Electricity stealing behavior detection method and system based on intelligent electric energy meter

Abstract

The invention discloses an electricity stealing behavior detection method and system based on an intelligent electric energy meter, and the method comprises the following steps: constructing a phase locking reference frame at the end of the intelligent electric energy meter, synchronously collecting a voltage phase and a current phase through high-precision time, generating a direction discrimination baseline, setting a zero-power crossing point and a direction overturning threshold value, and carrying out the detection of an electricity stealing behavior. Outputting a direction baseline consistency index as a direction identification reference; according to the invention, through phase locking, high-precision synchronization, neutral line drift analysis, multi-table data alignment and phase disturbance verification, accurate identification and causal confirmation of a current backward flow behavior are realized, a disturbance user is reversely positioned in combination with a disturbance fingerprint, an intervention instruction is output, and an identification-verification-positioning-control closed loop mechanism is constructed. And the anti-electricity-stealing capability and the metering accuracy of the intelligent electric energy meter in a complex power supply environment are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric energy metering, in particular to a method and system for detecting electricity stealing behavior based on a smart electric energy meter. BACKGROUND

[0002] "Electricity stealing behavior detection based on a smart electric energy meter" refers to using the multi-dimensional real-time collection, remote communication and data analysis capabilities of a smart electric energy meter to conduct fine monitoring of user electricity consumption, and identifying potential abnormal behavior by comparing with normal electricity consumption models to discover electricity stealing phenomena. The smart electric energy meter can not only record parameters such as current, voltage, active power, reactive power and power factor, but also generate data such as load curve and electricity fluctuation trajectory. When the system finds that the electricity quantity billing data of a user is significantly different from the actual electric energy characteristics, it can trigger a suspicion of electricity stealing. For example, if the collected current fluctuation is inconsistent with the billing electricity quantity displayed by the electric energy meter, or the load curve has breakpoints, reverse power and other abnormal characteristics, it may indicate that the user has electricity stealing behaviors such as bypassing the electric energy meter, reverse wiring and short circuiting the line. Through this detection method based on a smart electric energy meter, real-time discovery, rapid positioning and automatic early warning of electricity stealing behavior can be achieved, significantly improving the efficiency of anti-electricity stealing of power companies and ensuring the safe and economic operation of the power supply system.

[0003] The prior art has the following disadvantages: In the prior art, three-phase common-line power supply structure is widely used in residential areas and commercial buildings to reduce power distribution costs and simplify construction wiring. However, in this type of power supply structure, multiple users share the three-phase power supply and common neutral line output by the same transformer, and there is strong coupling between users at the electrical level. Under this background, the prior art cannot effectively prevent a new type of electricity stealing behavior that is extremely concealed, i.e., a stealing user intentionally disturbs the electrical balance of the common-line structure through dynamic switching of asymmetric loads, thereby interfering with the electric energy meter of the adjacent legal user.

[0004] Specifically, some electricity stealing users use controllable power electronic devices or specific combination loads to switch the current path between different phase lines at high speed, forming a transient asymmetric load state, and then inducing neutral line potential fluctuation or inter-phase voltage distortion. Although this interference behavior lasts for a very short time (usually in the order of milliseconds), it is enough to cause the electric energy meter of the adjacent user to misjudge the current direction in that period, i.e., the sampling system incorrectly identifies the "current backflow" state. Since the existing electric energy meter relies on the coupling relationship between voltage polarity and current phase for direction identification, in the face of this dynamic nonlinear interference, it is easy to produce abnormal data such as short-time negative power record or power vector reversal.

[0005] The aforementioned phenomena not only mislead data-driven electricity theft detection algorithms, causing legitimate users to be misclassified as abnormal load users, but also weaken the ability to trace genuine electricity theft activities and even interfere with the accuracy of functions used in smart metering systems for reverse identification and reconciliation. In long-term operation, such interference may also induce regional identification confusion, abnormal credit scores, and cumulative economic losses in large-scale deployment environments.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for detecting electricity theft based on smart meters, so as to solve the problems in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting electricity theft based on a smart meter, characterized by comprising the following steps: A phase-locked reference framework is constructed at the smart energy meter. Voltage and current phases are acquired through high-precision time synchronization to generate a direction discrimination baseline. Zero power crossing point and direction reversal threshold are set, and a direction baseline consistency index is output as a direction identification benchmark. Based on the direction discrimination baseline, the neutral line potential drift curve is calculated in real time, the transient characteristics of three-phase imbalance are extracted, the drift increment and phase difference mutation rate are calculated, and the dynamic change results are mapped to the direction discrimination baseline to determine the time window of suspected current backflow. For the suspected time window of current backflow, time alignment is performed on the load curves of multiple adjacent smart energy meters to construct a relative load response difference matrix, determine single-point anomalies or group consistency loss, and output the current backflow confidence level. Based on the suspected time window and the confidence level of current backflow, a small phase perturbation signal is injected into the phase-locked reference frame, the directional response gain is monitored, and the directional reversal feature is extracted after removing natural fluctuations to complete the causal verification. Based on the causal verification results, the current backflow confidence, phase difference mutation rate, direction response gain and direction reversal features are weighted and fused to generate a current backflow evidence score. When the score exceeds the threshold, an alarm is triggered, the direction recognition weight is frozen, and the perturbation feature fingerprint is recorded. Based on disturbance feature fingerprint information, the load sequences of multiple smart energy meters are aligned with short cycles to construct a multi-user response difference matrix, generate load disturbance fingerprints, and locate disturbing users through abnormal coupling features, output intervention control commands, and realize the identification and suppression of current backflow behavior.

[0009] Preferably, the specific steps for using the output direction baseline consistency index as a direction identification benchmark are as follows: A time synchronization sampling device is deployed at the smart energy meter. A clock reference signal is generated by a crystal oscillator with temperature compensation function, and clock alignment is established with the global positioning system timing module, the Beidou satellite synchronization module, or the power station broadcast time code receiving module. After clock alignment is completed, the A / D converter is started at fixed time intervals to synchronously sample the three-phase voltage and the user line current. Each phase is configured with a 12-bit or higher precision sampling channel, the sampling frequency is not less than 10kHz, and a high-resolution timestamp is added to the sampled value. Based on the sampled data, the zero crossover point of the voltage and current waveforms is extracted, the phase difference is calculated and a sliding window filter is applied to generate a continuous phase difference curve. A direction discrimination baseline is established based on the mean, variance, and maximum deviation of the phase difference during a stable period. A zero-power crossing point and a direction reversal threshold are set, and a direction baseline consistency index is output to provide a direction identification benchmark.

[0010] Preferably, the steps for determining the suspected time window for current backflow are as follows: The instantaneous voltage values ​​of the three-phase conductors and the neutral line-to-ground voltage values ​​are obtained under the time reference of the direction discrimination baseline. The data is sampled every 10 milliseconds to form a continuous neutral line potential curve and establish a synchronization relationship with the three-phase voltage data. By comparing the neutral line potential curve cycle by cycle, the sudden increase or decrease segment with a change amplitude exceeding 1.5 volts within one electrical cycle is extracted, and the start time, maximum amplitude and duration are recorded, and compared with the change in the three-phase voltage amplitude. Under the premise of neutral line potential disturbance, the voltage phase and current phase difference are used as a reference to identify the situation where the phase difference change amplitude exceeds 15 degrees and the change direction is consistent within three consecutive cycles. This is determined as a directional disturbance event, and the event parameters are recorded. Multiple directional disturbance events are integrated based on a time interval of less than 500 milliseconds, and extended by two electrical cycles before and after to form a suspected current backflow time window, which is then output as a basis for subsequent analysis.

