Electricity meter electricity stealing behavior identification method and system based on multi-feature fusion

Abstract

The application discloses a kind of based on multi-feature fusion electric energy meter electricity larceny behavior identification method and system, belong to electric power system electricity larceny detection technical field.The method includes: in the transmission sub-time slot of detection time slot, to power line inject detection signal;During the transmission of detection signal, the composition signal is synchronously sampled, the transmission signal component corresponding to detection signal is subtracted from composition signal, and network response component is extracted;The network response component is carried out frequency domain transformation, and the channel frequency response curve is obtained;The difference curve between channel frequency response curve and preset reference response curve is calculated, and the electricity larceny behavior identification result is determined according to the morphological characteristics of difference curve.Through the application, the problem that prior art lacks active incentive means and can only judge whether there is change but cannot identify electricity larceny type is solved, and active detection and accurate classification to electricity larceny behavior are realized.

Description

Technical Field

[0001] This invention relates to a method and system for identifying electricity theft behavior in electricity meters based on multi-feature fusion, belonging to the field of electricity theft detection technology in power systems. Background Technology

[0002] With the rapid development of smart grids and the comprehensive coverage of electricity information collection systems, electricity meters, as the core metering node between the power grid and users, directly affect the economic benefits and power supply order of the power system through their operational security and accuracy. To prevent users from illegally tampering with metering circuits or connecting bypasses to steal electricity, power departments have placed higher demands on the anti-theft detection capabilities of electricity meters.

[0003] Existing methods for detecting electricity theft from power meters typically rely on passively monitoring the communication signal characteristics on power lines or directly collecting and analyzing conventional metering data such as voltage and current. When the meter is in a communication silence period or there are no obvious abnormal fluctuations in electricity consumption, these passive detection methods are often ineffective and cannot detect physical topological changes in a timely manner. Furthermore, existing detection methods are relatively limited, often only providing a rough conclusion of "change" or "no change" by comparing signal waveforms or electricity consumption data before and after the theft.

[0004] This passive detection method lacks active stimulation and multi-dimensional feature extraction of power line network characteristics, making it impossible to implement effective detection during communication silence periods. Furthermore, it can only determine whether there is electricity theft but cannot accurately identify the specific type of electricity theft (such as metal wire bypass, capacitor or inductive coupling bypass, etc.), resulting in a lack of specificity in subsequent manual investigations and low efficiency in electricity theft investigation.

[0005] Therefore, a new solution is needed to address this problem. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for identifying electricity theft behavior based on multi-feature fusion. It solves the problem that the existing technology lacks active incentive means and can only judge whether there is a change but cannot identify the type of electricity theft, and realizes active detection and accurate classification of electricity theft behavior.

[0007] The technical problem to be solved by this invention is achieved by the following technical solution:

[0008] A method and system for identifying electricity theft behavior based on multi-feature fusion, including:

[0009] During the transmit sub-slot of the probe time slot, a probe signal is injected into the power line;

[0010] During the transmission of the detection signal, a composite signal is synchronously acquired, and the transmission signal component corresponding to the detection signal is subtracted from the composite signal to extract the network response component;

[0011] The network response components are subjected to frequency domain transformation to obtain the channel frequency response curve;

[0012] Calculate the difference curve between the channel frequency response curve and the preset reference response curve, and determine the electricity theft behavior identification result based on the morphological characteristics of the difference curve.

[0013] Optionally, the synchronous acquisition and synthesis signal includes:

[0014] The synthesized signal is attenuated by a high-impedance passive voltage divider network, and the attenuated synthesized signal is input to an analog-to-digital converter for sampling.

[0015] Optionally, the step of subtracting the transmitted signal component corresponding to the probe signal from the synthesized signal to extract the network response component includes:

[0016] The operating frequency band is divided into multiple sub-bands, with the detection signal as the reference input and the synthesized signal as the desired signal;

[0017] An adaptive filter is used for iterative calculation within each sub-band, and the adaptive filtering result of the synthesized signal is subtracted from that of the probe signal to obtain the network response component;

[0018] The filter coefficients of the adaptive filter are initialized to the convergence values ​​of the previous round of detection.

[0019] Optionally, injecting a detection signal into the power line includes:

[0020] A linear frequency modulated signal is generated as the detection signal, or a coded sequence at multiple preset frequency points is generated as the detection signal;

[0021] The step of performing frequency domain transformation on the network response components to obtain the channel frequency response curve includes:

[0022] A short-time Fourier transform is performed on the network response components, and the transform results of each sub-band or frequency point are combined to obtain the channel frequency response curve covering the entire frequency band.

