characteristic parameter in it. Taking the current effective value standard deviation of the incoming line L1 and the branch circuit L2 as an example, within the same time window (e.g., 1 hour), the difference between the two is calculated and the absolute value of the difference is marked as the inter-loop deviation parameter. The preset deviation threshold is set based on the characteristic difference range of the normal operation of the loop, for example, for branch loops of the same type of load, the inter-loop current standard deviation threshold is set to 10A. If exceeds the threshold (such as calculated , it is determined that the characteristic parameter has an abnormal deviation, and it is added to the abnormal power consumption detection reference set.

[0039] In addition, the reference construction unit needs to supplement the load power mutation characteristics according to the user terminal power consumption behavior data. The judgment standard of the load power mutation is that the load power change amplitude exceeds the preset threshold (such as 5kW). When the power change data reported by the user terminal meets the condition, the corresponding power change rate, mutation occurrence time and other characteristic parameters are extracted and directly included in the abnormal power consumption detection reference set. For example, the load power of a user jumps from 1kW to 7kW during 14:00-14:10, , the related characteristics (such as power values before and after the mutation, duration) of the mutation event are supplemented to the reference set.

[0040] The construction process of the abnormal power consumption detection reference set needs to ensure the timeliness and dynamic adaptability of the characteristic parameters. The system defaults to update the reference set at 00:00 every day, and recompute the dynamic threshold, inter-loop deviation threshold and load power mutation threshold based on all power consumption data of the previous day, and eliminate the characteristic parameters that have not triggered an anomaly for 7 consecutive days, and include newly identified abnormal characteristic parameters, to ensure that the reference set always reflects the current normal power consumption mode of the user. The reference set is stored in the form of key-value pairs, each characteristic parameter includes parameter name (such as "L1 loop current standard deviation"), data type (such as floating point type), threshold range (such as greater than 25A), weight coefficient (such as 0.2) and other attributes, for subsequent dynamic verification link to call.

[0041] The algorithm implementation of the feature modeling analysis module is based on the Python programming language, and uses the NumPy library for numerical calculation, and the Pandas library for data cleaning and structured processing. For a data set containing 4 loops and 20 characteristic parameters for each loop, the feature extraction unit can complete all parameter calculations within 500ms, and the dynamic threshold screening and cross-loop comparison analysis of the reference construction unit takes about 800ms, meeting the processing frequency requirement of once every 15 minutes. On the hardware level, this module can be deployed on an ARM processor of an edge computing terminal or an x86 architecture server of a cloud server, and multiple user feature modeling tasks can be processed in parallel through multi-threading technology to improve the overall processing efficiency of the system.

[0042] Example 3: The electrical relationship verification module includes a phase monitoring unit and a mismatch calculation unit, which are used to dynamically verify user electricity consumption data based on the abnormal electricity consumption detection benchmark set and analyze the mismatch status of current signals and voltage signals.

[0043] The phase monitoring unit uses a high-precision sampling circuit and phase-locked loop (PLL) technology to achieve real-time detection of the phase angle difference between current and voltage. Taking phase A of a three-phase four-wire circuit as an example, the current transformer (CT) and voltage transformer (PT) configured in the metering loop respectively collect the instantaneous current value. With instantaneous voltage value The sampling frequency is 10kHz, ensuring 200 sampling points are included within each power frequency cycle (20ms). The phase-locked loop circuit tracks the phase of the voltage signal in real time, generating a reference signal that is in phase and frequency with the fundamental voltage wave. ,in The fundamental angular frequency is denoted as . The fundamental component of the current and voltage signals is extracted using Discrete Fourier Transform (DFT) to obtain the fundamental current vector. and voltage fundamental vector ,in The amplitude of the fundamental current wave. The voltage fundamental amplitude, The phase angle of the fundamental current wave. This represents the phase angle of the fundamental voltage wave. The phase angle difference is... The calculation formula is: when hour( (The preset tolerance angle, in degrees), is used to determine an abnormal phase event. The default value is 15°, which can be adjusted through system parameter configuration according to different power consumption scenarios (such as residential power consumption and industrial power consumption). When an abnormal phase event occurs, the system records the event time and phase difference. Information such as loop number is involved, and the number of events is accumulated. .