[0011] Preferably, the current backflow confidence output steps are as follows: Within the suspected current backflow time window, taking the monitored user as the center, multiple smart energy meters connected to the same low-voltage branch are selected as references to extract active power, reactive power, current RMS value and power factor data. The sampling period is 20 milliseconds, and the data are aligned to a unified start and end time period through timestamps. The load data of the monitored electricity meter is compared with the data of each reference electricity meter hourly to construct a multi-dimensional difference matrix, forming a response difference structure with time as the horizontal axis and electricity meter number as the vertical axis. Statistical analysis of the difference matrix is ​​performed hourly to identify single-point anomalies or lack of group consistency, and the duration of the anomaly, the trend of power direction change, and the phase deviation are recorded. The behavioral consistency analysis results are linked with the neutral line potential disturbance amplitude, phase change rate and directional baseline score. The current backflow confidence is generated by weighting the power change amplitude, duration period and deviation degree, and a high confidence judgment is output when the result exceeds the threshold.

[0012] Preferably, the causal verification steps are as follows: When the built-in confidence score of the suspected current backflow time window exceeds the threshold, and the neutral line potential changes abruptly, the voltage and current phase difference jumps, and there is a single-point load anomaly, the voltage reference phase is adjusted by one to three degrees in the internal control chip of the energy meter based on the phase-locked reference frame, and the disturbance duration does not exceed one power frequency cycle. Before and after the disturbance injection, the voltage phase, current phase, phase difference and active power symbol are sampled every ten milliseconds. Data for ten consecutive cycles are recorded to generate a disturbance response time series. The direction discrimination difference, maximum response amplitude, number of response delay cycles and phase difference abrupt change amplitude before and after the disturbance are calculated to construct a direction response change profile. Using the natural fluctuation period without injected disturbance as a reference, the direction discrimination value, phase difference change and power direction trend are compared. When the phase difference change exceeds 15 degrees after the disturbance is triggered and the direction reversal lasts for more than two cycles, it is confirmed that the disturbance and the direction reversal have a causal relationship, and the direction reversal evidence data unit is output.

[0013] Preferably, the perturbation feature fingerprint generation steps are as follows: The confidence score of current backflow, the abrupt change rate of phase difference, the amplitude of directional response gain and the directional reversal characteristics were normalized and fused according to the weighted ratio to obtain the comprehensive current backflow evidence score. When the evidence score is greater than or equal to 8.0, a level 1 anomaly alarm is triggered, and the direction identification output of the target energy meter is frozen during the time window. At the same time, a data packet containing the energy meter address, event timestamp, score value and indicator details is sent to the monitoring platform, and the original voltage and current sampling values, phase trajectory and power symbol sequence are cached for no less than 72 hours. The temporal features, amplitude features, and associated behavioral features recorded in the triggered event are combined with the geographic electrical information in a structured manner to form disturbance feature fingerprint information. The event number and scoring result are attached, saved to the local database and uploaded to the central data platform for subsequent abnormal behavior comparison and repeated trigger identification.

[0014] Preferably, the steps for identifying and suppressing current backflow behavior are as follows: The user disturbed is located by reverse positioning based on abnormal coupling characteristics, and intervention control commands are output. Based on the start and end times, maximum response amplitude time points and phase reversal nodes recorded in the disturbance feature fingerprint, a unified alignment window is extracted. Multiple smart energy meters in the same distribution branch are selected to collect active power, reactive power, current amplitude and voltage phase data, and they are aligned with a granularity of ten milliseconds. The load parameters of each energy meter in the reference period before the disturbance are standardized and calculated to generate a two-dimensional response matrix. Based on standardization, the point-by-point response difference between the target energy meter and each adjacent energy meter is calculated to form a multi-user relative load response difference matrix. The center-diffusion abrupt change consistent with the disturbance feature fingerprint is identified to determine the suspected disturbance source. By combining the power supply topology and phase sequence wiring information, a reverse propagation chain is constructed to identify the disturbance source and compare the flip-over start point, voltage jump point and neutral line potential peak value. After confirming the user who is actively disturbing the system, an identification confirmation signal is sent to the monitoring platform to freeze the direction judgment result of the target electricity meter in the corresponding time period, mark the high-confidence disturbance source, and trigger an intervention control command to increase the data sampling frequency to five milliseconds.

[0015] A smart meter-based electricity theft detection system includes a phase-locked reference module, a neutral line drift analysis module, a multi-meter linkage difference analysis module, a perturbation injection causality verification module, an evidence fusion judgment module, and a perturbation fingerprint tracing and intervention module. The phase-locked reference module constructs a phase-locked reference framework at the smart energy meter. It acquires voltage and current phases through high-precision time synchronization, generates a direction discrimination baseline, sets the zero-power crossing point and direction reversal threshold, and outputs a direction baseline consistency index as a direction identification benchmark. The neutral line drift analysis module calculates the neutral line potential drift curve in real time based on the direction discrimination baseline, extracts the transient characteristics of three-phase imbalance, calculates the drift increment and phase difference abrupt change rate, and maps the dynamic change results to the direction discrimination baseline to determine the time window of suspected current backflow. The multi-meter linkage difference analysis module performs time alignment on the load curves of multiple adjacent smart energy meters for the suspected current backflow time window, constructs a relative load response difference matrix, judges single-point anomalies or group consistency loss, and outputs the current backflow confidence level. The perturbation injection causality verification module injects a micro-amplitude phase perturbation signal into the phase-locked reference frame based on the suspected time window and the current backflow confidence level. It monitors the directional response gain, removes natural fluctuations, extracts the directional reversal feature, and completes the causality verification. The evidence fusion and judgment module, based on the causal verification results, weights and fuses the current backflow confidence, phase difference mutation rate, direction response gain and direction reversal features to generate a current backflow evidence score. When the score exceeds the threshold, an alarm is triggered, the direction recognition weight is frozen, and the perturbation feature fingerprint is recorded. The disturbance fingerprint tracing and intervention module performs short-cycle alignment of the load sequences of multiple smart energy meters based on disturbance feature fingerprint information, constructs a multi-user response difference matrix, generates load disturbance fingerprints, and reverse-locates the disturbing user through abnormal coupling features, outputs intervention control commands, and realizes the identification and suppression of current backflow behavior.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention constructs a phase-locked reference frame to ensure high-precision synchronization of voltage and current phases, providing a stable benchmark for direction identification. Furthermore, by combining neutral line potential drift calculation and phase difference abrupt change rate analysis, it accurately captures abnormal disturbance behavior caused by three-phase imbalance. Through multi-meter load data linkage alignment and difference matrix calculation, it not only improves the spatial resolution of identification but also enhances the accuracy of judging group consistency deficiencies. Supplemented by injecting micro-amplitude phase disturbances to achieve causal verification, it significantly improves the robustness of direction reversal identification. Finally, by using a disturbance fingerprint backpropagation chain to lock the disturbance source user and generate precise intervention commands, it achieves closed-loop control from identification, confirmation to suppression. This method effectively overcomes the limitations of existing smart meters in handling dynamic nonlinear disturbance behavior, improving the anti-theft capability and power supply security of smart metering systems in complex power supply environments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart of a method for detecting electricity theft based on a smart energy meter according to the present invention.