[0023] Optionally, obtaining the preset benchmark response curve includes:

[0024] Obtain the historical response curve sequences corresponding to each of the multiple operating conditions within a historical time period;

[0025] At each frequency point, the median of the historical response curve sequence is taken as the preset reference response curve under the corresponding operating condition, and the absolute deviation of the median at each frequency point is calculated as the dispersion scale; the operating conditions include weekday daytime, weekday nighttime, and holidays.

[0026] Optionally, determining the electricity theft behavior identification result based on the morphological characteristics of the difference curve includes:

[0027] When the difference curve shows a uniform negative shift across the entire frequency band, the electricity theft behavior is identified as a metal wire bypass.

[0028] When the attenuation increment of the difference curve increases significantly with increasing frequency, the electricity theft behavior is identified as capacitive coupling bypass.

[0029] When the attenuation increment of the difference curve increases significantly as the frequency decreases, the electricity theft behavior is identified as inductively coupled bypass.

[0030] Optionally, it also includes:

[0031] When distortion of the difference curve is detected, switch to sensing mode to obtain the total received power spectrum of the environment;

[0032] The direction of change of the total received power spectrum in the environment is compared with the direction of distortion of the difference curve. When the two directions are consistent, the electricity theft behavior is confirmed.

[0033] Optionally, it also includes:

[0034] Calculate the maximum amplitude deviation or root mean square deviation of the difference curve at each frequency point, and use it as the severity of distortion.

[0035] The cumulative change is obtained by summing the positive differences between two adjacent detections within the time window for the severity of the distortion.

[0036] When the cumulative change exceeds the cumulative threshold, and the severity of the distortion in a single instance is lower than the preset trigger threshold, the electricity theft behavior identification result is determined to be a progressive structural change.

[0037] When the electricity theft behavior identification result is confirmed as electricity theft, the preset benchmark response curve is locked;

[0038] When the lock time window expires and at least one of the following conditions is met, the lock is released and the preset baseline response curve is reconstructed:

[0039] Received manual confirmation and exclusion instruction;

[0040] The distortion severity detected in multiple consecutive tests is less than half of the severity threshold, and the consistency between the distortion direction of the difference curve and the power spectrum change direction returns to normal.

[0041] Optionally, it also includes:

[0042] When the energy of the network response component detected multiple times is lower than a preset energy threshold, or the transmit power monitoring value is abnormal, or the feedback signal of the coupling transformer is abnormal, the detection capability is determined to be faulty.

[0043] The step of injecting a detection signal into the power line is paused, and conventional metering data is instead acquired.

[0044] When the power consumption in the conventional metering data suddenly drops or the power factor changes abruptly, a statistical anomaly is identified, and a degradation detection result is output.

[0045] A multi-feature fusion-based electricity meter theft detection system includes:

[0046] The signal injection module is used to inject a detection signal into the power line within the transmit sub-time slot of the detection time slot;

[0047] The synchronous acquisition module is used to synchronously acquire the synthesized signal through a hardware sampling front-end during the transmission of the detection signal, and to transmit the sampled data to the extraction module;

[0048] The extraction module is used to extract the network response component by subtracting the transmitted signal component corresponding to the detection signal from the synthesized signal through adaptive filtering, using the detection signal as a reference input and the synthesized signal as the desired signal.

[0049] The frequency domain transformation module is used to perform frequency domain transformation on the network response components to obtain the channel frequency response curve.

[0050] The distortion decision module is used to calculate the difference curve between the channel frequency response curve and the preset reference response curve, and to determine the electricity theft behavior identification result based on the morphological characteristics of the difference curve.

[0051] The above modules are software modules that run on the main control microcontroller and are managed uniformly by the main program scheduler.

[0052] The beneficial effects of this invention are:

[0053] 1. By actively injecting detection signals into the power line during the detection time slot of the communication silence period and extracting the network response components to generate the channel frequency response curve, the specific types of electricity theft such as metal wire bypass, capacitive coupling bypass or inductive coupling bypass can be accurately identified based on the morphological characteristics of the difference curve in the whole frequency band (such as uniform negative offset, high frequency attenuation increase or low frequency attenuation increase), providing a targeted basis for accurately combating electricity theft.

[0054] 2. When extracting network response components, a high-impedance passive voltage divider network is introduced to linearly attenuate the synthesized signal, physically bypassing the RF switch and low-noise amplifier of the conventional communication receiving link. This effectively avoids nonlinear saturation distortion caused by high-power transmission signals, providing a linear and distortion-free input signal for subsequent adaptive filtering and ensuring the accuracy of feature extraction.

[0055] 3. In the adaptive filtering stage, a frequency domain block processing strategy is adopted, which divides the entire frequency band into multiple sub-bands and uses a single-tap filter to model each sub-band. This effectively reduces the computational complexity of the low-computing-power microcontroller unit and ensures the stable operation of high-frequency, high-resolution real-time detection and sensing on low-cost embedded hardware. Attached Figure Description

[0056] Figure 1 This is a diagram illustrating the product composition of the present invention.