[0044] The mismatch calculation unit combines the load power variation data from user terminal electricity consumption behavior data. (Unit: kW) Quantitative analysis of phase mismatch status. Load power variation amplitude. Defined as the difference between the maximum and minimum active power within the current acquisition period, this value is collected in real time by smart sockets or load controllers and uploaded to the system. The mismatch calculation unit counts the number of abnormal phase events within a preset time period (e.g., 30 minutes). and simultaneously acquire data within that time period. The phase mismatch index is calculated using the following formula: wherein, is the number of abnormal phase events in the preset period, and is a positive integer; is the load power variation amplitude, and the value range is a non-negative number (unit: kW). The formula amplifies the influence of phase abnormalities in low power variation scenarios through the logarithmic term and reflects the comprehensive effect of high power variation and low event number through the fractional term . When , the phase mismatch index is 0, indicating that the phase relationship between current and voltage is normal.

[0045] The system sets a preset alarm threshold for the phase mismatch index (unit: dimensionless), and the default value is 100. When , the electricity theft suspicion flag is triggered, an alarm record containing the loop number, calculation period, phase mismatch index value, etc. is generated, and is pushed to the power monitoring center through wired network (such as TCP / IP protocol) or wireless communication (such as 4G). The alarm threshold can be dynamically adjusted according to historical data statistics, for example, by analyzing the normal electricity data of a user in the past 30 days, calculating the 95% quantile of the phase mismatch index as the new threshold to adapt to the electricity characteristics of different users.

[0046] The hardware implementation of the electrical relationship verification module relies on the cooperation of the metering chip and the edge computing terminal. The metering chip (such as ATT7022E) integrates multi-channel ADC sampling, DFT operation and phase calculation functions, and can output real-time parameters such as fundamental phase angle of current and voltage, active power, etc. The sampling accuracy is 0.2S level. The edge computing terminal uses ARM Cortex-M7 processor with a main frequency of 168MHz, reads the metering chip data through SPI interface, executes abnormal phase event detection and phase mismatch index calculation logic, and generates phase mismatch index calculation results every 10 minutes. The communication module supports Modbus RTU protocol, and can upload real-time data to cloud server for long-term storage and trend analysis.

[0047] In the data processing process, the phase monitoring unit and the mismatch calculation unit adopt a pipeline architecture design: the first 10 minutes are used for real-time acquisition and phase angle calculation, the middle 10 minutes are used for event statistics and power variation amplitude acquisition, and the last 10 minutes are used for phase mismatch index calculation and result output. This design ensures that the calculation tasks in each 30-minute period are evenly distributed, avoids the peak load of the processor caused by concentrated calculation, and improves the system stability. At the same time, the module has a watchdog timer, which automatically restarts the hardware unit when there is no valid result output for 3 consecutive calculation periods (i.e. 90 minutes), and restores the normal working state.

[0048] The calculation process of the phase mismatch index strictly follows the principle of dimensional consistency: the logarithmic term The unit of the constant 1 is consistent with the unit of , ensuring the legality of the logarithmic operation; the fractional term , the numerator is the power change amplitude (kW), and the denominator is the number of events (dimensionless), and the result unit is kW / time, which matches the dimensionless result of the logarithmic term through the coefficient product. The final phase mismatch index is a dimensionless quantity, which is convenient for cross-user and cross-period comparison and analysis.

[0049] The anti-interference design of this module includes digital filtering and outlier rejection mechanism: 50Hz notch filter is used for the collected current and voltage instantaneous value to suppress power frequency harmonic interference; the load power change amplitude is filtered by moving average filter (window length is 3 collection periods) to eliminate the influence of random noise. When it is detected that more than 1.5 times the maximum declared capacity of the user, it is automatically marked as invalid data, avoiding false calculation caused by equipment failure or communication error code.