[0019] Figure 2 This is a schematic diagram of a module of an electricity theft detection system based on a smart energy meter according to the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0021] This invention provides a method for detecting electricity theft based on a smart meter, as shown in the figure, comprising the following steps: A phase-locked reference framework is constructed at the smart energy meter. Voltage and current phases are acquired through high-precision time synchronization. A direction discrimination baseline is generated based on the phase data, the zero-power crossing point is located, and a direction reversal threshold is set. The direction baseline consistency index is output to provide a direction identification benchmark.

[0022] To achieve high-precision construction of the current direction determination benchmark in smart energy meters and maintain the stability of direction determination under the influence of three-phase imbalance disturbances, a direction determination reference method with high-precision synchronization of voltage and current phases as its core is proposed. The specific implementation process is as follows: A time synchronization sampling device is deployed at the smart energy meter to achieve nanosecond-level precision synchronous acquisition of grid voltage and current signals. To this end, a set of temperature-compensated crystal oscillators is integrated into the energy meter to generate a local, highly stable clock reference signal, which is then aligned with an external time source. The external time source can be a GPS timing module, a BeiDou satellite synchronization module, or a power station broadcast time code receiver module, and is connected to the smart energy meter's core chip via serial communication or an RS-485 bus. After synchronization is established, the sampling control logic starts the A / D converter at fixed time intervals to synchronously sample the three-phase voltage of the mains power and the current of the user's current line. Each phase is configured with at least one 12-bit or higher precision analog sampling channel, and the sampling frequency is not less than 10kHz to ensure that at least 200 sampling points can be acquired within each 50Hz electrical cycle. During the sampling process, a high-resolution timestamp is appended to each set of voltage and current sample values. This timestamp originates from a unified time reference locked by a local phase-locked loop and is recorded in a buffer register for subsequent processing. This step ensures a one-to-one correspondence between voltage and current signals in time, providing a precise synchronous data foundation for subsequent directional analysis.

[0023] After completing the synchronous acquisition of high-precision voltage and current data, the system proceeds to the phase identification and phase difference calculation stage. First, the acquired voltage waveform undergoes zero-crossing detection. A three-point fitting interpolation algorithm is used to accurately locate the zero-crossing point of the sinusoidal waveform to calculate the phase start point of the voltage signal. Next, the acquired current waveform is processed in the same way to locate its phase start point, and the phase difference between the current and voltage phases is calculated to obtain the phase difference value for the current cycle. To improve the accuracy of the results, the system uses a sliding window method to perform median filtering on the phase difference over several consecutive cycles (e.g., 10 cycles) to filter out high-frequency noise interference. The resulting continuous voltage-current phase difference curve is used to determine whether there is a stable phase lag relationship in the current load behavior. Normally, under normal electrical load conditions, the current phase should lag behind the voltage phase, with a lag angle generally between 20° and 80°. If the current phase precedes the voltage phase, i.e., a negative phase difference, it may indicate signs of reverse power or current backflow. Based on this, the phase difference curve is constructed as a preliminary source for subsequent direction discrimination baseline, and the stability of the phase difference over time is recorded to form the temporal evolution trajectory of direction behavior.

[0024] After obtaining the voltage-current phase difference curve, a direction discrimination baseline is further established, and stability parameters and a reversal discrimination threshold are extracted using statistical analysis methods. To this end, time periods with minimal fluctuations and stable power are selected from users' daily electricity consumption data. Phase difference data within these time periods is extracted, and their average value, variance, and maximum deviation are calculated. Based on this, the center value and normal tolerance range of the direction baseline are determined. For example, if the average phase difference within this stable period is 45° and the standard deviation is 5°, the baseline tolerance range can be set to 35° to 55°. Furthermore, to identify the critical state of power direction reversal, a "zero power crossing point" is constructed when the phase difference is close to 0° (i.e., voltage and current are in phase) or 180° (i.e., voltage and current are out of phase), and direction reversal judgment thresholds are set on both sides of this point. For example, if the phase difference falls within ±10° for more than three consecutive cycles, a direction reversal event can be identified. This threshold does not use a fixed value but is dynamically generated based on historical stable data to ensure that it adapts to the unique electricity consumption behavior patterns of different users, different time periods, and different electricity consumption states. Meanwhile, by continuously updating the baseline parameters and threshold range, the direction discrimination strategy has adaptive adjustment capabilities and will not make misjudgments due to small fluctuations in the power grid or changes in user electricity consumption characteristics.

[0025] Based on the logical structure jointly defined by the phase difference curve, direction discrimination baseline, zero-power crossing point, and flip judgment threshold, a direction baseline consistency index is constructed as an evaluation basis for the stability of direction recognition. In implementation, a sliding window approach is used to compare the current phase difference curve with the historical discrimination baseline point by point, calculating the average offset of the current phase difference curve, the overlap rate with the baseline, and the contact frequency with the threshold boundary. Weights are assigned to these three parameters to generate a comprehensive consistency score. For example, setting the offset weight to 0.5, the overlap rate weight to 0.3, and the contact frequency weight to 0.2, a weighted average is used to obtain the stability score of the current direction judgment. If the score remains high, it indicates high reliability of the current direction judgment; if the score drops significantly, or frequently experiences sudden drops in a short period, it indicates an abnormal state of the system's direction judgment. In this case, the consistency score result is jointly analyzed with the neutral line potential drift curve used to identify interference behavior, in order to identify potential periods of reverse current in advance. At this point, the direction recognition reference framework has been completed, and a stable, real-time updated direction recognition benchmark has been formed inside the electricity meter, providing highly reliable direction information support for the entire electricity theft detection method.

[0026] This step provides a physical benchmark with temporal continuity, phase accuracy, and stability for current direction identification in smart meters, ensuring accurate determination of energy flow direction even under disturbing load changes or asymmetrical power supply conditions. By constructing a high-precision phase-locked reference frame at the meter end, nanosecond-level alignment of voltage and current signals in the sampling time is achieved, enabling the system to maintain a high degree of synchronization of electrical parameters during grid transient fluctuations. Based on this, by analyzing the real-time phase difference between voltage and current, stable intervals are extracted, crossing points are identified, and direction reversal thresholds are set, thereby establishing a direction discrimination baseline. This baseline not only reflects the energy transmission direction characteristics under normal load conditions but also provides a sensitive standard for identifying abnormal reverse power behavior (such as current backflow). Compared to traditional methods that rely on the average power over an entire cycle to determine direction, this step significantly improves response speed and anti-interference capability, and is a key prerequisite for achieving high-precision current direction determination and accurate identification of electricity theft.

[0027] The drift curve of the neutral line potential is calculated in real time based on the direction discrimination baseline. The transient change characteristics of the three-phase imbalance are extracted, the phase difference mutation rate between the potential drift increment and the voltage and current is calculated, and the obtained dynamic change results are mapped to the direction discrimination baseline to identify and extract the time window suspected of current backflow. To identify potential current direction misjudgments caused by human interference during electricity metering, based on the established direction discrimination baseline, and combined with the dynamic characteristics of neutral line potential and phase difference, potential current backflow behavior is located, and high-confidence suspected time windows are extracted. The specific steps are as follows: Based on the established time reference for voltage and current direction discrimination using the electricity meter, the instantaneous voltage values ​​of the three-phase conductors and their potential values ​​relative to the neutral line are acquired, forming a set of electrical parameters for each sampling period. Within the sampling period, by measuring the three-phase voltage and the neutral line-to-ground voltage separately, the relative amplitude differences between the three-phase voltages are correlated point-by-point with the changes in the neutral line voltage. The neutral line potential refers to the actual voltage value of the neutral line relative to the ground. Its theoretical value should be close to zero, but in practical applications, it fluctuates significantly due to factors such as unbalanced three-phase loads and dynamic changes in connected equipment. By sampling every 10 milliseconds, the neutral line potential value is continuously recorded for several consecutive periods. These potential values ​​are arranged in chronological order to form a set of neutral line potential change curves, and a synchronous correspondence is established with the three-phase voltage data under the same time reference. Unlike existing methods that use integer cycle averages for voltage discrimination, this step, using voltage-to-ground change data in milliseconds, can clearly depict the voltage disturbance behavior caused by users switching specific loads or wiring methods in the three-phase load structure.