[0057] Figure 2 This is the main flowchart of the method of the present invention.

[0058] Figure 3 This is a flowchart of the distortion morphology classification of the present invention.

[0059] Figure 4 This is a flowchart of the severity and locking logic of the present invention.

[0060] Figure 5 This is the system architecture diagram of the present invention.

[0061] Figure 6 This is a diagram showing the relationship between software module scheduling and data flow in this invention. Detailed Implementation

[0062] To facilitate a clear understanding of the technical means, creative features, objectives, and effects of this invention, the invention will be further described below in conjunction with specific illustrations.

[0063] To address the technical problems in existing technologies, such as lack of active incentives, single detection dimensions, inability to perform detection during communication silence periods, and the inability to identify the type of electricity theft that can only determine whether there is a change, this application provides a method for identifying electricity theft behavior based on multi-feature fusion.

[0064] like Figure 1As shown, this method is integrated inside the electricity meter. The electricity meter, as a smart metering terminal, has the ability to transmit and receive half-duplex power line carrier signals. Its hardware components include: a main control microcontroller unit, a 4-inch color LCD display, a radio frequency-power line carrier dual-mode module, a wireless fidelity module, a Bluetooth module, a 4×4 keypad, a data storage module, and a power supply module.

[0065] The main control microcontroller unit uses the M0 core with a main frequency of 29 MHz. As the core processor of the energy meter, it is responsible for all management work and has software upgrade capabilities.

[0066] The power line carrier section of the RF-power line carrier dual-mode module operates in the high-speed power line carrier standard frequency band (2MHz to 12MHz), adopts a half-duplex architecture, and the transmission and reception are time-division controlled, sharing a coupling transformer. It has three working states: transmission mode, reception mode, and sensing mode. The main control microcontroller unit controls the module to switch between the above three modes (switching time ≥ 5 microseconds).

[0067] The electricity meter integrates a hardware front-end and a core algorithm layer for detecting electricity theft. At the physical layer, the electricity meter injects a detection signal into the power line through the transmission link of the radio frequency-power line carrier dual-mode module, and uses a high-impedance passive voltage divider network as the hardware sampling front-end to provide a distortion-free input signal for subsequent algorithms.

[0068] At the algorithm layer, the software modules running on the main control microcontroller unit include: a synchronous acquisition and extraction module, a frequency domain transformation module, a distortion decision module, and a response execution module. The synchronous acquisition and extraction module consists of a synchronous acquisition sub-function and an adaptive filtering sub-function, responsible for acquiring hardware sampling data and extracting network response components, respectively. The distortion decision module includes a dynamic benchmark management sub-module, a power spectrum verification sub-module, and a distortion severity and progressive detection sub-module. The response execution module includes a degradation detection sub-module. The sampling data output from the high-impedance passive voltage divider network is converted from analog to digital and then transmitted to the synchronous acquisition and extraction module for further processing.

[0069] As a standard configuration of smart meters, electricity meters are suitable for mass deployment at electricity user metering sites. During on-site installation, they connect to the power grid via a power line carrier communication interface, directly identifying electricity theft behavior in their own metering circuits and power line networks. No external independent equipment is required, making installation convenient and in line with the standardization requirements of smart grids.

[0070] refer to Figure 2 The method includes the following steps:

[0071] Step S1 is executed within the transmit sub-time slot of the probe time slot to inject a probe signal into the power line.

[0072] The intelligent unit generates a probe signal and injects it into the power line via a coupling transformer through the transmission link of the power line carrier module. This probe signal covers the high-speed power line carrier operating frequency band. The probe signal is a linear frequency modulated (LFM) signal, with the frequency linearly scanned from 2MHz to 12MHz. The scan duration is determined according to the preset frequency resolution to ensure that the sweep bandwidth covers 2MHz to 12MHz and meets the frequency resolution requirements. The preset frequency resolution is less than or equal to 0.1MHz, and the typical scan duration is 200 microseconds. Optionally, the probe signal can also be a coded sequence at multiple preset frequency points, transmitted sequentially in steps less than or equal to 0.5MHz within the 2MHz to 12MHz frequency band, with a dwell time at each frequency point greater than or equal to 10 microseconds. The coded sequence is a Barker code or a pseudo-random binary sequence. The transmit power spectral density of the probe signal meets the high-speed power line carrier standard limit, less than or equal to -45dBm / Hz.

[0073] While performing step S1, step S2 is performed. During the transmission of the detection signal, the synthesized signal is acquired synchronously, and the transmission signal component corresponding to the detection signal is subtracted from the synthesized signal to extract the network response component.

