Time-frequency feature extraction method and system for interference signal of electric energy meter clock synchronization

Abstract

The application discloses a method and system for extracting time-frequency characteristics of interference signals by using clock synchronization of electric energy meters, and relates to the field of data processing. The method comprises the following steps: obtaining a plurality of sampling signal groups of a plurality of electric energy meters in a power distribution area; detecting abnormal triggering under time sequence correlation according to a clock synchronization mechanism to determine a suspected interference event; intercepting a multi-measurement-point signal segment corresponding to the interference event, performing cross-measurement-point event matching and time deviation correction, and obtaining a signal set of the interference event; performing joint time-frequency analysis on the signal set to construct a joint time-frequency feature set; extracting cross-measurement-point propagation characteristics of the interference signal to perform reverse tracing compensation mining to obtain an interference tracing compensation vector; and performing anti-interference reconstruction on the signal set. The application solves the technical problems of low interference recognition accuracy and insufficient traceability in existing electric energy meter interference signal analysis and processing, and achieves the technical effects of improving interference signal recognition accuracy and realizing accurate tracing of interference sources and propagation paths.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method and system for extracting time-frequency features of interference signals using electricity meter clock synchronization. Background Technology

[0002] In power systems, the accuracy of sampling data from electricity meters in distribution areas is a core foundation for ensuring fair electricity metering and reliable grid operation monitoring. Various external interference signals can easily cause deviations in electricity meter sampling data, which not only affects the accuracy of electricity metering but may also mislead grid operation and maintenance decisions, leading to power dispatching errors, equipment failures, and other problems. Currently, the mainstream approach in the industry for handling interference signals from electricity meters is to combine independent monitoring of a single measuring point with conventional filtering. Specifically, a basic filtering module is configured on a single electricity meter to perform simple low-pass and high-pass filtering on the collected voltage and current signals to remove obvious interference noise. At the same time, the electricity meter operation logs are checked periodically by manual review, and historical normal data is compared to identify abnormal signal fluctuations, thereby determining whether there are interference events and initially investigating the scope of interference impact. However, the above-mentioned processing methods have obvious limitations. Independent monitoring of a single measuring point cannot take into account the signal correlation between multiple energy meters in the power distribution area. Conventional filtering can only remove simple noise and is difficult to distinguish between interference signals and normal power load fluctuations, resulting in low accuracy in identifying interference events and easy misjudgment or omission. At the same time, the lack of effective analysis of the propagation law of interference signals makes it impossible to accurately locate the source of interference and the propagation path. The efficiency of interference event processing is low and it is difficult to meet the needs of refined operation and maintenance in large-scale power distribution areas.

[0003] At present, the analysis and processing of interference signals from electricity meters suffers from technical problems such as low interference identification accuracy and insufficient source tracing capability. Summary of the Invention

[0004] This application provides a method and system for extracting time-frequency features of interference signals using electricity meter clock synchronization. The method involves acquiring sampling signal groups from multiple electricity meters within a distribution area, identifying suspected interference events through anomaly detection based on time-series correlation using a clock synchronization mechanism, extracting multi-measurement point signal segments corresponding to these events, and obtaining a set of signals related to the same interference event through cross-measurement point event matching and time deviation correction. Joint time-frequency analysis is then performed on this signal set to construct a joint time-frequency feature set reflecting the time-frequency correlation among multiple measurement points. Based on this feature set, cross-measurement point propagation features of the interference signal are extracted. An interference tracing compensation vector is obtained through reverse tracing compensation mining, and this compensation vector is used to reconstruct the set of signals related to the same interference event, thus completing the processing and optimization of the interference signal. This approach solves the technical problems of low interference identification accuracy and insufficient source tracing capability in existing electricity meter interference signal analysis and processing, achieving the technical effects of improving interference signal identification accuracy and realizing precise tracing of interference sources and propagation paths.

