Anti-electricity-stealing monitoring method, device and equipment based on intelligent universal meter and medium

Abstract

The invention relates to an anti-electricity-stealing monitoring method and device based on an intelligent universal meter, equipment and a medium. The method comprises the following steps: acquiring real-time electric sampling data of a universal meter, and preprocessing the data to obtain electric characteristic data; carrying out electricity stealing feature extraction on the electricity feature data by utilizing an edge electricity stealing detection model, and carrying out preliminary judgment on an electricity utilization type to obtain a preliminary judgment result; if the preliminary judgment result is at least one of an electricity larceny event and reverse feed, using a cloud electricity larceny detection model to perform secondary analysis on the electricity characteristic data to obtain an abnormal probability; and if the abnormal probability exceeds a preset threshold, generating a corresponding alarm work order, and pushing the alarm work order to a supervision platform. By adopting the method, the field acquisition capability of the intelligent multimeter can be fully combined, the local and cloud computing characteristics are combined, and the recognition and monitoring of electricity stealing behaviors such as low-current electricity stealing and reverse feed are effectively enhanced through a multi-source feature extraction and judgment mechanism.

Description

Technical Field

[0001] The present invention belongs to the field of electric power monitoring, and in particular relates to an anti-electricity theft monitoring method, device, equipment and medium based on an intelligent multimeter. Background Art

[0002] With the development of smart power distribution and digital grid technologies, user-side electricity usage monitoring technologies have emerged. They mainly use electricity metering facilities to analyze abnormal electricity usage to prevent users from stealing electricity.

[0003] In traditional technology, the electricity consumption data collected by smart meters is used to compare historical electricity consumption habits with the same period to analyze whether there are abnormal deviations; or by setting up segmented metering points in the distribution and transformation substation area, the sum of the electricity consumption of the main incoming line and the downstream branch meters is compared to identify the existence of electricity theft in a certain area; and by combining the data collected by the centralized meter reading system with manual audits, an alarm is generated and personnel are dispatched to the site for verification.

[0004] However, these traditional methods, which rely on back-end big data analysis or manual audits, struggle to respond to end-user electricity usage in real time and identify theft periods and evidence. Comparisons based solely on static values ​​like energy and power ignore dynamic behavior, such as voltage and current waveforms, harmonic disturbances, and power factor fluctuations. This makes it difficult to model complex electricity usage. If users employ covert methods like parallel connections or illegally connect to solar inverters, the main meter current remains extremely low, making it difficult to effectively identify theft and capture evidence. Summary of the Invention

[0005] Based on this, it is necessary to provide an anti-electricity theft monitoring method, device, equipment and medium based on a smart multimeter that can integrate edge recognition and cloud analysis, support high-frequency sampling and multi-dimensional feature extraction, in response to the above technical problems.

[0006] In a first aspect, the present application provides an anti-electricity theft monitoring method based on a smart multimeter, comprising:

[0007] Acquire real-time electrical sampling data from the multimeter and pre-process the data to obtain electrical characteristic data; the real-time electrical sampling data includes voltage sampling data and current sampling data; the electrical characteristic data includes current, voltage, load power, instantaneous power, real-time accumulated electric energy, harmonic distribution, and power factor distributed according to the sampling time sequence;

[0008] The edge electricity theft detection model is used to extract electricity theft features from the electricity feature data and make a preliminary judgment on the electricity usage type to obtain a preliminary judgment result; the preliminary judgment result includes normal electricity usage, electricity theft events and reverse power feeding;

[0009] If the preliminary determination result is at least one of the electricity theft event and reverse power feeding, the cloud-based electricity theft detection model is used to perform a secondary analysis of the electrical feature data to obtain the abnormality probability;

[0010] If the abnormal probability exceeds the preset threshold, a corresponding alarm work order will be generated and pushed to the supervision platform.

[0011] In one embodiment, real-time electrical sampling data from a multimeter is acquired and pre-processed to obtain electrical characteristic data, including:

[0012] Perform digital filtering on real-time electrical sampling data to obtain current and voltage;

[0013] Calculate active power and apparent power based on voltage and current to obtain load power and power factor;

[0014] Calculate instantaneous power based on voltage, current and power factor;

[0015] The instantaneous power is accumulated using a numerical integration algorithm to obtain real-time accumulated electric energy;

[0016] The voltage sampling data and the current sampling data are subjected to fast Fourier transform using a sliding time window, and the amplitudes of multiple harmonic orders are extracted to obtain the harmonic distribution.

[0017] In one embodiment, the power factor is obtained by the following formula:

[0018]

[0019] S=U rms I rms

[0020]

[0021] Where PF is the power factor; P is the active power; S is the apparent power; N is the number of sampling times in the sampling period; n is the sampling point; u[n] is the voltage; i[n] is the current; U rms is the effective value of voltage; I rms is the effective value of current;

[0022] The instantaneous power is obtained by the following formula:

[0023] P real [n]=u[n]×i[n]×PF[n]

[0024] Among them, P real [n] is the instantaneous power;

[0025] The real-time accumulated electric energy is obtained through the following formula:

[0026]

[0027] Where E(t) is the real-time accumulated energy and Δt is the sampling period.