[0074] In a conventional half-duplex power line carrier architecture, the transmit and receive links are connected to the power line via a coupling transformer. During the transmission signal's duration, the signal on the coupling transformer simultaneously contains both the transmitted signal component and the network response component. However, high-power transmitted signals can cause saturation of the low-noise amplifier in the receive link, leading to nonlinear distortion and preventing subsequent adaptive filtering from effectively extracting the network response component. This embodiment introduces a high-impedance passive voltage divider network as a hardware-level anti-saturation mechanism. At the physical level, this linearly attenuates the high-power transmitted signal to within the linear quantization range of the analog-to-digital converter, effectively avoiding the aforementioned nonlinear distortion risk and providing a linear, distortion-free input signal for subsequent adaptive filtering. Specifically, during the transmission signal's duration, the receive link synchronously samples the synthesized signal on the coupling transformer via the high-impedance passive voltage divider network. This high-impedance passive voltage divider network consists of passive resistors connected in parallel between the front end of the coupling transformer or the power amplifier output and the spare sampling channel of the analog-to-digital converter. The high-impedance passive voltage divider network has less than 0.1% impact on the normal communication RF impedance. The synthesized signal is attenuated using a high-impedance passive voltage divider network. This attenuated signal is then input to an analog-to-digital converter (ADC) for sampling, physically bypassing the RF switches and low-noise amplifiers of conventional communication receiver links, thus avoiding saturation distortion in the analog front-end. The ADC has a sampling rate greater than or equal to 24 MHz and a resolution greater than or equal to 12 bits. The high-impedance passive voltage divider network linearly attenuates the synthesized signal to within the ADC's full-scale range of -3 dB to -6 dB.

[0075] The process of subtracting the transmitted signal component from the synthesized signal is implemented using adaptive filtering. Since full-band time-domain adaptive filtering faces computational bottlenecks on a microcontroller unit with a 29MHz main frequency, this embodiment employs a buffered time-division processing strategy. The synthesized signal acquired during the detection time slot is buffered in the data storage module, and the main control microcontroller unit reads the data for computation during non-detection periods, effectively circumventing computational limitations. During computation, the operating frequency band from 2MHz to 12MHz is divided into multiple sub-bands, each with a bandwidth of 1MHz. Since the coupling coefficient within each sub-band is approximately constant, a single-tap filter can be used for effective modeling, significantly reducing computational complexity. Using the detection signal as the reference input and the synthesized signal as the desired signal, iterative computation is performed using an adaptive filter within each sub-band. The adaptive filter employs the least mean square algorithm, subtracting the adaptive filtering result of the synthesized signal from that of the detection signal to obtain the network response component. The filter coefficients of the adaptive filter are initialized to the convergence value of the previous detection round; only fine-tuning is needed in steady state, without needing to converge from zero each time. During the initial detection, the initial values ​​of the filter coefficients are obtained through short-time calibration measurements. These measurements are taken during the learning phase after the smart unit is installed. A known detection signal is injected into the power line, and the amplitude ratio of the detection signal to the synthesized signal on the coupling transformer is measured to determine the initial estimate of the coupling coefficients. The coupling coefficients are determined by the frequency response of the coupling transformer and vary smoothly within the operating frequency band; therefore, a single-tap adaptive filter can be effectively used for modeling.

[0076] Optionally, the adaptive filter can also employ a recursive least squares algorithm. When the probe signal is a coded sequence at multiple preset frequency points, the above subband filtering process is performed at each frequency point.

[0077] After extracting the network response components, step S3 is executed to perform a frequency domain transformation on the network response components to obtain the channel frequency response curve. Specifically, a short-time Fourier transform is performed on the network response components to obtain the frequency response values ​​of each frequency point or sub-band. The transformation results of each sub-band or frequency point are combined to obtain a channel frequency response curve covering the entire frequency band from 2MHz to 12MHz. When using a linear frequency modulated signal, a frequency resolution corresponding to a 200-microsecond sweep duration is 5kHz. When using coded sequences at multiple preset frequency points, the frequency response values ​​extracted during the dwell period of each frequency point are stitched together according to the frequency coordinates to form a discrete channel frequency response curve covering the entire frequency band. The horizontal axis of the channel frequency response curve represents frequency, and the vertical axis represents signal attenuation, with the unit being decibels (dB).

[0078] After obtaining the channel frequency response curve at the current moment, a preset reference response curve needs to be obtained. Step S4 is executed to obtain the preset reference response curve. Specifically, the intelligent unit enters a learning phase after initial installation, which lasts for 7 days. During the learning phase, the intelligent unit performs detection according to a preset cycle, obtaining one response curve per cycle. The intelligent unit obtains the historical response curve sequences corresponding to multiple operating conditions within a historical time period. Operating conditions include weekday daytime, weekday nighttime, and holidays. Weekday daytime is from 06:00 to 18:00, weekday nighttime is from 18:00 to 06:00 the next day, and the entire day of a holiday is considered an independent operating condition.