[0050] Example 4: The electricity stealing judgment module includes a deviation calculation unit and a judgment unit, which are used to generate electricity consumption behavior deviation index and judge electricity stealing behavior. The deviation calculation unit quantifies the abnormality of electricity consumption behavior by analyzing the difference between the current electricity consumption data of the user and the historical benchmark; the judgment unit realizes the comprehensive judgment of electricity stealing behavior based on the continuous triggering of the deviation index.

[0051] Take a residential user as an example, whose electricity consumption equipment includes air conditioner (2kW), water heater (3kW), refrigerator (0.2kW) and so on. The system first extracts the historical electricity consumption data of the user in the past 30 days, and constructs the historical benchmark for the feature parameters in the abnormal electricity consumption detection benchmark set. For example, the benchmark set includes "line loop current effective value standard deviation", "L1 and L2 loop current standard deviation difference", "load power mutation amplitude", "phase mismatch index" and other feature parameters. In the historical benchmark, the average value of the line loop current effective value standard deviation is 12A, reflecting the average level of the user's daily load fluctuation; the average value of the current standard deviation difference between L1 (air conditioner loop) and L2 (water heater loop) is 5A, reflecting the difference in load characteristics of different equipment; the common amplitude of load power mutation is 3-4kW (such as the air conditioner starting power from 0.5kW to 2.5kW, the change amplitude is 2kW, but occasionally more than 5kW mutation occurs due to the simultaneous start of multiple equipment); the historical maximum value of the phase mismatch index is 80, corresponding to the temporary phase fluctuation when the air conditioner starts.

[0052] ​When the current user electricity data enters the judgment link, the deviation calculation unit compares the current value of each characteristic parameter with the historical reference value one by one. Taking the "line loop current effective value standard deviation" as an example, if the current calculation value is 25A, which is significantly higher than the historical average value of 12A, the system calculates the deviation of the parameter through the distance measurement method (such as Euclidean distance), reflecting the difference between the current value and the historical reference. Similarly, if the current value of the difference between the current value of the current standard deviation of L1 and L2 loops is 18A, which is far higher than the historical average value of 5A, it indicates that the load characteristics of the two loops deviate abnormally, and there may be unauthorized equipment access or loop wiring abnormalities.

[0053] The analysis of the load power mutation feature combines with the device start-stop time sequence. For example, the user terminal reports that the air conditioner and water heater are started at 14:00, the load power jumps from 0.3kW (refrigerator running) to 5.3kW, the mutation amplitude is 5kW, which reaches the judgment standard of "power change exceeding 5kW within 10 minutes" in the reference set, and the system includes the related features (such as mutation occurrence time, involved device type) of the mutation event into the deviation calculation. In terms of phase mismatch index, if 5 abnormal phase events (phase angle difference value exceeds 15°) are detected within the current 30 minutes, the load power change amplitude is 6kW, according to the pre-designed calculation logic, the index reflects the abnormality of the phase relationship between current and voltage and the relevance to power change.

[0054] The deviation of each characteristic parameter is weighted and summed through pre-set weight, forming a comprehensive electricity behavior deviation index. The weight distribution is based on the characteristic of the feature to the electricity stealing behavior, for example, the phase mismatch index weight is higher (such as 30%), because it directly reflects the abnormality of the electrical relationship of the metering loop; the load power mutation weight is second (such as 20%), which reflects the irregular change of electricity behavior; the inter-loop deviation parameter and the current standard deviation weight are 25% and 15% respectively, and the weight of other parameters such as voltage kurtosis coefficient is 10% in total. Assuming that the current parameter deviation is: phase mismatch index deviation 0.9, inter-loop deviation deviation 0.8, load power mutation deviation 0.7, and current standard deviation deviation 0.6, the weighted sum of the comprehensive index is 0.9x30%+0.8x25%+0.7x20%+0.6x15%=0.79, which is close to the pre-set threshold (such as 0.8).