[0028] In one embodiment, an edge electricity theft detection model is used to extract electricity theft features from electricity feature data and perform a preliminary determination of the electricity usage type, obtaining a preliminary determination result including:

[0029] Performing time-series electrical behavior feature extraction on the electrical feature data to obtain an electrical feature vector; the time-series electrical behavior feature extraction corresponds to steady-state load detection, low-current electricity theft detection, and reverse current detection;

[0030] Generate electrical feature labels based on the electrical feature vector; the electrical feature labels include normal power output, suspected power theft, and reverse current;

[0031] Based on a sliding window with a preset fixed time length, the electrical feature tags at the sampling moments in multiple consecutive windows are statistically determined to obtain a preliminary determination result;

[0032] Among them, steady-state load detection corresponds to extracting the abnormal deviation characteristics of load power compared with the historical average load power in the same period;

[0033] Low-current electricity theft detection corresponds to the deviation characteristics of the extracted voltage and the jitter harmonics in the harmonic distribution that reflect the actual power consumption status;

[0034] Reverse current detection is corresponding to extracting the abnormal characteristics of the direction and value of the current.

[0035] In one embodiment, if the preliminary determination result is at least one of electricity theft and reverse power feeding, a cloud-based electricity theft detection model is used to perform a secondary analysis of the electrical feature data to obtain an abnormality probability, including:

[0036] The monitoring data of the multimeters on the same circuit during the same period are extracted to obtain the group average curve;

[0037] The multi-dimensional anomaly detection algorithm based on information entropy is used to compare and analyze the electrical characteristic data with the group average curve, and the anomaly probability corresponding to the preliminary judgment result is output.

[0038] In one embodiment, if the abnormal probability exceeds a preset threshold, a corresponding alarm work order is generated and pushed to the supervision platform, including:

[0039] If the abnormal probability corresponding to the electricity theft event exceeds the preset threshold, the reverse power compensation method is used to estimate the amount of electricity theft based on the line loss and the actual load power entering the household.

[0040] If the abnormal probability corresponding to reverse power feeding exceeds the preset threshold, the abnormal power feedback amount is identified through the reverse current amplitude and harmonic distribution of the secondary side small signal of the multimeter current transformer, and the amount of stolen energy is obtained;

[0041] Generate an alarm work order based on the amount of electricity stolen, multimeter number, geographic location, and a snapshot of the evidence of theft;

[0042] Among them, the reverse power compensation method estimates the following steps:

[0043] Calculate the theoretical power consumption of the distribution circuit under normal conditions and obtain the theoretical value;

[0044] Based on the line loss, the deviation between the actual load power entering the home and the theoretical value is accumulated to obtain the amount of stolen electricity.

[0045] In one embodiment, the method further comprises:

[0046] If the preliminary determination result is at least one of the electricity theft event and reverse power feeding, trigger local storage of real-time electricity sampling data within the current preset time interval to obtain a snapshot of the electricity theft evidence;

[0047] Based on GPRS wireless remote communication technology, the electricity theft evidence snapshot, electrical characteristic data, corresponding multimeter number and geographic location data are packaged and sent to the cloud platform.

[0048] In a second aspect, the present application further provides an anti-electricity theft monitoring device based on a smart multimeter, comprising:

[0049] The data acquisition module is used to obtain the real-time electrical sampling data of the multimeter and pre-process the data to obtain electrical characteristic data;

[0050] The local processing module is used to extract the electricity theft features from the electricity feature data using the edge electricity theft detection model, and to make a preliminary determination of the electricity usage type to obtain a preliminary determination result;

[0051] A cloud processing module is configured to perform a secondary analysis of the electrical characteristic data using a cloud-based electricity theft detection model to obtain an abnormality probability if the preliminary determination result is at least one of an electricity theft event and reverse power feeding;

[0052] The supervision module is used to generate a corresponding alarm work order if the abnormal probability exceeds the preset threshold, and push the alarm work order to the supervision platform.

[0053] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of any of the above-mentioned anti-electricity theft monitoring methods based on a smart multimeter.

[0054] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned anti-electricity theft monitoring methods based on a smart multimeter.

[0055] The aforementioned smart multimeter-based anti-electricity theft monitoring method, device, equipment, and medium utilize the smart multimeter as the data sensing foundation and integrate edge model judgment with cloud-based model re-judgment mechanisms to construct a closed-loop electricity usage behavior monitoring method that covers the entire process of data collection, anomaly identification, behavior determination, and automatic alarms. This method accurately determines normal electricity usage and efficiently, in real time, and quantifiably identifies and addresses a variety of new types of electricity theft. By locally collecting and preprocessing high-frequency voltage and current sampled data in real time on the smart multimeter, electrical signature data containing multi-dimensional indicators such as active power, load power, power factor, instantaneous power, and harmonic distribution is generated. Using edge models, initial classification of electricity usage behavior is performed at the on-site terminal, effectively improving event detection response speed and meeting the need for immediate identification of sudden, short-term electricity theft. By uploading electrical signature data to a cloud platform, leveraging enhanced computing power and swarm behavior models, anomaly events are re-judged and anomaly probability assessed, significantly enhancing the stability and accuracy of overall judgment and reducing false alarm and missed detection rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 FIG. 1 is an application environment diagram of the anti-electricity theft monitoring method based on the intelligent multimeter of the present invention;

[0058] Figure 2 Schematic diagram of the flow of the anti-electricity theft monitoring method based on the intelligent multimeter of the present invention;

[0059] Figure 3 Schematic diagram of the step-by-step process of step S202;

[0060] Figure 4 This is a structural diagram of the anti-electricity theft monitoring device based on the intelligent multimeter of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0062] The anti-electricity theft monitoring method based on the smart multimeter provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the local server 100 communicates with the cloud server 200 through GPRS (general packet radio service) wireless remote communication. The data storage system can store the data that the cloud server 200 needs to process. The data storage system can be integrated on the cloud server 200, or it can be placed on the cloud or other network servers. Among them, the local server 100 is connected to the smart multimeter 300, the smart multimeter 300 supports 220V AC power supply, and has a built-in backup lithium battery to ensure continuous operation in the power-off state. The current sampling adopts an open or bolt-fixed transformer to ensure measurement accuracy and long-term stability. The voltage sampling is carried out through high-impedance voltage division sampling, and has overvoltage protection and isolation protection functions. The smart multimeter 300 has a timed self-test mechanism, that is, the current transformer and voltage sensor channel integrity self-test can be automatically started during non-peak hours every day, and remote calibration and local calibration are supported.