[0028] Based on the obtained neutral line potential curve, the potential change amplitude of each time period is compared cycle by cycle, and the time points of potential surges and drops are extracted. To eliminate slow trend changes caused by external environment or system noise, only segments with potential changes exceeding 1.5 volts within one electrical cycle are extracted. For each detected segment, its start time, maximum change amplitude, and duration are recorded, and compared with the voltage difference between the three phases in the corresponding time period. By matching point by point, it is determined whether at least one phase voltage experiences a jump in amplitude or polarity of more than 5% before or after the neutral line potential abrupt change. If the above conditions are met, it indicates that the neutral line potential disturbance is likely caused by a momentary imbalance of the three-phase load. Such imbalances often occur when some users deliberately switch high-power single-phase loads to different phase lines, causing distortion of the direction of neighboring households by disrupting voltage symmetry. This step, through refined comparison of potential amplitude and correspondence with three-phase voltage abrupt changes, can not only detect the existence of voltage disturbance behavior but also trace the origin of the disturbance back to its inception.

[0029] Assuming the existence of neutral line potential disturbance, the voltage and current phase differences calculated in the previous steps are used as a reference indicator to identify whether abnormal changes in current direction occur during the disturbance period. Specifically, all phase differences within the neutral line potential abrupt change period are sorted chronologically, and the magnitude of the difference change between adjacent cycles is calculated. If the phase difference change magnitude exceeds 15 degrees for three consecutive cycles, and the phase change direction is consistent (both showing rapid increase or decrease), it can be identified as a directional disturbance. In this process, the phase difference calculation method is consistent with the previous phase locking process, still using zero-crossing point positioning and high-frequency sampling to ensure the time and amplitude accuracy of the results. Simultaneously, each neutral line potential change and phase abrupt change matching event is assigned a number, and each event unit is appended with numerical parameters such as time length, maximum voltage difference, maximum potential fluctuation amplitude, and total phase change angle to form a complete disturbance event description.

[0030] By combining the occurrence characteristics of each disturbance event with its position in the direction discrimination baseline, these events are integrated into one or more suspected current backflow time windows. During integration, the time interval between two adjacent disturbance events is used as the criterion; if the interval between two events is less than 500 milliseconds, they are considered the same disturbance process, and the time window is extended to cover the entire time period between the two events. After the time window is determined, it is further extended forward and backward by two electrical cycles to ensure that the system response phase before and after the disturbance is included, ultimately forming a high-confidence suspected backflow time interval. Each time interval contains complete neutral line potential data, voltage change data, and current direction change data, which can serve as a reference for subsequent load curve linkage analysis and disturbance behavior confirmation.

[0031] This step aims to identify abnormal electrical changes caused by three-phase load imbalance or malicious disturbances through real-time calculation and analysis of the neutral line potential. This allows for the precise extraction of potential time windows within the electricity meter that may indicate misjudgment of current direction. The neutral line potential should remain stable in a three-phase co-line power supply structure. However, when a user intentionally switches to an unbalanced load or connects a high-frequency disturbance source, it causes a sudden change in the neutral line-to-ground potential, simultaneously interfering with the phase relationship between voltage and current. This can lead to the smart meter misjudging the behavior as reverse power or current backflow. This step, by continuously monitoring the amplitude of neutral line potential changes and combining this with the abrupt changes in three-phase voltage and phase difference, can accurately pinpoint the start and end times of electrical disturbances at the millisecond level. This time window is then mapped to an established direction discrimination baseline, ultimately extracting a high-confidence potential time window for current backflow. This time window provides a crucial time anchor for subsequent multi-meter load data alignment and disturbance behavior identification, making it a core component of the entire anti-theft detection process, enabling "proactive identification and early warning."

[0032] Based on the suspected current backflow time window, the load curves of multiple adjacent smart energy meters are aligned by multi-meter linkage time, a relative load response difference matrix is ​​constructed, and it is determined whether there is a single meter abnormality or a lack of consistency in the group load. The confidence level of current backflow behavior is calculated and output.

[0033] To further verify whether there are current direction disturbances in the target user's current within the suspected time window, and to accurately distinguish them from natural load fluctuations or group responses, this step, based on time-consistent load data, performs cycle-by-cycle alignment and difference analysis on the curve responses of multiple adjacent smart meters to quantify the spatial uniqueness and directional anomaly of the disturbance behavior. Specifically, the implementation steps include the following: Within the suspected current backflow time window, all smart energy meters connected to the same low-voltage branch or terminal block as the monitored user are selected as reference meters. For each energy meter, load data including active power, reactive power, RMS current, and power factor are extracted, ensuring that all data use the same sampling period and timestamp. In this implementation, a sampling period of 20 milliseconds is recommended to cover the time resolution required for typical disturbances in the power system. To achieve time consistency, all meter data must be aligned to the same start and end time period according to the timestamp, covering the complete time interval from the start of the suspected time window to the end of the next power frequency cycle. At each time point, each energy meter should have a complete set of load data so that subsequent hourly analysis and feature matching operations can be performed on the basis of complete data.

[0034] Load data from all smart meters that have completed time alignment are expanded under the same time base to construct a response difference structure with the monitored user as the comparison reference. At each sampling moment, the active power value of the monitored meter is subtracted from the active power value of each reference meter at the same moment, resulting in multiple sets of one-to-one active power differences. Simultaneously, differences in reactive power, RMS current, and power factor are calculated and recorded using the same structure. The difference set at each moment reflects the dynamic offset between the monitored meter's load response on various physical parameters and that of neighboring users. The difference results from all sampling moments are arranged by time, forming a multi-dimensional difference matrix with time as the horizontal axis and meter number as the vertical axis. Each value in this matrix represents the response difference of the monitored meter relative to a neighboring meter on a specific parameter at a specific moment, serving as the fundamental data characterizing the spatial features of abnormal disturbance behavior. Compared to the offline evaluation of meter response trends in traditional methods, this step captures the instantaneous response of disturbance behavior within each cycle and represents the intensity of behavioral anomalies using multi-dimensional differences.

[0035] Based on the constructed difference matrix, hourly difference analysis and behavioral consistency determination are performed. At each sampling moment, the number of all reference energy meters and the monitored energy meter with differences exceeding a preset threshold is counted. Taking active power as an example, if at a certain moment, the difference between more than 70% of the reference meters and the monitored energy meter is less than 10% of the average power of the monitored meter, then that moment is considered "response consistent"; conversely, if most reference energy meters have no load change, but only the monitored meter shows a large fluctuation, then that moment is considered "single-point anomaly". The total number of "single-point anomaly" moments, the longest consecutive period, and the power direction change trend corresponding to these anomalies are counted, i.e., the total number of cycles from positive to negative or from negative to positive. If the above indicators exceed the empirical preset, such as the existence of abnormal direction changes for more than three consecutive power frequency cycles without synchronous response from reference meters, it is preliminarily determined that the monitored user has disturbance characteristics during that time period. In this step, the differences in real-time response behavior of various energy meters under the same electrical physical environment are directly analyzed without relying on modeling and historical comparison. The uniqueness and intensity of load anomalies are determined from the perspective of physical measurement, which has higher judgment accuracy and anti-interference ability than existing methods.