[0055] The judgment unit determines the electricity stealing behavior according to the number of continuous triggering of the deviation degree index. The system sets the detection period to be 15 minutes. If the deviation degree indexes of the user in the three continuous periods of 14:00, 14:15 and 14:30 are 0.79, 0.82 and 0.85 respectively, all of which exceed the preset threshold 0.8, the electricity stealing judgment logic is triggered. At this time, the system automatically locks the multi-source data of the period of 14:00-14:30, including the current and voltage waveforms of each loop, the device start-stop record, the phase abnormal event details and the like, and generates an alarm report containing an abnormal feature list. The alarm report content includes: the current standard deviation of the incoming line loop is increased by 108% compared with the historical average value, the difference between the current standard deviations of L1 and L2 loops exceeds the normal range by 260%, 5 phase abnormal events are detected and accompanied by 6kW power mutation and the like.

[0056] In hardware implementation, the deviation degree calculation unit and the judgment unit can be deployed in a cloud server, and the distributed computing framework is used to process the judgment tasks of multiple users in parallel. The historical baseline data is stored in a distributed database (such as HBase) to support millisecond-level query response; the current power consumption data is pushed to the judgment module in real time through a message queue (such as Kafka) to ensure that the processing delay is less than 1 minute. The judgment result is pushed to the power inspection system through a WebAPI interface, and the on-site inspection task dispatching process is triggered.

[0057] In order to adapt to the change of the power consumption mode of different users, the system automatically updates the historical baseline data every week, eliminates the long-term deviation characteristics caused by seasonal loads (such as winter electric heaters), and includes the normal operation characteristics of new devices. For example, after the user adds a 2kW dehumidifier in summer, the system learns the start-stop rule and load characteristics of the device within one week, adjusts the baseline value of the related feature parameters, and avoids false judgment caused by normal device changes.

[0058] The whole electricity stealing judgment process follows the principle of “multi-feature fusion, multi-period verification”. Through the dynamic baseline updating and weight adjustment mechanism, the detection sensitivity and anti-interference ability are balanced. At the same time, the system retains all the intermediate data of the deviation degree calculation process, supports manual backtracking analysis, and provides complete evidence chain for the power inspection personnel, including the time sequence curve of abnormal features, parameter comparison report and detailed record of electrical relationship abnormality.

[0059] Embodiment 5: The electricity stealing judgment module includes a deviation degree calculation unit and a judgment unit, which are used to generate a power consumption behavior deviation degree index and judge electricity stealing behavior. The embodiments are described in detail in combination with specific user scenarios.

[0060] Take a commercial user as an example, its electrical equipment includes lighting system (total power 5kW), air conditioning unit (15kW), freezer cabinet (3kW) and office equipment (2kW), normal power mode presents regularity: weekdays 8:00-18:00 is high load period, air conditioning and lighting equipment continuous operation, freezer cabinet start-stop 3-4 times per hour, load power fluctuation range is between 10-25kW; Non-working hours only freezer and security system runs, power is maintained at 3-5kW. The system is based on the user's past 30 days of electricity data to build a set of abnormal electricity detection benchmarks, including the following characteristic parameters: the normal range of standard deviation of line loop current effective value is 8-12A, the difference between the standard deviation of each branch loop is usually less than 10A, the load power mutation amplitude exceeds 6kW, the daily average is not more than 2 times, and the phase mismatch index is generally less than 90.

[0061] On a working day at 9:00, the system analyzes the current power data of the user. The deviation calculation unit first extracts the current value of each characteristic parameter: the standard deviation of the line loop current effective value is 22A, which is significantly higher than the upper limit of the historical benchmark of 12A; The difference between the current standard deviation of the lighting loop (L1) and the air conditioning loop (L2) is 25A, more than twice the normal range; During 9:00-9:10, the load power suddenly dropped from 18kW to 5kW, with a mutation amplitude of 13kW, far exceeding the 6kW threshold in the benchmark; 7 abnormal phase events were detected at the same time, with a phase angle difference of more than 15°, and the phase mismatch index calculated by combining the power change amplitude is 120, higher than the historical maximum value of 90.