[0063] In an exemplary embodiment, Figure 2 As shown in the figure, a method for anti-electricity theft monitoring based on intelligent multimeter is provided. Figure 1 The application environment is described, including:

[0064] S201. Acquire real-time electrical sampling data from a multimeter and preprocess the data to obtain electrical characteristic data; the real-time electrical sampling data includes voltage sampling data and current sampling data; the electrical characteristic data includes current, voltage, load power, instantaneous power, real-time accumulated electric energy, harmonic distribution, and power factor distributed according to the sampling time sequence.

[0065] Schematically, the multimeter integrates high-precision voltage and current sampling modules. The voltage channel utilizes a high-impedance voltage-divider sampling network, combined with common-mode filtering and overvoltage protection circuitry, to stably step down the 220V mains voltage to within the ADC's operating range. The current channel utilizes a split-type current transformer or Hall effect sensor for contactless current sampling, adapting to a variety of current ranges. The multimeter controller utilizes a synchronous sampling mechanism during the sampling process, acquiring alternating electrical signals at a sampling rate of several thousand times per second to ensure timing alignment and waveform integrity.

[0066] Optionally, after obtaining the original analog sampling data, the MCU or SoC digital processing unit inside the multimeter will perform digital conversion on the data.

[0067] Furthermore, the preprocessing process mainly includes signal filtering, normalization, period segmentation, effective value calculation, and feature extraction. Specifically, the filtering stage uses a digital low-pass filter to suppress high-frequency interference, while normalization and windowing operations are used to construct stable periodic signal segments. For example, multiple key electrical parameters are calculated to form electrical characteristic data, including the effective values ​​of current and voltage, which are the basic operating status of the instantaneous load in the electrical circuit; load power, as a reflection of actual working capacity, reflects whether there is active load operation behavior; instantaneous power further reveals the product relationship between voltage and current, and has strong discrimination capabilities under harmonic interference or carrier frequency conversion conditions; real-time accumulated electric energy is obtained by integrating instantaneous power, which can be used as a basis for determining energy consumption in a certain time interval and assist in analyzing abnormal periods; power factor reveals the phase relationship between voltage and current and is an important indicator for determining whether it is a normal resistive load or an inductive load. Harmonic distribution uses fast Fourier transform (FFT) to obtain the high-frequency components of current or voltage, which is particularly useful for detecting power consumption caused by nonlinear loads such as variable-frequency air conditioners and rectifiers.

[0068] S202 , using an edge electricity theft detection model to extract electricity theft features from the electricity feature data, and to perform a preliminary determination of the electricity usage type to obtain a preliminary determination result; the preliminary determination result includes normal electricity usage, electricity theft events, and reverse power feeding.

[0069] Schematically, a locally deployed edge electricity theft detection model—which can be a lightweight machine learning classifier such as a decision tree model, a minimization neural network, or a support vector machine—is invoked to quickly perform a preliminary assessment of current electricity usage. The model inputs electrical feature data and, after feature mapping, generates a set of high-dimensional vectors representing the current electricity usage behavior. Based on the feature distribution and label space of the existing training samples, the model outputs the assessment result for the current window. Specifically, if all extracted feature indicators remain stable, the power factor is normal, and the load power and harmonic characteristics are within the expected range, the model identifies the current behavior as normal electricity usage. If the extracted features show extremely low current values, near-zero load power but frequent harmonic disturbances, or significantly abnormal power factor with short-cycle repetitive features, it may be considered an electricity theft event. Furthermore, if the extracted features show a reversal of current direction (i.e., a change in the polarity of the CT (current transformer) output signal), and there is reverse energy feedback (i.e., negative apparent power), it is flagged as reverse feeding.

[0070] S203: If the preliminary determination result is at least one of the electricity theft event and reverse power feeding, a cloud-based electricity theft detection model is used to perform a secondary analysis on the electricity feature data to obtain an abnormality probability.

[0071] Furthermore, a cloud-based re-judgment mechanism is introduced. That is, when the preliminary judgment result is at least one of the electricity theft incident and reverse feeding, the remote data reporting mechanism will be automatically triggered, and the electrical characteristic data of the current window, the preliminary judgment result, the timestamp, the device number and other contents will be packaged and uploaded to the cloud server. In the cloud platform, a cloud-based electricity theft detection model based on LSTM (long short-term memory neural network), Transformer (network architecture based on attention mechanism) or stacked autoencoder network obtained through large-scale sample training is deployed. It has a high ability to extract time series data patterns and can explore the potential laws in nonlinear electricity consumption behavior. By inputting the historical behavior curves of multiple devices, the cloud-based model can not only review the degree of abnormal behavior of the current device, but also form a horizontal behavior comparison with other multimeters in the same distribution area, construct a dynamic reference curve based on group behavior, and thus improve the accuracy of abnormality judgment.

[0072] S204. If the abnormal probability exceeds the preset threshold, a corresponding alarm work order is generated and pushed to the supervision platform.

[0073] If the anomaly probability exceeds a set threshold, a standardized alarm work order is generated. This work order includes key fields such as the device number, geographic location, preliminary determination type, anomaly time period, power deviation estimate, sampled data snapshot summary, harmonic distribution diagram, and power factor curve. Once generated, the alarm work order is pushed to the supervision platform or the power company's operation and maintenance terminal. Relevant operation and maintenance personnel can access anomaly records, view data curves, and download electrical parameter snapshots on the platform to further decide whether to dispatch personnel for on-site inspection or directly initiate remote power outage protection measures.