[0036] By linking the behavioral consistency analysis results with parameters such as the amplitude of previous neutral line potential disturbances, phase change rate, and direction discrimination baseline score, a confidence index for the current backflow behavior of the target user within the suspected window is calculated. The confidence level is calculated using a segmented cumulative statistical method. For each "single-point anomaly" event, a level is assigned based on four indicators: power change amplitude, duration period, deviation from the direction baseline, and neutral line potential fluctuation intensity. For example, if the power change amplitude exceeds 30% of the normal load, it is assigned a level 3; if the duration period exceeds 5, it is assigned a level 2, and so on. Finally, all indicators are weighted to generate a backflow suspicion score for that time period. When the score is greater than a preset trigger threshold (e.g., 8 points), a high-confidence current backflow determination is output, and this result is submitted to the next stage of direction causality verification and feature identification.

[0037] This step aims to identify whether a target user exhibits significantly different load fluctuation characteristics compared to neighboring users by comparing the load response behavior of multiple smart meters within a suspected current backflow time window under a unified time benchmark. This allows for the determination of whether the target user is the active initiator of the electrical disturbance. Due to the physical coupling between users in a three-phase co-line power supply structure, relying solely on data from a single meter is insufficient to accurately determine the attribution of abnormal current direction. This step synchronizes the load data of the target user with multiple adjacent users within the same power supply branch at millisecond-level time precision, constructing a relative response difference matrix. This analysis examines whether a "single-point anomaly" occurs within the suspected time window, where only the target user experiences a power surge while other users remain stable, or whether a structural feature of "lack of group response consistency" exists. If such behavior occurs continuously, exhibiting a trend of directional abrupt changes and a consistent duration, the target user can be identified as exhibiting significant disturbance characteristics. This step ultimately outputs the confidence level of the current backflow behavior, providing crucial quantitative evidence for subsequent determinations of whether the disturbance constitutes malicious behavior, whether intervention is necessary, or whether causal verification is required.

[0038] Based on the suspected time window of current backflow and the confidence level of current backflow, a micro-amplitude phase perturbation signal is injected into the phase-locked reference frame to monitor the amplitude change of the direction indication response, construct the direction response gain profile, eliminate natural fluctuations through the response differential signal, extract the direction reversal feature, and form the basis for causal verification.

[0039] To verify whether the abnormal directional behavior of the monitored user within the suspected current backflow time window exhibits characteristics of human control, based on the established directional discrimination baseline and phase-locked structure, a micro-amplitude phase perturbation injection method is adopted. By monitoring the dynamic response process of the directional judgment results, the causal relationship between the perturbation and directional reversal is identified. Specifically, the following steps are included: Based on the obtained suspected current backflow time window and corresponding confidence score, it is determined whether the time period meets the triggering conditions for activating micro-amplitude disturbance verification. When the confidence score of a certain time window exceeds the preset confidence threshold (e.g., a confidence score of 8 or above out of 10), and a sudden change in neutral line potential, a jump in voltage and current phase difference, and abnormal single-point load response behavior of the target user occur within the corresponding time period, it is determined that the time period has sufficient disturbance characteristics and enters the micro-amplitude phase disturbance excitation process. Before executing this process, the original direction discrimination results, voltage and current waveform data, power direction change trend, and relevant timestamp information within the time window are uniformly cached for comparison before and after the intervention. When the intervention action is officially activated, based on the phase locking structure inside the smart energy meter, the voltage reference phase within the current cycle is adjusted by a very small amplitude through the internal control chip. The adjustment amplitude is controlled between 1 and 3 degrees of the original phase of the voltage signal, and the disturbance time does not exceed one complete power frequency cycle (20 milliseconds) to ensure that the normal energy metering results are not affected. This disturbance is constructed in a way that simulates phase shift and does not affect the user's electrical load; it only changes the internal direction judgment reference of the electricity meter.

[0040] After injecting a micro-amplitude disturbance signal, the system enters the dynamic response monitoring phase of the direction determination result. The system continuously samples and records the direction determination results for a total of 10 cycles before and after the disturbance, in 10-millisecond increments. Within each sampling cycle, the system records the voltage phase, current phase, phase difference, active power value, and their positive and negative directional signs. Through continuous recording, a complete disturbance response time series curve is generated, which accurately shows the entire process from the steady-state direction before the disturbance signal injection to the change in direction response during the disturbance. If, within 1 to 2 cycles after the start of the disturbance signal, the system's direction determination result changes from positive to negative, or the previously stable phase difference suddenly becomes less than 10 degrees or even undergoes a jump to the opposite polarity, it indicates that the direction recognition mechanism has a significant response to the injected micro-disturbance signal. Based on this, the system further calculates the difference in direction determination values ​​before and after the disturbance, the maximum amplitude of the direction change, the number of response delay cycles, and the amplitude of the phase difference abrupt change, extracting these values ​​in a structured form to construct a direction response change profile. This profile is used to describe the sensitivity and response strength of the direction determination mechanism of the electricity meter to specific time disturbances, and is a key quantitative basis for determining whether the direction reversal is triggered by external factors.

[0041] After obtaining the directional response change profile, a natural fluctuation removal operation needs to be performed to ensure the causal authenticity of the identification results. To this end, multiple natural fluctuation time periods with similar load states are selected as control samples within the time period without injected disturbances. In each control time period, the directional discrimination value, phase difference change, and power direction trend are recorded for horizontal comparison with the disturbance response profile. If no directional abrupt change trajectory similar to the disturbance response profile appears in the natural fluctuation sample, and the phase difference change under natural fluctuations remains within the set tolerance range (e.g., ±5 degrees), but after the disturbance is triggered, the phase difference changes drastically by more than 15 degrees, the power direction flips, and this continues for more than two cycles, then the directional change can be considered to have significant causal response. Furthermore, the disturbance injection time, the maximum amplitude of the directional response change, the response start time, the response delay time, the phase abrupt change angle, and the power direction flip indicator are collectively considered as components of the causal characteristics of the directional flip in that time period, forming a complete directional flip evidence data unit. This data unit will be used in the subsequent multi-dimensional feature fusion stage to construct the current backflow behavior evidence scoring result along with other indicators, for the final identification and judgment.

[0042] This step, by introducing a small perturbation into the original orientation recognition framework and observing the system response characteristics, enables direct verification of the "triggerable orientation reversal" behavior. It effectively solves the problem of identification in existing technologies that rely on trend judgment and statistical reasoning and are difficult to establish a causal chain, and provides technical support for moving from statistical anomaly identification to physical response confirmation.

[0043] Based on the causal verification results, the backflow confidence, phase difference mutation rate, direction response gain and direction reversal features are weighted and fused to generate a backflow evidence score. If the evidence score exceeds the set threshold, an abnormal alarm is triggered, the direction recognition weight is frozen, and the fingerprint information of the disturbance features is recorded.