[0079] At each frequency point, the median of all response values ​​in the historical response curve sequence is taken as the preset reference response curve for the corresponding operating condition. The number of detections under this operating condition during the learning phase is greater than or equal to 100. Simultaneously, the median of the absolute values ​​of the deviations between the historical response values ​​and the preset reference response curve at each frequency point is calculated as the dispersion scale. Under normal conditions, when no distortion is detected, the preset reference response curve is gradually updated with a preset weight of 0.01 to adapt to the long-term, slow changes in household load.

[0080] After obtaining the current channel frequency response curve and the preset reference response curve, step S5 is executed to calculate the difference curve between the channel frequency response curve and the preset reference response curve, and the electricity theft behavior identification result is determined based on the morphological characteristics of the difference curve.

[0081] refer to Figure 3 Specifically, the difference between the current channel frequency response curve and the preset reference response curve at each frequency point is calculated to obtain the difference curve. A positive value on the difference curve indicates enhanced response, while a negative value indicates increased attenuation. Based on the morphological characteristics of the difference curve across the entire frequency band, distortion types are classified, and the identification result of electricity theft is determined. For ease of understanding, typical numerical references for each type of difference curve are given below: Under normal conditions where no electricity theft has occurred, the absolute value of the deviation of the difference curve at each frequency point is usually less than 2 dB; when the deviation exceeds this normal fluctuation range and exhibits a specific shape, it indicates suspected electricity theft.

[0082] When the difference curve exhibits a uniform negative shift across the entire frequency band from 2MHz to 12MHz, meaning the attenuation increment at each frequency point increases by an equal amount and is greater than 3 dB, while the fluctuation between frequency points is less than 1 dB, the electricity theft behavior is identified as a metal wire bypass. For example, in a typical metal wire bypass scenario, the attenuation increment of the difference curve at each frequency point across the entire frequency band is approximately 6 dB, and the fluctuation between frequency points does not exceed 0.5 dB. The mechanism of the uniform negative shift lies in the fact that the metal wire bypass branch increases the signal shunt loss across the entire frequency band.

[0083] When the attenuation increment of the difference curve increases significantly with increasing frequency—that is, the difference between the attenuation increment at high frequencies (greater than 6 MHz) and low frequencies (less than 4 MHz) exceeds 5 dB—the electricity theft is identified as capacitive coupling bypass. For example, in a typical capacitive coupling bypass scenario, the attenuation increment at 2 MHz is approximately 2 dB, while at 12 MHz it can reach over 15 dB. The mechanism of capacitive coupling bypass lies in the fact that the capacitive reactance of the capacitor decreases with increasing frequency, making high-frequency current shunting more significant.

[0084] When the attenuation increment of the difference curve increases significantly with decreasing frequency—that is, the difference between the attenuation increment in the low-frequency band (less than 4 MHz) and the attenuation increment in the high-frequency band (greater than 6 MHz) exceeds 5 dB—the electricity theft is identified as inductively coupled bypass. For example, in a typical inductively coupled bypass scenario, the attenuation increment at 12 MHz is approximately 2 dB, while at 2 MHz it can reach over 15 dB. The mechanism of inductively coupled bypass lies in the fact that the inductive reactance of the inductor decreases with decreasing frequency, making low-frequency current shunting more significant.

[0085] Further, in step S6, when distortion of the difference curve is detected, the system switches to sensing mode to acquire the total received power spectrum of the environment and verify the electricity theft identification result. In sensing mode, the analog-to-digital converter directly samples the output of the coupling transformer without going through a demodulation link to measure the original environmental signal. The direction of change of the total received power spectrum of the environment is compared with the direction of distortion of the difference curve. When the two directions are consistent, the electricity theft identification result is confirmed. When no distortion is detected, or when the direction of change of the total received power spectrum of the environment no longer shifts in the same direction as the direction of distortion of the difference curve, it is determined that the consistency has returned to normal.

[0086] When the morphological characteristics of the difference curve match a specific type of electricity theft, and the distortion severity exceeds the severity threshold and persists, while the power spectrum consistency verification passes, the electricity theft behavior is identified as confirmed electricity theft.

[0087] Further, step S7 is performed to calculate the distortion severity and perform progressive structural change detection, referencing... Figure 4 The distortion severity is the maximum amplitude deviation of the difference curve at each frequency point. Optionally, the distortion severity can also be the root mean square deviation of the difference curve.

[0088] The severity threshold is determined statistically based on the fluctuations of the response curve under normal conditions in historical data. Specifically, the 3σ criterion is adopted, which means that the sum of the mean of distortion severity in the most recent 50 normal detections and three times the standard deviation is taken as the severity threshold. Normal fluctuations within three times the standard deviation will not trigger an alarm.