[0074] The aforementioned smart multimeter-based anti-electricity theft monitoring method acquires and preprocesses multi-dimensional data locally at the multimeter, including current, voltage, load power, instantaneous power, accumulated energy, harmonic distribution, and power factor. This enables comprehensive end-level power usage behavior, enhancing data representation and behavioral identifiability. By deploying an edge-based electricity theft detection model, a preliminary assessment of electricity usage behavior is made locally at the smart multimeter, providing real-time output indicating suspected theft or reverse power feeding. This edge model is lightweight and computationally efficient, enabling rapid response without relying on continuous communication, effectively compensating for the time delay and data lag associated with centralized analysis solutions. Once the edge model identifies potential electricity theft, the cloud-based electricity theft detection model further performs high-dimensional, multi-period, and multi-user comparative analysis on the electrical signature data. Using a more complex algorithmic model and a group behavior reference mechanism, it eliminates misjudgments due to short-term fluctuations and load disturbances, thereby enhancing the reliability and scientific nature of the final anomaly identification. When the abnormality probability exceeds the set threshold, a standardized alarm work order is automatically generated. This work order provides a structured and traceable processing basis for the operation and maintenance end or the supervision platform, improving the closed-loop efficiency from identification to disposal.

[0075] In one embodiment, real-time electrical sampling data from a multimeter is acquired and pre-processed to obtain electrical characteristic data, including:

[0076] S21. Digitally filter the real-time electrical sampling data to obtain current and voltage.

[0077] Because actual power systems often contain complex factors such as power frequency disturbances, high-frequency interference, and ground noise, directly using unfiltered data can easily introduce fluctuation misjudgments. Therefore, after the raw real-time electrical sampling data undergoes analog-to-digital conversion (ADC), digital filtering is used to denoise the voltage and current sampling sequences. For example, a finite impulse response (FIR) or infinite impulse response (IIR) filter is used to filter out high-frequency components above a certain frequency threshold in the sampled data, while retaining frequency components such as the fundamental and low-order harmonics that are relevant to power consumption behavior.

[0078] S22. Calculate active power and apparent power based on voltage and current to obtain load power and power factor.

[0079] Schematically, based on the filtered voltage and current signals, the effective RMS values ​​are calculated to obtain apparent power, and the average value is calculated by combining the instantaneous product to obtain the active power. The effective RMS value is calculated using the square root mean square method, which squares the sampling points within a certain time window, averages them, and then takes the square root to reflect the actual amplitude of the signal during that time period. Apparent power is represented by the product of the effective voltage and effective current values, reflecting the total flow of electrical energy in the circuit. Active power, on the other hand, is the time average of the point-to-point product of voltage and current, representing the actual work done, or load power. The ratio of these two values ​​is then used to determine the power factor.

[0080] S23. Calculate the instantaneous power based on the voltage, current and power factor.

[0081] Instantaneous power can be regarded as a fine-grained characterization of load state changes, with good sensitivity and timeliness. Especially when the load starts and stops frequently, the current jumps briefly, or the behavior is abnormally disturbed, the instantaneous power change precedes traditional indicators such as RMS and can be used to identify event boundaries in subsequent electricity theft feature extraction.

[0082] S24. Accumulate the instantaneous power using a numerical integration algorithm to obtain real-time accumulated electric energy.

[0083] Real-time accumulation of electrical energy is achieved through a numerical integration algorithm. Since instantaneous power reflects the unit energy flow rate at each point in time, its integral value is the total electrical energy over a specific period of time. Standard numerical integration techniques such as the trapezoidal method and the Simpson method can be used for integration. In embedded systems, this is often implemented as a sliding summation. The accumulated energy results support power consumption comparisons and behavioral pattern recognition, playing a key role in cross-period data modeling or comparing energy consumption during unusual time periods.

[0084] S25. Perform fast Fourier transform on the voltage sampling data and the current sampling data using a sliding time window, and extract multiple order harmonic amplitudes to obtain harmonic distribution.

[0085] For the filtered voltage and current sampling data, a fast Fourier transform (FFT) is performed through a sliding time window to extract the amplitudes of multiple harmonics such as the 3rd, 5th, 7th, and 9th orders to form a harmonic distribution. Specifically, the transformation process is performed in a periodic sliding manner to ensure that the data in each analysis window is complete and periodically stable. The extracted harmonic distribution can be regarded as the spectrum signature of the current load, which is highly sensitive to identifying electromagnetic interference sources, nonlinear load behavior, and illegal wiring that intentionally disrupts metering. For example, if high-amplitude 3rd and 5th harmonic jitters are frequently observed when the current is close to zero, it is very likely that there is an improperly connected frequency converter or power electronic equipment after the meter, indicating potential hidden electricity theft.

[0086] In one embodiment, the power factor is obtained by the following formula:

[0087]

[0088] S=U rms I rms

[0089]

[0090] Where PF is the power factor; P is the active power; S is the apparent power; N is the number of sampling times in the sampling period; n is the sampling point; u[n] is the voltage; i[n] is the current; U rms is the effective value of voltage; I rms is the effective value of current.

[0091] The instantaneous power is obtained by the following formula:

[0092] P real [n]=u[n]×i[n]×PF[n]

[0093] Among them, P real [n] is the instantaneous power.

[0094] The real-time accumulated electric energy is obtained through the following formula:

[0095]

[0096] Where E(t) is the real-time accumulated energy and Δt is the sampling period.