[0044] To quantitatively evaluate the multidimensional anomaly features obtained from the causal verification of direction reversal, and to trigger judgment and response actions accordingly, a unified current backflow evidence scoring mechanism is constructed based on the fusion of multi-indicator data. This mechanism then performs alarm triggering, identification freezing, and disturbance feature archiving operations, specifically including the following steps: After completing the micro-perturbation injection and directional response observation, several indicator parameters obtained in the previous steps were invoked, including the current backflow confidence score, phase difference mutation rate, directional response gain amplitude, and directional reversal state characteristics. These indicators all originated from the previous stage's joint analysis of target user load behavior, voltage-current phase relationship, neutral line potential change, and directional judgment output within the suspected time window. Each indicator corresponds to a specific value, timestamp, and original sampled data. To ensure comparability between indicators, the four parameters were first normalized, mapping each indicator to a scoring range of 0 to 10. For example, the current backflow confidence score, originally expressed as a percentage, was divided by 10 to convert it into a score value; the directional response gain was calculated based on the difference in directional judgment output before and after the perturbation, also converted to a 0-10 score; and the phase difference mutation rate was based on the maximum mutation angle change value. After normalization, a weighted average is assigned to each indicator. The weights are set based on their contribution to the anomaly detection results in actual measurements. For example, the confidence level for current backflow is set to 0.4, the directional response gain and directional reversal feature are set to 0.25, and the phase abrupt change rate is set to 0.1. A comprehensive score, namely the current backflow evidence score, is obtained through a weighted summation calculation. This score represents the overall confidence index of the target user's perturbation behavior within the specified time window.

[0045] After obtaining the current backflow evidence score, the system determines whether to trigger an abnormal event alarm and whether to freeze the identification results based on the score. To ensure sufficient discriminativeness of the response actions, two levels of response thresholds are set. For example, a first-level abnormal alarm is triggered when the score is greater than or equal to 8.0; a second-level observation alarm is triggered when the score is between 6.0 and 8.0; and no response is taken when the score is below 6.0. After triggering the first-level alarm, the following response actions are executed: First, the direction identification output of the target energy meter within the current time window is frozen, meaning that the direction identification results for this time period will no longer participate in the training and correction calculation of the subsequent direction discrimination model, thereby preventing disturbance data from polluting the overall identification accuracy; Second, an alarm signal is sent to the monitoring platform, along with a complete data packet consisting of the energy meter address, event timestamp, score value, and detailed indicator items; Third, all original data of this alarm event, including voltage and current sampling values, phase change trajectories, power symbol sequences, and corresponding disturbance injection times, are cached for a period of no less than 72 hours for subsequent data tracing and manual review. This method of linking freezing and alarm avoids the cumulative misleading effect of disturbances on the long-term direction of the system, while ensuring that abnormal events can be promptly known and recorded externally.

[0046] To support subsequent targeted identification and recurring behavior detection, key feature parameters involved in the current disturbance behavior are structured to construct disturbance feature fingerprint information. Specifically, this includes: First, temporal features, including the start time, end time, maximum response delay period, and duration period of the anomaly; second, amplitude features, including the maximum value of the directional response gain, the directional reversal angle range, and the duration of continuous current direction reversal; third, related behavior features, including whether the neutral line potential fluctuation amplitude exceeds 2 volts, whether the phase change angle exceeds 20 degrees, whether there are linked load anomalies between adjacent energy meters, and whether the disturbance is triggered by a single user; and fourth, geographic electrical information, including the transformer number connected to the energy meter, the distribution branch code, and the wiring sequence number in the same row. All of these features constitute a unique fingerprint set, appended with the current event number and scoring result, stored in a local database, and synchronously uploaded to the central data platform via remote communication. This feature fingerprint will be compared with historical fingerprint data when new suspected disturbance events are encountered to determine whether the current disturbance behavior is repeatedly triggered by the same user or whether similar variations are being reproduced.

[0047] This step achieves a closed-loop process for identifying, quantifying, and archiving current backflow disturbance behavior, ensuring the traceability, learnability, and self-enhancing capabilities of the identification system. It is an important technical support for the present invention to achieve accurate identification and response control of complex interference-type electricity theft behavior.

[0048] Based on the fingerprint information of disturbance features, the load sequences of multiple smart energy meters are aligned in a short period of multi-meter linkage to construct a multi-user relative load response difference matrix, generate load disturbance fingerprints, and locate active disturbance users by combining the back propagation chain of abnormal coupling features, and output intervention control commands to achieve accurate identification and suppression of current backflow behavior.

[0049] To achieve the final identification and control of current backflow behavior, based on the disturbance feature fingerprint information generated in the previous steps, a short-cycle linkage analysis is performed on the load response of multiple smart meters during the disturbance period. By constructing a load difference matrix among multiple users, the abnormal response propagation path is identified, and the active disturbance source user is determined by combining the back propagation chain. Precise intervention commands are then output, specifically including the following steps: After extracting the disturbance feature fingerprint, based on the start and end times of the disturbance, the time point of maximum response amplitude, and the phase reversal node recorded in the feature, the corresponding time period is extracted as a unified alignment window for linkage analysis. This window needs to cover the complete main cycle of the disturbance and extend forward by at least one complete power frequency cycle to establish a steady-state comparison before the disturbance; it should also extend backward by two cycles to observe the aftershock propagation characteristics of the disturbance. Based on this, all smart meters located on the same distribution branch as the target energy meter are selected as participants. Active power, reactive power, current amplitude, and voltage phase data of each energy meter are collected within the aforementioned time window, and these are constructed into a time series array with a sampling granularity of 10 milliseconds. Through timestamp alignment and sampling period normalization, strict synchronization of all energy meter data is ensured, providing consistent basic data for subsequent difference matrix calculation.

[0050] The raw load response data of all electricity meters are standardized. For each meter, the mean and standard deviation of its load parameters are extracted during the reference period before the disturbance. Based on this, the data for subsequent time periods are standardized, transforming the instantaneous load value of each meter into its variation relative to its steady state. This process effectively eliminates baseline offsets caused by different connected equipment types, power phase configurations, or user load habits, making the response curves of all meters comparable. Furthermore, the standardized data is arranged into a two-dimensional response matrix according to meter number and time series, allowing for the calculation of the response difference between any two meters at the same time.

[0051] Based on standardized and aligned multi-meter load data, the response difference between the target meter and each of the other meters at each sampling point is calculated, and a multi-user relative load response difference matrix is ​​constructed according to the time axis. In this matrix, each element represents the standardized response difference between the target meter and another meter at a certain time point. Subsequently, the difference matrix is ​​scanned to identify whether there are significant abrupt change regions before and after the disturbance trigger point, paying particular attention to whether there is a rapid and continuous response shift between a certain meter and most other meters, and whether the shift amplitude is consistent with the flip characteristics in the disturbance feature fingerprint. If the difference abrupt change exhibits a "center-diffusion" spatial distribution, that is, it expands step by step from a specific meter to multiple adjacent meters along the time axis, the meter can be preliminarily identified as a possible active disturbance source.

[0052] Based on the identification results of the difference matrix, and combined with the electrical connection structure and phase sequence wiring information, a reverse propagation path chain on the physical topology is constructed. Specifically, firstly, according to the power supply transformer branch line topology diagram, all participating energy meters are sorted according to their physical connection order; then, combining the starting time of each energy meter's response abrupt change in the difference matrix, the temporal sequence is analyzed to determine whether it conforms to the physical law of current propagation from a particular energy meter to its downstream electrical users. If this sequence is valid, and the delay time is within two sampling periods, the disturbance behavior is confirmed as an electrical-level propagation behavior rather than a data error or occasional disturbance. Finally, this propagation chain is compared with the flip-over starting point, voltage jump point, and neutral line potential drift peak in the disturbance feature fingerprint. If they all originate from the same energy meter and the propagation path direction is consistent, the energy meter can be determined as the active initiator of this disturbance behavior.