[0089] The positive difference in distortion severity between two adjacent detections within the time window is calculated. When the difference is negative, it is set to zero. The cumulative sum of these positive differences is then calculated to obtain the cumulative change. When the cumulative change exceeds the cumulative threshold and the severity of a single distortion is below the preset trigger threshold, the electricity theft behavior is determined to be a progressive structural change.

[0090] When the distortion severity exceeds the severity threshold and the distortion persists in multiple consecutive detections, it is judged as a suspected persistent topological change.

[0091] After confirming the electricity theft detection result, the smart unit responds to the judgment by executing alarm display, cloud push, or power outage command. The alarm display is shown on a color LCD screen, including information on the type of electricity theft. The cloud push is sent to the user terminal or main station via a wireless fidelity module or Bluetooth module.

[0092] Further, in step S8, when the electricity theft behavior is confirmed, a preset baseline response curve is locked. During the locking period, the preset baseline response curve for the current operating condition remains unchanged. The locking time window is 24 hours. During the locking period, each newly detected response curve is stored as continuous evidence of electricity theft in the local storage and uploaded to the cloud with a timestamp.

[0093] When the lockout window expires and at least one of the following conditions is met, the lockout is lifted and the preset reference response curve is reconstructed. The first condition is receiving a manual confirmation instruction to eliminate the theft. The second condition is that the distortion severity of multiple consecutive detections is less than half of the severity threshold, and the distortion direction of the difference curve returns to normal consistency with the direction of change of the total received power spectrum of the environment. If the lockout window expires but the electricity theft is not eliminated, the preset reference response curve remains locked and is not automatically released. The system continues to record evidence of electricity theft and encrypts and uploads it until manual confirmation or on-site repair is performed. After the lockout is lifted, the preset reference response curve for the current operating condition is reconstructed using the median frequency of multiple consecutive detection results after the lockout.

[0094] During the detection process, the intelligent unit also executes step S9, degradation detection. After extracting the network response components, it determines whether the energy of the network response components detected multiple times is lower than a preset energy threshold, or whether the transmitted power monitoring value is abnormally low or zero, or whether the feedback signal of the coupling transformer is abnormal, such as the feedback voltage being lower than the normal coupling threshold or the waveform showing severe distortion. If any of the above conditions are met, the detection capability is determined to be ineffective, and the execution of subsequent steps and the step of injecting detection signals into the power line is suspended. Instead, conventional metering data is acquired, which includes voltage, current, power, energy consumption, and power factor. When there is a sudden drop in energy consumption or a sudden change in the power factor in the conventional metering data, a statistical anomaly is identified, and a degradation detection result is output. The degradation detection result is labeled as "suspected electricity theft" and marked with a low confidence level. If the above conditions are not met, step S3 is executed.

[0095] refer to Figure 5 This invention also provides a multi-feature fusion-based electricity meter theft identification system to implement the aforementioned multi-feature fusion-based electricity meter theft identification method. The system includes a multi-modal data intelligent processing unit, which has at least half-duplex power line carrier signal transmission and reception capabilities. The aforementioned modules are software modules running on the main control microcontroller unit and are uniformly managed by the main program scheduler. The specific module division is as follows.

[0096] The signal injection module is used to inject a probe signal into the power line within the transmit sub-slot of the probe time slot. The probe signal covers the high-speed power line carrier operating frequency band from 2 MHz to 12 MHz, and is a linear frequency modulated signal or a coded sequence at multiple preset frequency points. The transmit power spectral density meets the high-speed power line carrier standard limits.

[0097] The synchronous acquisition module is used to synchronously acquire the synthesized signal through a hardware sampling front-end during the transmission of the detection signal, and then transmit the sampled data to the extraction module. Specifically, the hardware sampling front-end is a high-impedance passive voltage divider network, which attenuates the synthesized signal before inputting it to the analog-to-digital converter for sampling.

[0098] The extraction module uses the probe signal as a reference input and the synthesized signal as the desired signal. It extracts the network response component by subtracting the transmitted signal component corresponding to the probe signal from the synthesized signal through adaptive filtering. Specifically, within each subband, an adaptive filter is used for iterative calculations, and the network response component is obtained by subtracting the adaptive filtering result of the synthesized signal from that of the probe signal.

[0099] The frequency domain transformation module is used to perform frequency domain transformation on the network response components to obtain the channel frequency response curve. Specifically, a short-time Fourier transform is performed on the network response components, and the transformation results of each sub-band or frequency point are combined to obtain a channel frequency response curve covering the entire frequency band.