[0097] In one embodiment, Figure 3 As shown, the edge electricity theft detection model is used to extract electricity theft features from the electricity feature data and make a preliminary judgment on the electricity usage type, obtaining preliminary judgment results, including:

[0098] S301 , extracting time-series electrical behavior features from electrical feature data to obtain an electrical feature vector; the time-series electrical behavior feature extraction corresponds to steady-state load detection, low-current electricity theft detection, and reverse current detection.

[0099] Among them, steady-state load detection corresponds to extracting abnormal deviation characteristics of load power compared with the historical average load power during the same period.

[0100] Low-current electricity theft detection corresponds to the deviation characteristics of the extracted voltage and the jitter harmonics in the harmonic distribution that reflect the actual power consumption status.

[0101] Reverse current detection is corresponding to extracting the abnormal characteristics of the direction and value of the current.

[0102] Schematically, the time-series electrical behavior characteristics refer to the dynamic evolution trajectory of users' electricity consumption behavior in terms of current, voltage, power, harmonics and other indicators, which can reflect the key data patterns of load operation rules, state transitions and potential anomalies.

[0103] Among them, steady-state load detection mainly analyzes the power consumption fluctuations of long-term running equipment. Under normal power consumption conditions, users' load power usually has obvious daily or weekly cycles. Especially in residential areas, industrial and commercial users, the differences in power consumption between weekdays and holidays, daytime and nighttime are predictable. Therefore, the load power in the current detection period is processed by sliding average and compared with the average load power in the same period in history. When it is found that the current power value is significantly lower than the historical average, or is in a no-load operation state, while indicators such as voltage and electric energy remain stable, abnormal deviation features are extracted to mark that the load operation in this period is inconsistent with historical behavior, and there is potential suspicion of human intervention or meter bypassing.

[0104] Low-current electricity theft detection focuses on users who bypass the main meter to draw electricity through branch lines, and continue to operate high-power electrical equipment when the current recorded by the main meter is extremely low, close to zero. Such behavior is often accompanied by high-frequency interference and harmonic disturbances. Therefore, the jitter harmonic characteristics under voltage stability conditions are extracted. Jitter harmonics refer to the short-term change amplitude or disturbance frequency of certain orders in the harmonic distribution. Unlike the static harmonic content, jitter reflects the volatility of electric energy and the startup rhythm of the equipment. For example, when it is detected that the current is extremely low but a specific harmonic term continues to jitter in multiple time windows, especially when the frequency related to household variable-frequency equipment is significantly prominent, it is determined to be a low-current abnormality feature and extracted as a judgment vector.

[0105] Reverse current detection identifies situations where illegal grid-connected power generation or reverse power connection occurs. In a normal power supply structure, current should flow from the power source to the load. If a current reversal is detected through CT polarity or negative active power, accompanied by a continuous feedback energy flow and reverse harmonic peaks, reverse power feeding may be present, including the illegal installation of photovoltaic inverters, generators, or non-regular grid-connected systems. Extracted features include current sign changes, current and voltage phase reversals, and abnormal energy accumulation direction and power factor signs, which constitute reverse direction features.

[0106] S302: Generate an electrical feature label based on the electrical feature vector; the electrical feature label includes normal power output, suspected power theft, and reverse current.

[0107] In this example, a rapid classification operation is performed on the electrical feature vector. Specifically, based on the similarity between the feature vector and known category samples, the power usage behavior type label for the current period is output. The labels include normal power output, suspected power theft, and reverse current.

[0108] S303: Based on a sliding window of a preset fixed time length, statistically determine the electrical feature tags at sampling moments in a plurality of consecutive windows to obtain a preliminary determination result.

[0109] Since actual electricity consumption behaviors such as user equipment startup, tripping, or load switching often have short-term fluctuations, making decisions directly based on the judgment results of a single moment is prone to false alarms. Furthermore, a time sliding window mechanism is introduced to enhance the stability of the label. Specifically, a sliding window of a fixed time length is used as the judgment period, and the electrical feature labels of each sampling moment in the window are statistically summarized. For example, if more than 70% of the moments are marked as suspected electricity theft within 30 consecutive seconds, the window is determined to be an abnormal electricity usage window, and the final label is output as the preliminary judgment result of the window. This mechanism effectively avoids misjudgments caused by short-term disturbances, and also ensures that abnormal behaviors can be captured in time when they are continuously identified.

[0110] In one embodiment, if the preliminary determination result is at least one of electricity theft and reverse power feeding, a cloud-based electricity theft detection model is used to perform a secondary analysis of the electrical feature data to obtain an abnormality probability, including:

[0111] S31. Extract the monitoring data of the multimeters of the same circuit during the same period to obtain a group average curve.

[0112] Illustratively, after receiving the preliminary determination result and the electrical signature data for the corresponding time period, monitoring data from other smart multimeters on the same power supply line as the device is extracted for the same time period. The same line can be identified based on topology or substation code. Exemplarily, clustering is performed based on the logical affiliation of all terminal devices downstream of the distribution transformer node, or extraction is performed using the power supply circuit identification field preset in the power operation and maintenance platform. The same time period is consistent with or temporally adjacent to the current suspected power theft period, exemplarily within a range of ±5 to ±15 minutes. Extracted data includes, but is not limited to, the voltage, current, power factor, load power, total harmonic content, order harmonic distribution, and instantaneous power curves for each device. By structuring and time-aligning the electrical signature data from the multimeters on the same line, a group average curve is generated. This curve, numerically represented as a sequence of statistical means or medians across multiple dimensions, represents the typical electricity usage patterns of all users on the distribution line within the same time window. It can be used to construct a load commonality reference model, that is, the common trend of electricity consumption behavior of user groups in the same area under the influence of macro variables such as temperature, weather, and holidays; on the other hand, when the current user's electricity characteristics deviate significantly from this curve, it can directly reflect the degree of abnormality in their electricity consumption behavior, thereby providing a comparison baseline for subsequent abnormality judgment.