[0005] This application provides a method for extracting time-frequency features of interference signals using electricity meter clock synchronization, comprising: acquiring multiple sampling signal groups from multiple electricity meters within a power distribution area, and performing anomaly trigger detection on the multiple sampling signal groups under time-series correlation according to a clock synchronization mechanism to determine suspected interference events; extracting multi-measurement point signal segments corresponding to the suspected interference events, and performing cross-measurement point event matching and time deviation correction on the multi-measurement point signal segments to obtain a set of signals with the same interference events; performing joint time-frequency analysis on the set of signals with the same interference events to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measurement points; extracting cross-measurement point propagation features of interference signals based on the joint time-frequency feature set, and performing reverse tracing compensation mining based on the cross-measurement point propagation features of interference signals to obtain an interference tracing compensation vector; and performing anti-interference reconstruction on the set of signals with the same interference events based on the interference tracing compensation vector.

[0006] In a possible implementation, anomaly trigger detection under time-correlation is performed on the multiple sampled signal groups according to the clock synchronization mechanism to determine suspected interference events, and the following processing is performed: the multiple sampled signal groups are uniformly time-aligned according to the clock synchronization mechanism to obtain a multi-measurement-point aligned signal set; time-correlation fluctuation anomaly detection is performed on the multi-measurement-point aligned signal set to determine suspected interference trigger points; the suspected interference trigger points are expanded with preceding and following nodes according to the multi-measurement-point aligned signal set to establish a suspected interference trigger window; interference event features are captured on the multi-measurement-point aligned signal set according to the suspected interference trigger window to generate the suspected interference event.

[0007] In a possible implementation, time-series correlation fluctuation anomaly detection is performed on the multi-measurement point aligned signal set to determine suspected interference trigger points, and the following processing is performed: Time-frequency fluctuation indices are obtained, including time-domain fluctuation indices and frequency-domain offset indices; time-frequency fluctuation analysis is performed on the multi-measurement point aligned signal set based on the time-frequency fluctuation indices to obtain time-frequency fluctuation matrices for each signal; an interference suspected detection network is trained based on the electricity meter interference historical event set; the time-frequency fluctuation matrices of each signal are input into the interference suspected detection network to obtain multiple interference suspected scores; suspected interference signals are determined based on the multiple interference suspected scores, and the multi-dimensional signal feature parameters of the suspected interference signals are output as the suspected interference trigger points.

[0008] In a possible implementation, cross-measurement point event matching and time deviation correction are performed on the multi-measurement point signal segments to obtain a set of signals with the same interference events. The following processing is then performed: Time deviation characteristic analysis is conducted on the multi-measurement point signal segments based on the sampled suspected interference events to obtain multi-dimensional time deviation characteristics, including clock residual deviation, sampling delay deviation, and communication timestamp deviation; time compensation and correction are performed on the multi-dimensional time deviation characteristics to obtain a set of corrected signal segments; a suspected interference time-frequency vector is constructed based on the sampled suspected interference events, and time-frequency similarity analysis is performed on the corrected signal segment set based on the suspected interference time-frequency vector to obtain an interference time-frequency similarity distribution; the corrected signal segment set is then filtered based on the interference time-frequency similarity distribution to generate the set of signals with the same interference events.

[0009] In a possible implementation, joint time-frequency analysis is performed on the set of co-interference event signals to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measurement points. The following processes are then performed: independent time-frequency transformation is performed on each signal in the set of co-interference event signals to obtain time-frequency feature information of multiple signals; synchronization analysis is performed based on the time-frequency feature information of multiple signals to obtain synchronization time-frequency features; mutual information is calculated on the time-frequency feature information of multiple signals to obtain the time-frequency mutual information distribution; time-frequency interaction modeling is performed on the set of co-interference event signals based on the synchronization time-frequency features and the time-frequency mutual information distribution to obtain a time-frequency interaction model; and the time-frequency feature information of multiple signals is fused based on the time-frequency interaction model to generate the joint time-frequency feature set.