[0100] The distortion decision module is used to calculate the difference curve between the channel frequency response curve and the preset reference response curve, and to determine the electricity theft behavior identification result based on the morphological characteristics of the difference curve. Specifically, based on the different shapes of the difference curve, such as uniform negative offset across the entire frequency band, significant increase in attenuation increment in the high frequency band, or significant increase in attenuation increment in the low frequency band, the electricity theft behavior identification result is determined to be metal wire bypass, capacitive coupling bypass, or inductive coupling bypass, respectively.

[0101] The dynamic reference management module maintains the historical response curve sequence and establishes and manages the normal state reference response curve. The preset reference response curve is the median of the frequency points in the historical response curve sequence, and the absolute deviation of the median at each frequency point is used as the dispersion scale. After electricity theft is confirmed, the dynamic reference management module executes a locking and condition release mechanism. The release condition is that the locking time window expires and meets the manual confirmation and exclusion command, or the distortion severity of multiple consecutive detections is less than half of the severity threshold, and the consistency between the distortion direction of the difference curve and the change direction of the total received power spectrum of the environment returns to normal.

[0102] The power spectrum verification module is used to switch to the sensing mode to obtain the total received power spectrum of the environment when distortion of the difference curve is detected. It compares the direction of change of the total received power spectrum of the environment with the direction of distortion of the difference curve. When the two directions are consistent, the electricity theft behavior identification result is confirmed.

[0103] The distortion severity and progressive detection module is used to calculate the maximum amplitude deviation or root mean square deviation of the difference curve at each frequency point as the distortion severity, and to calculate the cumulative sum of the positive differences in distortion severity between two adjacent detections within the time window as the cumulative change. When the cumulative change exceeds the cumulative threshold and the single distortion severity is lower than the preset trigger threshold, the electricity theft behavior identification result is determined to be a progressive structural change.

[0104] The multi-dimensional feature fusion decision module is used to fuse distortion morphology, distortion severity, temporal persistence, and power spectrum consistency for decision-making. Electricity theft is confirmed when the distortion morphology matches the feature library, the distortion persists, and the power spectrum consistency is satisfactory.

[0105] The response execution module is used to respond to the judgment result execution alarm display, cloud push, or power outage command.

[0106] The degradation detection module is used to automatically switch to a degradation detection mode based on conventional metering data when the detection capability fails. When the network response component energy is lower than the preset energy threshold or the transmit power monitoring value is abnormally low or zero after multiple consecutive probes, the degradation detection module obtains the power and power factor from the conventional metering data. When the power suddenly drops or the power factor changes abruptly, the degradation detection result is output.

[0107] refer to Figure 6The electricity theft detection function of this invention is integrated into the embedded software system of the main control microcontroller unit of the electricity meter in the form of a task. The main program of the electricity meter includes two levels: a regular task loop and an electricity theft detection task. The regular task loop is responsible for daily functions such as metering data acquisition, power line carrier communication protocol stack processing, human-machine interaction, and cloud communication, and runs in the main time slice of the main program. The electricity theft detection task is periodically triggered by the timer interrupt service routine of the main program. The interval between two adjacent detections is configurable, with a typical value of 10 minutes.

[0108] In a complete detection process, the data flow path is as follows: First, the main program calls the driver interface of the RF-power line carrier dual-mode module to control it to switch to transmission mode and inject a detection signal within a specified time slot. Simultaneously, the high-impedance passive voltage divider network attenuates and samples the synthesized signal on the coupling transformer, and the analog-to-digital converter stores the sampled data in a dedicated buffer of the data storage module. Then, after the detection is complete, the main program notifies the electricity theft detection task to enter the data processing stage via a callback function.

[0109] Due to the limited computing power of the main microcontroller unit, a time-sharing strategy is adopted in the data processing stage. Specifically, the main program schedules the execution of the synchronous acquisition module, extraction module, frequency domain transformation module, and distortion decision module sequentially only during non-detection time slots (i.e., communication quiet periods or idle times). After the calculations are completed, the decision result is transmitted to the response execution module via a message queue. This module then calls the display driver, communication protocol stack, or power outage control interface to execute the specific response action. Throughout the process, the main program interacts with each software module through well-defined application programming interfaces. Each module is integrated into the main program image as a static link library, and its lifecycle is managed uniformly by the main program's scheduler.

[0110] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and various changes and modifications can be made without departing from the spirit and scope of the invention, all of which fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying electricity theft behavior based on multi-feature fusion, characterized in that, include: During the transmit sub-slot of the probe time slot, a probe signal is injected into the power line; During the transmission of the detection signal, a composite signal is synchronously acquired, and the transmission signal component corresponding to the detection signal is subtracted from the composite signal to extract the network response component; The network response components are subjected to frequency domain transformation to obtain the channel frequency response curve; Calculate the difference curve between the channel frequency response curve and the preset reference response curve, and determine the electricity theft behavior identification result based on the morphological characteristics of the difference curve.