[0053] Upon successfully locating the user actively causing the disturbance, a targeted intervention and control instruction is immediately generated. This instruction includes the following execution content: First, send an identification confirmation signal to the monitoring platform, along with the disturbance feature fingerprint number, the target electricity meter address code, the start and end times of the disturbance, and the evidence score.

[0054] Second, at the target electricity meter, a direction judgment weight freeze operation is performed, which means that all direction judgment results within this time period are excluded from the reference range of subsequent model iterations.

[0055] Third, mark the user's disturbance event information as a high-confidence disturbance source and set an automatic reporting strategy when similar disturbance behavior is triggered again within three months.

[0056] Fourth, trigger the command to increase the local data sampling frequency, temporarily setting the sampling period of the target user's electricity meter to 5 milliseconds, so as to capture more granular load response changes in its subsequent behavior.

[0057] Through the above operations, a closed-loop process from disturbance identification to user locking and then to response control was realized. This not only significantly improved the accuracy of judging complex electrical disturbance behavior, but also established an identification and protection mechanism with dynamic self-learning capabilities, providing key technical support for electricity theft prevention and control in the smart grid environment.

[0058] This invention constructs a phase-locked reference frame to ensure high-precision synchronization of voltage and current phases, providing a stable benchmark for direction identification. Furthermore, by combining neutral line potential drift calculation and phase difference abrupt change rate analysis, it accurately captures abnormal disturbance behavior caused by three-phase imbalance. Through multi-meter load data linkage alignment and difference matrix calculation, it not only improves the spatial resolution of identification but also enhances the accuracy of judging group consistency deficiencies. Supplemented by injecting micro-amplitude phase disturbances to achieve causal verification, it significantly improves the robustness of direction reversal identification. Finally, by using a disturbance fingerprint backpropagation chain to lock the disturbance source user and generate precise intervention commands, it achieves closed-loop control from identification, confirmation to suppression. This method effectively overcomes the limitations of existing smart meters in handling dynamic nonlinear disturbance behavior, improving the anti-theft capability and power supply security of smart metering systems in complex power supply environments.

[0059] This invention provides, for example Figure 2 The system shown is a smart meter-based electricity theft detection system, comprising a phase-locked reference module, a neutral line drift analysis module, a multi-meter linkage difference analysis module, a perturbation injection causality verification module, an evidence fusion judgment module, and a perturbation fingerprint tracing and intervention module. The phase-locked reference module constructs a phase-locked reference framework at the smart energy meter. It acquires voltage and current phases through high-precision time synchronization, generates a direction discrimination baseline, sets the zero-power crossing point and direction reversal threshold, and outputs a direction baseline consistency index as a direction identification benchmark.

[0060] The neutral line drift analysis module calculates the neutral line potential drift curve in real time based on the direction discrimination baseline, extracts the transient characteristics of three-phase imbalance, calculates the drift increment and phase difference abrupt change rate, and maps the dynamic change results to the direction discrimination baseline to determine the time window of suspected current backflow.

[0061] The multi-meter linkage difference analysis module performs time alignment on the load curves of multiple adjacent smart energy meters for the suspected current backflow time window, constructs a relative load response difference matrix, judges single-point anomalies or group consistency loss, and outputs the current backflow confidence level.

[0062] The perturbation injection causality verification module injects a small-amplitude phase perturbation signal into the phase-locked reference frame based on the suspected time window and the current backflow confidence level. It monitors the directional response gain, removes natural fluctuations, extracts the directional reversal feature, and completes the causality verification.

[0063] The evidence fusion judgment module, based on the causal verification results, weights and fuses the current backflow confidence, phase difference mutation rate, direction response gain and direction reversal features to generate a current backflow evidence score. When the score exceeds the threshold, an alarm is triggered, the direction recognition weight is frozen, and the perturbation feature fingerprint is recorded.

[0064] The disturbance fingerprint tracing and intervention module performs short-cycle alignment of the load sequences of multiple smart energy meters based on disturbance feature fingerprint information, constructs a multi-user response difference matrix, generates load disturbance fingerprints, and reverse-locates the disturbing user through abnormal coupling features, outputs intervention control commands, and realizes the identification and suppression of current backflow behavior.

[0065] The present invention provides a method for detecting electricity theft based on a smart meter, which is implemented through the above-mentioned electricity theft detection system based on a smart meter. For details of the specific method and process of the electricity theft detection system based on a smart meter, please refer to the above-mentioned embodiment of the method for detecting electricity theft based on a smart meter, which will not be repeated here.

[0066] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for detecting electricity theft based on a smart meter, characterized in that, Includes the following steps: A phase-locked reference framework is constructed at the smart energy meter. Voltage and current phases are acquired through high-precision time synchronization to generate a direction discrimination baseline. Zero power crossing point and direction reversal threshold are set, and a direction baseline consistency index is output as a direction identification benchmark. Based on the direction discrimination baseline, the neutral line potential drift curve is calculated in real time, the transient characteristics of three-phase imbalance are extracted, the drift increment and phase difference mutation rate are calculated, and the dynamic change results are mapped to the direction discrimination baseline to determine the time window of suspected current backflow. For the suspected time window of current backflow, time alignment is performed on the load curves of multiple adjacent smart energy meters to construct a relative load response difference matrix, determine single-point anomalies or group consistency loss, and output the current backflow confidence level. Based on the suspected time window and the confidence level of current backflow, a small phase perturbation signal is injected into the phase-locked reference frame, the directional response gain is monitored, and the directional reversal feature is extracted after removing natural fluctuations to complete the causal verification. Based on the causal verification results, the current backflow confidence, phase difference mutation rate, direction response gain and direction reversal features are weighted and fused to generate a current backflow evidence score. When the score exceeds the threshold, an alarm is triggered, the direction recognition weight is frozen, and the perturbation feature fingerprint is recorded. Based on disturbance feature fingerprint information, the load sequences of multiple smart energy meters are aligned with short cycles to construct a multi-user response difference matrix, generate load disturbance fingerprints, and locate disturbing users through abnormal coupling features, output intervention control commands, and realize the identification and suppression of current backflow behavior.

2. The method for detecting electricity theft based on a smart meter according to claim 1, characterized in that, The specific steps for outputting the directional baseline consistency index as a directional identification benchmark are as follows: A time synchronization sampling device is deployed at the smart energy meter. A clock reference signal is generated by a crystal oscillator with temperature compensation function, and clock alignment is established with the global positioning system timing module, the Beidou satellite synchronization module, or the power station broadcast time code receiving module. After clock alignment is completed, the A / D converter is started at fixed time intervals to synchronously sample the three-phase voltage and the user line current. Each phase is configured with a 12-bit or higher precision sampling channel, the sampling frequency is not less than 10kHz, and a high-resolution timestamp is added to the sampled value. Based on the sampled data, the zero crossover point of the voltage and current waveforms is extracted, the phase difference is calculated and a sliding window filter is applied to generate a continuous phase difference curve. A direction discrimination baseline is established based on the mean, variance, and maximum deviation of the phase difference during a stable period. A zero-power crossing point and a direction reversal threshold are set, and a direction baseline consistency index is output to provide a direction identification benchmark.