[0062] The deviation calculation unit quantifies the deviation degree of each characteristic parameter by comparing the current value with the historical benchmark value. For example, the deviation of the standard deviation of the line loop current is measured by the difference between the current value and the historical average, and the current value 22A is 120% higher than the average 10A, which is judged as highly abnormal; The deviation of the inter-loop deviation parameter is based on the ratio of the difference to the preset threshold, and the difference of 25A exceeds 150% of the threshold of 10A, which is marked as significant abnormality; The load power mutation amplitude of 13kW exceeds 117% of the benchmark threshold of 6kW, which is a strong abnormal feature; The phase mismatch index of 120 is 33% higher than the historical maximum value of 90, which is considered to be medium abnormal. The deviation of each characteristic parameter is calculated by a pre-set weight, in which the phase mismatch index accounts for 30%, the inter-loop deviation parameter accounts for 25%, the load power mutation accounts for 20%, the current standard deviation accounts for 15%, and other parameters (such as voltage kurtosis coefficient) account for 10%.

[0063] Assuming that the deviation of each parameter quantization value is: phase mismatch index deviation 0.8, inter-loop deviation deviation 0.9, load power mutation deviation 1.0, current standard deviation deviation 0.8, weighted sum after generation of power consumption behavior deviation index is 0.8x30%+0.9x25%+1.0x20%+0.8x15%=0.875, which exceeds the preset threshold 0.8. The judgment unit further checks the continuous triggering of the index: in the three consecutive 15-minute detection periods of 8:45, 9:00, 9:15, the deviation index is 0.82, 0.875, 0.85, all of which meet the threshold requirement, and the system determines that there is a suspicion of electricity stealing.

[0064] After abnormality judgment, the system automatically retrieves multi-source data in the abnormal period to integrate the evidence chain: the current waveform data shows that the current of L2 loop (air conditioning loop) drops sharply during 9:00-9:10, and the L3 loop (non-metering loop) appears abnormal current fluctuation; the user terminal power consumption behavior data shows that there is no device start-stop record in this period, but the load power mutation amplitude does not match the device running state; the phase monitoring data shows that the phase angle difference between current and voltage reaches 25°-30° at 9:02, 9:05, etc., far exceeding the normal range of ±15°. These data collectively point to the possible existence of bypass electricity stealing behavior (such as bypassing the metering loop and directly connecting) or intentional destruction of metering devices (such as changing the connection polarity of current transformer).

[0065] The alarm report generated by the system contains the following key information: ① Abnormal period: May 27, 2025 8:45-9:15; ② Abnormal feature list: incoming line loop current standard deviation abnormal rise, multi-loop current difference exceeds the limit, no record of power mutation, phase relationship mismatch; ③ Data evidence: each loop current and voltage waveform file (with timestamp), device start-stop log, phase abnormal event record; ④ Judgment conclusion: the deviation index exceeds the standard for 3 consecutive periods, it is recommended to check the metering device connection and whether there is an unauthorized power consumption loop on site.

[0066] In the hardware implementation layer, the deviation calculation unit and the judgment unit work cooperatively through edge computing terminal and cloud server. The edge computing terminal collects power consumption data in real time and completes preliminary feature extraction, uploads key parameters (such as current standard deviation, phase mismatch index) to the cloud; the cloud server stores the historical baseline data of each user, performs multi-period deviation calculation and trend analysis, dynamically optimizes feature weights using machine learning models (such as random forest), and improves the accuracy of judgment. The communication link uses encrypted transmission protocol (such as TLS1.3) to ensure the security of data in the transmission and storage process.