[0113] S32. Utilize a multi-dimensional anomaly detection algorithm based on information entropy to compare and analyze the electrical characteristic data with the group average curve, and output an anomaly probability corresponding to the preliminary determination result.

[0114] As an example, the electrical characteristic data uploaded by the current user is compared and analyzed with the group average curve constructed above. To comprehensively assess the degree of deviation in the multidimensional feature space, a multidimensional anomaly detection algorithm based on information entropy is introduced. Specifically, if a user's electrical behavior exhibits high uncertainty in each characteristic dimension or significantly deviates from the group average, then this behavior has high information entropy in the sense of information theory and can be identified as a potential anomaly.

[0115] Information entropy essentially measures the degree of confusion or uncertainty of a variable. Within a multidimensional electrical feature vector, the deviation of each electrical feature data point from the group mean is calculated, and these deviations are then mapped into a discretized probability distribution to construct a probabilistic model of the local features. Furthermore, the entropy value of each feature is calculated using the Shannon entropy formula, and the average entropy level of the sample across the entire feature space is synthesized. Optionally, the algorithm can also incorporate metrics such as joint entropy or mutual information to assess the degree of coupling between different features, thereby identifying complex electricity theft strategies with jointly anomalous behaviors. Ultimately, a numerical anomaly probability indicator is output, which comprehensively reflects the rarity of the current user's electrical feature behavior within the population in the area.

[0116] For example, if a user's current is only 0.1A in the afternoon of a certain day, with a power factor close to 0, but the third and fifth harmonics are significantly higher than the average of users in the same area by more than 10%, and the user's voltage curve is highly stable with no significant fluctuations, it will be judged as a suspected low-current power theft scenario with a high abnormal probability. In another scenario, if the power feedback direction of a device reverses, resulting in a short period of negative power, accompanied by a sharp increase in harmonic reverse disturbances within 5 seconds, and no other device in the same area experiences similar situations, a high abnormal probability will also be output, indicating potential illegal grid connection.

[0117] In one embodiment, if the abnormal probability exceeds a preset threshold, a corresponding alarm work order is generated and pushed to the supervision platform, including:

[0118] S41. If the abnormal probability corresponding to the electricity theft event exceeds a preset threshold, the amount of electricity theft is estimated using the reverse electricity compensation method in combination with the line loss and the actual load power entering the household.

[0119] Among them, the reverse power compensation method estimates the following steps:

[0120] Calculate the theoretical power consumption of the distribution circuit under normal conditions and obtain the theoretical value.

[0121] Based on the line loss, the deviation between the actual load power entering the home and the theoretical value is accumulated to obtain the amount of stolen electricity.

[0122] Schematically, the reverse energy compensation method aims to quantify the amount of electricity a user may steal during an abnormal period. Specifically, based on the user's electricity usage history, device type, and the typical load model for the area, the theoretical amount of electricity the user should consume under normal conditions during the abnormal period is calculated. Optionally, this theoretical value can be constructed by comparing historical data for the same period, the average of neighboring meters, or the average power of a clustered user group. Furthermore, a line loss correction factor compares the difference between the total incoming power of the distribution circuit and the outgoing power measured by the user's meter to eliminate losses caused by normal physical factors such as line aging and conductor voltage drop, thereby avoiding misinterpreting natural line losses as electricity theft. By adjusting the actual measured energy data, the effective deviation between the actual incoming power and the theoretical value is calculated. This deviation is accumulated over a sustained period to determine the amount of electricity stolen corresponding to the suspected theft.

[0123] S42. If the abnormal probability corresponding to the reverse power feeding exceeds a preset threshold, the abnormal electric energy feedback amount is identified through the reverse current amplitude and harmonic distribution of the secondary side small signal of the multimeter current transformer to obtain the stolen amount.

[0124] Schematically, a method for identifying power feedback based on the secondary-side current signal and spectral characteristics of a multimeter's current transformer is used to determine whether illegal grid-connected power generation or inverter reverse power feeding occurs. Reverse power feeding is characterized by a reversal of current direction, meaning that power flows from the user side to the grid. This is common when users unauthorizedly connect to distributed photovoltaic (PV) systems, diesel generators, or other energy storage devices. Specifically, a multimeter's CT sensing can directly capture the weak reverse current signal on the secondary side. If this signal is continuous over multiple sampling periods, rather than a brief pulse, transient jumps or harmonic interference can be ruled out as potential interference factors. Further analysis is performed on the harmonic spectrum characteristics associated with the reverse current, specifically the peak values, rate of change, and phase shift of the low-order odd harmonics and high-order harmonics during this period. Since inverters typically generate stable frequency disturbances during energy conversion, the reverse current amplitude, duration, and harmonic disturbances are combined to calculate the amount of power feedback, resulting in the total suspected energy stolen from illegal grid connection or reverse power feeding.

[0125] S43. Generate an alarm work order based on the amount of electricity stolen, the multimeter number, the geographical location, and the snapshot of the electricity theft evidence.

[0126] In principle, a structured, standardized alarm work order is automatically generated, containing the multimeter number, geographic location, amount of electricity stolen, alarm type, abnormal time period, and a snapshot of evidence. Once generated, the work order is pushed in encrypted format via the GPRS communication network to the power supply company's operation and maintenance platform or power regulatory center, where an alarm is issued via a terminal or mobile app. Based on the work order content, platform personnel can decide whether to conduct an on-site review, issue a remote power outage command, or blacklist the abnormal behavior.