[0010] In a possible implementation, reverse tracing compensation mining is performed based on the cross-measuring point propagation characteristics of the interference signal to obtain the interference tracing compensation vector, and the following processing is performed: the cross-measuring point propagation characteristics of the interference signal include interference arrival time sequence characteristics, interference energy propagation attenuation characteristics, interference frequency offset consistency, and interference signal correlation strength matrix; using the installation location topology of the multiple energy meters as the node reference, the interference arrival time sequence characteristics as the edge weight reference, and the interference energy propagation attenuation characteristics as the node weight reference, an interference propagation weighted directed graph is constructed; the interference propagation weighted directed graph is traversed in reverse optimization according to the interference frequency offset consistency and the interference signal correlation strength matrix to determine the interference path tracing result; compensation parameters are configured according to the interference path tracing result to generate the interference tracing compensation vector.

[0011] In a possible implementation, the interference propagation weighted directed graph is traversed in reverse optimization based on the interference frequency offset consistency and the interference signal correlation strength matrix to determine the interference path tracing result, and the following processing is performed: a reverse search is performed based on the interference propagation weighted directed graph to obtain a set of search interference paths; frequency consistency matching is performed on the set of search interference paths based on the interference frequency offset consistency to obtain a set of candidate interference paths; and time-frequency feature intensity weighted filtering is performed on the set of candidate interference paths based on the interference signal correlation strength matrix to obtain the interference path tracing result.

[0012] In a possible implementation, the following process is performed: the clock synchronization mechanism includes synchronizing the clocks of the plurality of energy meters to a standard time source via a network time protocol.

[0013] In a possible implementation, the following processing is performed: the time-domain fluctuation index includes amplitude kurtosis, pulse interval deviation, and short-time energy mutation rate; the frequency-domain offset index includes center frequency drift, harmonic distortion rate, and frequency band energy percentage change rate.

[0014] This application also provides a system for extracting time-frequency features of interference signals using electricity meter clock synchronization, comprising: an abnormal trigger detection module, used to acquire multiple sampling signal groups from multiple electricity meters in a power distribution area, and perform abnormal trigger detection on the multiple sampling signal groups under time-series correlation according to the clock synchronization mechanism to determine suspected interference events; a cross-measuring point event matching module, used to extract multi-measuring point signal segments corresponding to the suspected interference events, and perform cross-measuring point event matching and time deviation correction on the multi-measuring point signal segments to obtain a set of signals with the same interference events; a joint time-frequency analysis module, used to perform joint time-frequency analysis on the set of signals with the same interference events to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measuring points; a reverse tracing compensation mining module, used to extract cross-measuring point propagation features of interference signals according to the joint time-frequency feature set, and perform reverse tracing compensation mining according to the cross-measuring point propagation features of interference signals to obtain an interference tracing compensation vector; and an anti-interference reconstruction module, used to perform anti-interference reconstruction on the set of signals with the same interference events according to the interference tracing compensation vector.

[0015] The proposed method and system for extracting time-frequency features of interference signals using electricity meter clock synchronization first acquires multiple sampling signal groups from multiple electricity meters within a power distribution area. Based on the clock synchronization mechanism, abnormal trigger detection is performed on these sampling signal groups under time-series correlation to identify suspected interference events. Next, multi-measurement point signal segments corresponding to the suspected interference events are extracted, and cross-measurement point event matching and time deviation correction are performed on these multi-measurement point signal segments to obtain a set of signals with the same interference event. Then, joint time-frequency analysis is performed on this set of signals with the same interference event to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measurement points. Based on this joint time-frequency feature set, cross-measurement point propagation features of the interference signal are extracted, and reverse tracing compensation mining is performed based on these cross-measurement point propagation features to obtain an interference tracing compensation vector. Finally, the set of signals with the same interference event is reconstructed using the interference tracing compensation vector to improve anti-interference capabilities. The proposed method and system achieve the technical effects of improving the accuracy of interference signal identification and realizing precise tracing of interference sources and propagation paths. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the method for extracting time-frequency features of interference signals using electricity meter clock synchronization, as provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of the interference signal time-frequency feature extraction system that utilizes the clock synchronization of an electricity meter, as provided in an embodiment of this application.