[0067] To avoid misjudgment, the system sets multiple verification mechanisms: for the first time triggering the deviation threshold, it automatically enters the early warning state instead of directly determining electricity theft, and needs to be continuously monitored in the subsequent period; if the user changes the load mode due to seasonal equipment investment (such as adding new refrigeration equipment in summer), the system updates the historical baseline within 7 days through the adaptive learning module, for example, identifies the start-stop rules and power characteristics of the new equipment, and recalculates the normal range of each parameter. In addition, the artificial review link can intervene in the abnormal judgment process, and the inspectors can compare the current data with the historical curve through the visual interface to exclude false positive alarms caused by equipment failure (such as air conditioner compressor anomaly) or data acquisition error.

[0068] The entire electricity theft judgment process is strictly based on the historical rules of user electricity consumption behavior and the internal logical relationship of electrical parameters, and through cross-validation of multi-dimensional characteristics and continuity analysis of time series, it realizes accurate identification of electricity theft behavior. This module not only can detect traditional physical electricity theft methods (such as short-circuiting current transformers, disconnecting voltage loops), but also can identify new non-intrusive electricity theft behaviors (such as interfering with the sampling signal of the metering device through electronic equipment), providing comprehensive technical support for the anti-electricity theft management of smart grids.

[0069] It should be noted that in this text, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between or among the entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed, or inherent to such a process, method, article, or apparatus.

[0070] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, alternatives, and variations can be made thereto without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An electricity larceny prevention method for electric energy meter multi-source data fusion, characterized in that, The method comprises the following steps: S1: obtaining current waveform data, voltage waveform data and user terminal power consumption behavior data of the electric energy meter, and performing data preprocessing on the current waveform data, voltage waveform data and user terminal power consumption behavior data; S2: analyzing the preprocessed data to generate multi-dimensional power consumption characteristic parameters, performing feature screening to form a feature vector set according to the multi-dimensional power consumption characteristic parameters, and constructing an abnormal power consumption detection benchmark set according to the feature vector set; S3: dynamically verifying the user power consumption data according to the abnormal power consumption detection benchmark set, identifying the phase relationship between the current signal and the voltage signal in the metering loop by using a multi-source data synchronization technology, and quantitatively analyzing the mismatch state of the current signal and the voltage signal; S4: generating a power consumption behavior deviation degree index according to the verification result of the abnormal power consumption detection benchmark set, and judging whether there is electricity stealing behavior according to the power consumption behavior deviation degree index.

2. The electricity larceny prevention method of claim 1, wherein: In S1, a system preset acquisition period is obtained, and multi-loop current waveform data, multi-loop voltage waveform data and user terminal uploaded power consumption behavior data of the electric energy meter in the preset acquisition period are obtained; the current waveform data includes fundamental component and harmonic component; the voltage waveform data includes voltage effective value and instantaneous fluctuation data; the power consumption behavior data includes device start-stop time sequence and load power change data; The data preprocessing on the current waveform data, voltage waveform data and power consumption behavior data includes signal filtering, data alignment, abnormal value elimination and normalization processing; The current waveform data is preprocessed to generate a current feature sequence, and the voltage waveform data is preprocessed to generate a voltage feature sequence; the current feature sequences and the voltage feature sequences of different loops are associated and matched to form a multi-loop electrical feature set. 3.The electricity meter multi-source data fusion anti-theft method of claim 2, wherein: In S2, the following steps are included: S201: extracting statistical characteristic parameters from the preprocessed current feature sequence, voltage feature sequence and power consumption behavior data, respectively, the statistical characteristic parameters including standard deviation, kurtosis coefficient and zero-crossing rate, combining each statistical characteristic parameter according to a preset weight to generate multi-dimensional power consumption characteristic parameters; S202: setting a dynamic threshold for the multi-dimensional power consumption characteristic parameters in the multi-loop electrical feature set, screening the characteristic parameters meeting the threshold condition to form an initial feature vector set, and discarding the characteristic parameters not meeting the threshold condition; S203: performing cross-loop comparative analysis on each characteristic parameter in the initial feature vector set, extracting the difference amount of the characteristic parameters in different loops within the same time window, calculating the absolute value of the difference amount and marking it as a loop deviation parameter; S204: comparing the loop deviation parameter with a preset deviation threshold, screening the characteristic parameters exceeding the deviation threshold to add to the abnormal power consumption detection benchmark set, and supplementing the load power mutation feature to the abnormal power consumption detection benchmark set according to the user terminal power consumption behavior data.