[0127] In one embodiment, the method further comprises:

[0128] 51. If the preliminary determination result is at least one of the electricity theft event and reverse power feeding, trigger local storage of real-time electricity sampling data within the current preset time interval to obtain a snapshot of the electricity theft evidence.

[0129] Indicatively, when the edge model completes a preliminary judgment on the electrical characteristic data, and the result shows any one of the power theft events or reverse feeding, the local snapshot recording mechanism will be triggered immediately. This mechanism is based on preset time parameters, illustratively covering a total of 30 to 60 seconds before and after the event occurs. It automatically calls the circular buffer area to extract and save the original electrical sampling data sequence in the current period, and can restore the state of the power theft or reverse feeding behavior at that time. Optionally, the preservation method can adopt a compression sealing strategy, such as differential encoding and block compression (such as LZ4 algorithm) of the sampling point data to reduce storage occupancy. The sealed data is also bound to the event label, multimeter number, judgment model version and timestamp to ensure the uniqueness and integrity of the evidence chain, which is convenient for subsequent cloud analysis or on-site verification.

[0130] 52. Based on GPRS wireless remote communication technology, the snapshot of electricity theft evidence, electrical characteristic data, corresponding multimeter number and geographic location data are packaged and sent to the cloud platform.

[0131] Schematically, data is packaged and encapsulated via GPRS wireless communication and sent to a cloud-based monitoring platform. Optionally, to ensure transmission security and integrity, an encrypted transmission mechanism is employed at the communication protocol layer to prevent data tampering or theft during public network transmission. Encapsulated data can be structured in lightweight formats such as JSON and CBOR, or in a customized binary message format to suit communication link performance and server decompression efficiency requirements. The high-resolution raw data provided by snapshots of electricity theft evidence is often essential objective evidence for retrials and law enforcement.

[0132] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0133] Based on the same inventive concept, embodiments of the present application further provide an intelligent multimeter-based anti-electricity theft monitoring device for implementing the aforementioned intelligent multimeter-based anti-electricity theft monitoring method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the intelligent multimeter-based anti-electricity theft monitoring device provided below can be found in the aforementioned limitations of the intelligent multimeter-based anti-electricity theft monitoring method and will not be further elaborated here.

[0134] In an exemplary embodiment, Figure 4 As shown, an anti-electricity theft monitoring device based on a smart multimeter is provided, comprising:

[0135] The data acquisition module 401 is used to acquire real-time electrical sampling data from the multimeter and pre-process the data to obtain electrical characteristic data;

[0136] The local processing module 402 is used to extract electricity theft features from the electricity feature data using the edge electricity theft detection model, and perform a preliminary determination of the electricity usage type to obtain a preliminary determination result;

[0137] The cloud processing module 403 is configured to perform a secondary analysis on the electrical characteristic data using a cloud-based electricity theft detection model to obtain an abnormality probability if the preliminary determination result is at least one of electricity theft and reverse power feeding;

[0138] The supervision module 404 is used to generate a corresponding alarm work order if the abnormality probability exceeds a preset threshold, and push the alarm work order to the supervision platform.

[0139] In one embodiment, it further includes:

[0140] A feature extraction module is used to extract the time-series electrical behavior features of the electrical feature data to obtain an electrical feature vector; the time-series electrical behavior feature extraction corresponds to steady-state load detection, low-current electricity theft detection, and reverse current detection;

[0141] A label determination module is used to generate an electrical feature label based on the electrical feature vector; the electrical feature label includes normal power output, suspected power theft, and reverse current;

[0142] The event result module is used to perform statistical judgment on the electrical feature tags at the sampling moments in multiple consecutive windows based on a sliding window with a preset fixed time length to obtain a preliminary judgment result.

[0143] In one embodiment, it further includes:

[0144] The group feature module is used to extract the monitoring data of the multimeters on the same line during the same period and obtain the group average curve;

[0145] The secondary recognition module is used to compare and analyze the electrical characteristic data with the group average curve using a multi-dimensional anomaly detection algorithm based on information entropy, and output the anomaly probability corresponding to the preliminary judgment result.

[0146] In one embodiment, it further includes:

[0147] The electricity theft module is used to estimate the amount of electricity stolen using the reverse electricity compensation method based on line loss and actual household load power if the abnormal probability corresponding to the electricity theft event exceeds a preset threshold;

[0148] The power theft module is also used to identify the abnormal power feedback amount through the reverse current amplitude and harmonic distribution of the secondary side small signal of the multimeter current transformer if the abnormal probability corresponding to the reverse feeding exceeds the preset threshold, and obtain the stolen power amount;

[0149] The alarm work order module is used to generate alarm work orders based on the amount of electricity stolen, multimeter number, geographic location, and snapshots of electricity theft evidence.

[0150] In one embodiment, it further includes:

[0151] An evidence module is configured to trigger local storage of real-time electricity sampling data within a preset time interval to obtain a snapshot of electricity theft evidence if the preliminary determination result is at least one of electricity theft and reverse power feeding;

[0152] The data transmission module is used to encapsulate the snapshot of electricity theft evidence, electrical characteristic data, corresponding multimeter number and geographic location data based on GPRS wireless remote communication technology and send them to the cloud platform.