[0019] Figure labeling: 10 for abnormal trigger detection module, 20 for cross-measuring point event matching module, 30 for joint time-frequency analysis module, 40 for reverse tracing compensation mining module, and 50 for anti-interference reconstruction module. Detailed Implementation

[0020] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments; however, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides a method for extracting the time-frequency features of interference signals using electricity meter clock synchronization, such as... Figure 1 As shown, the method includes:

[0024] Step S100: Obtain multiple sampling signal groups from multiple energy meters within the power distribution area, and perform abnormal trigger detection on the multiple sampling signal groups under time-series correlation according to the clock synchronization mechanism to determine suspected sampling interference events. The clock synchronization mechanism includes synchronizing the clocks of the multiple energy meters to a standard time source through a network time protocol.

[0025] Specifically, multiple smart meters installed within the power distribution area are selected as measurement points. Each meter is equipped with a built-in sampling module and a network communication module. Each meter collects a set of sampling data containing voltage and current signals. Through the meter's network communication module, the sampling data collected by all meters is transmitted in real time to an edge computing gateway. The edge computing gateway classifies and stores the received sampling data, forming multiple sampling signal groups. Each sampling signal group corresponds to all the sampling data collected by one meter. A network time protocol is used for clock synchronization among the multiple meters, as follows: The edge computing gateway connects to the internet and obtains a standard time source provided by a time synchronization center. The time accuracy of the standard time source is controlled at the millisecond level. The edge computing gateway sends a clock synchronization command to each meter at fixed intervals through a network time protocol client. After receiving the clock synchronization command, the meter adjusts its built-in clock to match the standard time source. After synchronization, each meter adds a corresponding standard timestamp to the sampling data when collecting it. Based on the aforementioned clock synchronization mechanism, anomaly trigger detection is performed on multiple sampled signal groups under time-series correlation to identify suspected interference events. Specifically, each sampled signal group is first timestamped to ensure that the time axes of all sampled signal groups are fully aligned. Then, the signal processing unit built into the edge computing gateway performs time-series correlation analysis on the calibrated sampled signal groups to identify areas where the signal amplitude and frequency fluctuate abnormally, thereby identifying suspected interference events. The suspected interference events include key information such as the start time, end time, signal amplitude, frequency, and energy characteristics of the interference signal.

[0026] In one possible implementation, anomaly trigger detection is performed on the multiple sampling signal groups under time-series correlation based on a clock synchronization mechanism to determine suspected interference events. Step S100 further includes step S110, which performs unified time alignment on the multiple sampling signal groups according to the clock synchronization mechanism to obtain a multi-measurement point aligned signal set. Specifically, the timestamps of all sampled data in each sampling signal group are obtained. Using the time of the standard time source as a reference, the timestamps of each sampling signal group are verified one by one. If the timestamp of a certain sampled data deviates from the standard time, the timestamp of the sampled data is corrected by linear interpolation to ensure that the timestamp of each sampled data is completely consistent with the standard time source. A unified time alignment window is determined. The start time of the time alignment window is set to the earliest sampling time in all sampling signal groups, and the end time is set to the latest sampling time in all sampling signal groups. The time interval of the time alignment window is consistent with the sampling interval of the energy meter. For each time node, sampled data corresponding to all electricity meters at that time node is extracted. If no sampled data is collected for a certain electricity meter at that time node, linear interpolation is performed using sampled data from the two adjacent time nodes to supplement the data, ensuring that sampled data corresponding to all electricity meters is available at each time node. The sampled data from all time nodes are arranged in chronological order to form a multi-measurement point aligned signal set. The multi-measurement point aligned signal set is stored in the database of the edge computing gateway, with time as the horizontal axis and the sampled data of each electricity meter as the vertical axis.