4. The electricity meter multi-source data fusion anti-theft method according to claim 3, characterized in that: In S3, the quantitatively analyzing the mismatch state of the current signal and the voltage signal comprises the following steps: S301: real-time collecting the current instantaneous value and the voltage instantaneous value of the metering loop, calculating the phase angle difference value of the current and the voltage, and determining an abnormal phase event if the phase angle difference value exceeds a preset tolerance range; S302: count the number of abnormal phase events in a preset period of time as , and the load power change amplitude in the user terminal power consumption behavior data as ; S303: According to the number of abnormal phase events With the load power change amplitude The phase mismatch index is calculated, and the specific calculation formula is as follows: ； If the phase mismatch index exceeds the preset alarm threshold, the electricity stealing suspicion mark is triggered.

5. The electricity meter multi-source data fusion anti-stealing electricity method according to claim 4, characterized in that: In S4, the check result of the abnormal power consumption detection reference set in the same user historical power consumption data is extracted, the deviation degree of each detection reference is weighted and summed to generate a power consumption behavior deviation degree index, and if the power consumption behavior deviation degree index exceeds the preset threshold for a continuous number of times, it is determined that there is electricity stealing behavior.

6. An electricity stealing prevention system based on multi-source data fusion of an electric energy meter, the system being applied to the electricity stealing prevention method based on multi-source data fusion of an electric energy meter according to any one of claims 1-5, characterized in that, The system comprises a multi-source data acquisition module, a feature modeling analysis module, an electrical relationship verification module, and an electricity stealing judgment module. The multi-source data acquisition module is configured to acquire current waveform data, voltage waveform data, and user terminal power consumption behavior data of an electric energy meter and perform data preprocessing. The feature modeling analysis module is configured to analyze the preprocessed data to generate multi-dimensional power consumption feature parameters and construct an abnormal power consumption detection reference set. The electrical relationship verification module is configured to dynamically verify user power consumption data according to the abnormal power consumption detection reference set and analyze the mismatch state of the current signal and the voltage signal. The electricity stealing judgment module is configured to generate a power consumption behavior deviation degree index and determine electricity stealing behavior.

7. The electricity larceny prevention system of claim 6, wherein: The multi-source data acquisition module comprises an electrical quantity acquisition unit and a user behavior acquisition unit. The electrical quantity acquisition unit is configured to acquire multi-loop current waveform data and voltage waveform data of an electric energy meter. The user behavior acquisition unit is configured to acquire device start-stop time series and load power change data of a user terminal.

8. The electricity larceny prevention system of claim 6, wherein: The feature modeling analysis module comprises a feature extraction unit and a reference construction unit. The feature extraction unit is configured to extract standard deviation, kurtosis coefficient, and zero-crossing rate feature parameters from the preprocessed data. The reference construction unit is configured to filter feature parameters that meet threshold conditions and supplement load power mutation features to the abnormal power consumption detection reference set.

9. The electricity larceny prevention system with multi-source data fusion of electric energy meter according to claim 6, characterized in that: The electrical relationship verification module comprises a phase monitoring unit and a mismatch calculation unit. The phase monitoring unit is configured to detect the phase angle difference between the current and the voltage and identify abnormal phase events. The mismatch calculation unit is configured to calculate a phase mismatch index in combination with the load power change amplitude.

10. The electricity larceny prevention system with multi-source data fusion of electric energy meter according to claim 6, characterized in that: The electricity stealing judgment module comprises a deviation degree calculation unit and a judgment unit. The deviation degree calculation unit is configured to weight and sum the deviation degrees of the abnormal power consumption detection references. The judgment unit is configured to determine electricity stealing behavior according to the continuous triggering number of the deviation degree index.

Images