3. The method for detecting electricity theft based on a smart meter according to claim 2, characterized in that, The steps for determining the suspected time window for current backflow are as follows: The instantaneous voltage values ​​of the three-phase conductors and the neutral line-to-ground voltage values ​​are obtained under the time reference of the direction discrimination baseline. The data is sampled every 10 milliseconds to form a continuous neutral line potential curve and establish a synchronization relationship with the three-phase voltage data. By comparing the neutral line potential curve cycle by cycle, the sudden increase or decrease segment with a change amplitude exceeding 1.5 volts within one electrical cycle is extracted, and the start time, maximum amplitude and duration are recorded, and compared with the change in the three-phase voltage amplitude. Under the premise of neutral line potential disturbance, the voltage phase and current phase difference are used as a reference to identify the situation where the phase difference change amplitude exceeds 15 degrees and the change direction is consistent within three consecutive cycles. This is determined as a directional disturbance event, and the event parameters are recorded. Multiple directional disturbance events are integrated based on a time interval of less than 500 milliseconds, and extended by two electrical cycles before and after to form a suspected current backflow time window, which is then output as a basis for subsequent analysis.

4. The method for detecting electricity theft based on a smart meter according to claim 3, characterized in that, The steps for outputting the confidence level of current backflow are as follows: Within the suspected current backflow time window, taking the monitored user as the center, multiple smart energy meters connected to the same low-voltage branch are selected as references to extract active power, reactive power, current RMS value and power factor data. The sampling period is 20 milliseconds, and the data are aligned to a unified start and end time period through timestamps. The load data of the monitored electricity meter is compared with the data of each reference electricity meter hourly to construct a multi-dimensional difference matrix, forming a response difference structure with time as the horizontal axis and electricity meter number as the vertical axis. Statistical analysis of the difference matrix is ​​performed hourly to identify single-point anomalies or lack of group consistency, and the duration of the anomaly, the trend of power direction change, and the phase deviation are recorded. The behavioral consistency analysis results are linked with the neutral line potential disturbance amplitude, phase change rate and directional baseline score. The current backflow confidence is generated by weighting the power change amplitude, duration period and deviation degree, and a high confidence judgment is output when the result exceeds the threshold.

5. The method for detecting electricity theft based on a smart meter according to claim 4, characterized in that, The steps for verifying causality are as follows: When the built-in confidence score of the suspected current backflow time window exceeds the threshold, and the neutral line potential changes abruptly, the voltage and current phase difference jumps, and there is a single-point load anomaly, the voltage reference phase is adjusted by one to three degrees in the internal control chip of the energy meter based on the phase-locked reference frame, and the disturbance duration does not exceed one power frequency cycle. Before and after the disturbance injection, the voltage phase, current phase, phase difference and active power symbol are sampled every ten milliseconds. Data for ten consecutive cycles are recorded to generate a disturbance response time series. The direction discrimination difference, maximum response amplitude, number of response delay cycles and phase difference abrupt change amplitude before and after the disturbance are calculated to construct a direction response change profile. Using the natural fluctuation period without injected disturbance as a reference, the direction discrimination value, phase difference change and power direction trend are compared. When the phase difference change exceeds 15 degrees after the disturbance is triggered and the direction reversal lasts for more than two cycles, it is confirmed that the disturbance and the direction reversal have a causal relationship, and the direction reversal evidence data unit is output.

6. The method for detecting electricity theft based on a smart meter according to claim 5, characterized in that, The steps for generating perturbation feature fingerprints are as follows: The confidence score of current backflow, the abrupt change rate of phase difference, the amplitude of directional response gain and the directional reversal characteristics were normalized and fused according to the weighted ratio to obtain the comprehensive current backflow evidence score. When the evidence score is greater than or equal to 8.0, a level 1 anomaly alarm is triggered, and the direction identification output of the target energy meter is frozen during the time window. At the same time, a data packet containing the energy meter address, event timestamp, score value and indicator details is sent to the monitoring platform, and the original voltage and current sampling values, phase trajectory and power symbol sequence are cached for no less than 72 hours. The temporal features, amplitude features, and associated behavioral features recorded in the triggered event are combined with the geographic electrical information in a structured manner to form disturbance feature fingerprint information. The event number and scoring result are attached, saved to the local database and uploaded to the central data platform for subsequent abnormal behavior comparison and repeated trigger identification.

7. The method for detecting electricity theft based on a smart meter according to claim 6, characterized in that, By using abnormal coupling characteristics to reverse locate the disturbing user and output intervention control commands, the following steps are taken to identify and suppress current backflow behavior: Based on the start and end times, maximum response amplitude time points and phase reversal nodes recorded in the disturbance feature fingerprint, a unified alignment window is extracted. Multiple smart energy meters in the same distribution branch are selected to collect active power, reactive power, current amplitude and voltage phase data, and they are aligned with a granularity of ten milliseconds. The load parameters of each energy meter in the reference period before the disturbance are standardized and calculated to generate a two-dimensional response matrix. Based on standardization, the point-by-point response difference between the target energy meter and each adjacent energy meter is calculated to form a multi-user relative load response difference matrix. The center-diffusion abrupt change consistent with the disturbance feature fingerprint is identified to determine the suspected disturbance source. By combining the power supply topology and phase sequence wiring information, a reverse propagation chain is constructed to identify the disturbance source and compare the flip-over start point, voltage jump point and neutral line potential peak value. After confirming the user who is actively disturbing the system, an identification confirmation signal is sent to the monitoring platform to freeze the direction judgment result of the target electricity meter in the corresponding time period, mark the high-confidence disturbance source, and trigger an intervention control command to increase the data sampling frequency to five milliseconds.

8. A smart meter-based electricity theft detection system, used to implement the smart meter-based electricity theft detection method according to any one of claims 1-7, comprising a phase-locked reference module, a neutral line drift analysis module, a multi-meter linkage difference analysis module, a perturbation injection causality verification module, an evidence fusion judgment module, and a perturbation fingerprint tracing and intervention module: The phase-locked reference module constructs a phase-locked reference framework at the smart energy meter. It acquires voltage and current phases through high-precision time synchronization, generates a direction discrimination baseline, sets the zero-power crossing point and direction reversal threshold, and outputs a direction baseline consistency index as a direction identification benchmark. The neutral line drift analysis module calculates the neutral line potential drift curve in real time based on the direction discrimination baseline, extracts the transient characteristics of three-phase imbalance, calculates the drift increment and phase difference abrupt change rate, and maps the dynamic change results to the direction discrimination baseline to determine the time window of suspected current backflow. The multi-meter linkage difference analysis module performs time alignment on the load curves of multiple adjacent smart energy meters for the suspected current backflow time window, constructs a relative load response difference matrix, judges single-point anomalies or group consistency loss, and outputs the current backflow confidence level. The perturbation injection causality verification module injects a micro-amplitude phase perturbation signal into the phase-locked reference frame based on the suspected time window and the current backflow confidence level. It monitors the directional response gain, removes natural fluctuations, extracts the directional reversal feature, and completes the causality verification. The evidence fusion and judgment module, based on the causal verification results, weights and fuses the current backflow confidence, phase difference mutation rate, direction response gain and direction reversal features to generate a current backflow evidence score. When the score exceeds the threshold, an alarm is triggered, the direction recognition weight is frozen, and the perturbation feature fingerprint is recorded. The disturbance fingerprint tracing and intervention module performs short-cycle alignment of the load sequences of multiple smart energy meters based on disturbance feature fingerprint information, constructs a multi-user response difference matrix, generates load disturbance fingerprints, and reverse-locates the disturbing user through abnormal coupling features, outputs intervention control commands, and realizes the identification and suppression of current backflow behavior.

Images