[0153] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0154] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0155] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0156] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. An anti-electricity theft monitoring method based on a smart multimeter, characterized in that: The method comprises: Acquire real-time electrical sampling data from a multimeter and preprocess the data to obtain electrical characteristic data; the real-time electrical sampling data includes voltage sampling data and current sampling data; the electrical characteristic data includes current, voltage, load power, instantaneous power, real-time accumulated electric energy, harmonic distribution, and power factor distributed according to the sampling time sequence; The edge electricity theft detection model is used to extract electricity theft features from the electricity feature data, and a preliminary determination of the electricity usage type is performed to obtain a preliminary determination result; the preliminary determination result includes normal electricity usage, electricity theft events, and reverse power feeding; If the preliminary determination result is at least one of the electricity theft event and the reverse power feeding, performing a secondary analysis on the electricity feature data using a cloud-based electricity theft detection model to obtain an abnormality probability; If the abnormal probability exceeds a preset threshold, a corresponding alarm work order is generated and pushed to the supervision platform.

2. The method according to claim 1, characterized in that The method of obtaining real-time electrical sampling data from a multimeter and preprocessing the data to obtain electrical characteristic data includes: Performing digital filtering on the real-time electrical sampling data to obtain the current and the voltage; Calculating active power and apparent power according to the voltage and the current to obtain the load power and the power factor; Calculating the instantaneous power according to the voltage, the current and the power factor; Accumulating the instantaneous power using a numerical integration algorithm to obtain the real-time accumulated electric energy; A fast Fourier transform is performed on the voltage sampling data and the current sampling data using a sliding time window, and multiple order harmonic amplitudes are extracted to obtain the harmonic distribution.

3. The method according to claim 2, wherein: The power factor is obtained by the following formula: S=U rms ·AND rms Wherein, PF is the power factor; P is the active power; S is the apparent power; N is the number of sampling times in the sampling period; n is the sampling point; u[n] is the voltage; i[n] is the current; U rms is the effective value of voltage; I rms is the effective value of current; The instantaneous power is obtained by the following formula: P real [n]=u[n]×i[n]×PF[n] Among them, P real [n] is the instantaneous power; The real-time accumulated electric energy is obtained by the following formula: Wherein, E(t) is the real-time accumulated electric energy; Δt is the sampling period.

4. The method according to claim 1, wherein The edge electricity theft detection model is used to extract electricity theft features from the electricity feature data, and a preliminary determination of the electricity usage type is performed to obtain a preliminary determination result, including: Performing time-series electrical behavior feature extraction on the electrical feature data to obtain an electrical feature vector; the time-series electrical behavior feature extraction corresponds to steady-state load detection, low-current electricity theft detection, and reverse current detection; generating an electrical feature tag according to the electrical feature vector; the electrical feature tag including normal power output, suspected power theft, and reverse current; Based on a sliding window of a preset fixed time length, statistically determine the electrical feature tags at sampling moments within a plurality of consecutive windows to obtain a preliminary determination result; Wherein, the steady-state load detection corresponds to extracting abnormal deviation characteristics of the load power compared with the historical average load power during the same period; The low current electricity theft detection corresponds to extracting the deviation characteristics of the voltage and the jitter harmonics in the harmonic distribution that reflect the actual power consumption status; The reverse current detection corresponds to extracting the direction and value abnormality characteristics of the current.

5. The method according to claim 1, wherein If the preliminary determination result is at least one of the electricity theft event and the reverse power feeding, a secondary analysis is performed on the electricity feature data using a cloud-based electricity theft detection model to obtain an abnormality probability, including: The monitoring data of the multimeters on the same circuit during the same period are extracted to obtain the group average curve; The electrical characteristic data is compared and analyzed with the group average curve by using a multi-dimensional anomaly detection algorithm based on information entropy, and the anomaly probability corresponding to the preliminary determination result is output.

6. The method according to claim 1, characterized in that If the abnormal probability exceeds the preset threshold, a corresponding alarm work order is generated and pushed to the supervision platform, including: If the abnormal probability corresponding to the electricity theft event exceeds a preset threshold, the amount of electricity stolen is estimated using a reverse electricity compensation method in combination with the line loss and the actual load power entering the household; If the abnormal probability corresponding to the reverse power feeding exceeds a preset threshold, the abnormal power feedback amount is identified by the reverse current amplitude of the secondary side small signal of the multimeter current transformer and the harmonic distribution to obtain the amount of stolen power; Generate an alarm work order based on the amount of electricity stolen, the multimeter number, the geographic location, and a snapshot of the electricity theft evidence; The reverse power compensation method corresponds to the following steps: Calculate the theoretical power consumption of the distribution circuit under normal conditions and obtain the theoretical value; Based on the line loss, the deviation between the actual load power entering the household and the theoretical value is accumulated to obtain the stolen power amount.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: If the preliminary determination result is at least one of the electricity theft event and the reverse power feeding, triggering local storage of real-time electricity sampling data within a current preset time interval to obtain a snapshot of electricity theft evidence; Based on GPRS wireless remote communication technology, the electricity theft evidence snapshot, the electrical characteristic data, the corresponding multimeter number and geographic location data are packaged and sent to the cloud platform. 8.An anti-electricity theft monitoring device based on a smart multimeter, characterized in that: The device comprises: The data acquisition module is used to obtain the real-time electrical sampling data of the multimeter and pre-process the data to obtain electrical characteristic data; A local processing module is used to extract electricity theft features from the electricity feature data using an edge electricity theft detection model, and to perform a preliminary determination of the electricity usage type to obtain a preliminary determination result; a cloud processing module configured to perform a secondary analysis of the electrical characteristic data using a cloud-based electricity theft detection model to obtain an abnormality probability if the preliminary determination result is at least one of an electricity theft event and reverse power feeding; The supervision module is used to generate a corresponding alarm work order if the abnormal probability exceeds a preset threshold, and push the alarm work order to the supervision platform.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

Images