2. The method according to claim 1, characterized in that, The synchronously acquired and synthesized signal includes: The synthesized signal is attenuated by a high-impedance passive voltage divider network, and the attenuated synthesized signal is input to an analog-to-digital converter for sampling.

3. The method according to claim 1, characterized in that, The step of subtracting the transmitted signal component corresponding to the probe signal from the synthesized signal to extract the network response component includes: The operating frequency band is divided into multiple sub-bands, with the detection signal as the reference input and the synthesized signal as the desired signal; An adaptive filter is used for iterative calculation within each sub-band, and the adaptive filtering result of the synthesized signal is subtracted from that of the probe signal to obtain the network response component; The filter coefficients of the adaptive filter are initialized to the convergence values ​​of the previous round of detection.

4. The method according to claim 1, characterized in that, The injection of a detection signal into the power line includes: A linear frequency modulated signal is generated as the detection signal, or a coded sequence at multiple preset frequency points is generated as the detection signal; The step of performing frequency domain transformation on the network response components to obtain the channel frequency response curve includes: A short-time Fourier transform is performed on the network response components, and the transform results of each sub-band or frequency point are combined to obtain the channel frequency response curve covering the entire frequency band.

5. The method according to claim 1, characterized in that, Obtaining the preset benchmark response curve includes: Obtain the historical response curve sequences corresponding to each of the multiple operating conditions within a historical time period; At each frequency point, the median of the historical response curve sequence is taken as the preset reference response curve under the corresponding operating condition, and the absolute deviation of the median at each frequency point is calculated as the dispersion scale; the operating conditions include weekday daytime, weekday nighttime, and holidays.

6. The method according to claim 1, characterized in that, The step of determining the electricity theft behavior identification result based on the morphological characteristics of the difference curve includes: When the difference curve shows a uniform negative shift across the entire frequency band, the electricity theft behavior is identified as a metal wire bypass. When the attenuation increment of the difference curve increases significantly with increasing frequency, the electricity theft behavior is identified as capacitive coupling bypass. When the attenuation increment of the difference curve increases significantly as the frequency decreases, the electricity theft behavior is identified as inductively coupled bypass.

7. The method according to claim 6, characterized in that, Also includes: When distortion of the difference curve is detected, switch to sensing mode to obtain the total received power spectrum of the environment; The direction of change of the total received power spectrum in the environment is compared with the direction of distortion of the difference curve. When the two directions are consistent, the electricity theft behavior is confirmed.

8. The method according to claim 6, characterized in that, Also includes: Calculate the maximum amplitude deviation or root mean square deviation of the difference curve at each frequency point, and use it as the severity of distortion. The cumulative change is obtained by summing the positive differences between two adjacent detections within the time window for the severity of the distortion. When the cumulative change exceeds the cumulative threshold, and the severity of the distortion in a single instance is lower than the preset trigger threshold, the electricity theft behavior identification result is determined to be a progressive structural change. When the electricity theft behavior identification result is confirmed as electricity theft, the preset benchmark response curve is locked; When the lock time window expires and at least one of the following conditions is met, the lock is released and the preset baseline response curve is reconstructed: Received manual confirmation and exclusion instruction; The distortion severity detected in multiple consecutive tests is less than half of the severity threshold, and the consistency between the distortion direction of the difference curve and the power spectrum change direction returns to normal.

9. The method according to claim 1, characterized in that, Also includes: When the energy of the network response component detected multiple times is lower than a preset energy threshold, or the transmit power monitoring value is abnormal, or the feedback signal of the coupling transformer is abnormal, the detection capability is determined to be faulty. The step of injecting a detection signal into the power line is paused, and conventional metering data is instead acquired. When the power consumption in the conventional metering data suddenly drops or the power factor changes abruptly, a statistical anomaly is identified, and a degradation detection result is output.

10. A system for identifying electricity theft behavior based on multi-feature fusion, characterized in that, include: The signal injection module is used to inject a detection signal into the power line within the transmit sub-time slot of the detection time slot; The synchronous acquisition module is used to synchronously acquire the synthesized signal through a hardware sampling front-end during the transmission of the detection signal, and to transmit the sampled data to the extraction module; The extraction module is used to extract the network response component by subtracting the transmitted signal component corresponding to the detection signal from the synthesized signal through adaptive filtering, using the detection signal as a reference input and the synthesized signal as the desired signal. The frequency domain transformation module is used to perform frequency domain transformation on the network response components to obtain the channel frequency response curve. The distortion decision module is used to calculate the difference curve between the channel frequency response curve and the preset reference response curve, and to determine the electricity theft behavior identification result based on the morphological characteristics of the difference curve. The above modules are software modules that run on the main control microcontroller and are managed uniformly by the main program scheduler.

Images