[0027] Step S120: Time-series correlation fluctuation anomaly detection is performed on the multi-measurement point aligned signal set to identify suspected interference trigger points. Specifically, a time-series correlation fluctuation anomaly detection system is built, integrated into an edge computing gateway. This system consists of a data input unit, a time-frequency fluctuation index calculation unit, a signal analysis unit, a detection network unit, and a result output unit. The data input unit inputs the multi-measurement point aligned signal set to the time-frequency fluctuation index calculation unit, which calculates the time-frequency fluctuation index of the sampled signal at each measurement point and each time node. The time-frequency fluctuation index includes a time-domain fluctuation index and a frequency-domain offset index. After calculation, the time-frequency fluctuation index data for each measurement point is output. The signal analysis unit receives the time-frequency fluctuation index data, performs time-series correlation processing on it, analyzes the time-frequency fluctuation patterns of different measurement points at the same time node and the same measurement point at different time nodes, and identifies the time nodes with abnormal fluctuations. The detection network unit further verifies the time nodes with abnormal fluctuations, excluding normal signal fluctuations, such as small fluctuations caused by changes in power load, and finally determines the suspected interference trigger points, including the corresponding specific measurement point, time node, and signal fluctuation parameters of that node.

[0028] Step S130: Expand the suspected interference trigger point by nodes before and after it according to the multi-measurement point aligned signal set to establish a suspected interference trigger window. Specifically, determine the expansion range of the suspected interference trigger window. Combining the propagation characteristics of interference signals in the power distribution area and the duration of historical interference events, the expansion range is preferably set to 50 time nodes before and after the time node corresponding to the suspected interference trigger point, that is, each suspected interference trigger window contains 101 time nodes. If the suspected interference trigger point is located at the beginning of the multi-measurement point aligned signal set, and the first 50 time nodes are insufficient, expand from the beginning time node of the multi-measurement point aligned signal set; if the suspected interference trigger point is located at the end of the multi-measurement point aligned signal set, and the last 50 time nodes are insufficient, expand to the end time node of the multi-measurement point aligned signal set. For each suspected interference trigger point, extract the sampled signal data of all measurement points at all time nodes within the expansion range from the multi-measurement point aligned signal set, including voltage, current amplitude, timestamp, and other information. The extracted sampled signal data are sorted and organized according to time sequence and measurement points to form suspected interference trigger windows. Each suspected interference trigger window corresponds to a suspected interference trigger point, and the window contains sampled signal data from multiple measurement points within a certain period before and after the trigger point.

[0029] Step S140: Based on the suspected interference trigger window, interference event features are captured from the multi-measurement point aligned signal set to generate the sampled suspected interference event. Specifically, for each suspected interference trigger window, the feature capture unit built into the edge computing gateway is used to capture interference event features from the multi-measurement point sampled signal data within the window. The captured features include the start time, end time, signal amplitude, frequency, and energy characteristics of the interference signal. The energy characteristics include the total signal energy and the energy percentage of each frequency band. The start and end times of the interference signal are determined by: traversing all time nodes within the suspected interference trigger window and calculating the short-time energy mutation rate for each time node. When the short-time energy mutation rate first exceeds a preset upper threshold, the time node is determined as the start time of the interference signal; when the short-time energy mutation rate is less than a preset lower threshold for 10 consecutive time nodes, the first time node of these 10 consecutive nodes is determined as the end time of the interference signal. The amplitude characteristics of the interference signal are extracted by calculating the maximum, minimum, and average signal amplitude of each measurement point from the start time to the end time of the interference signal, which are used as the amplitude characteristics of the interference signal at that measurement point. Frequency characteristics of the interference signal are extracted using a Fast Fourier Transform (FFT) algorithm to calculate the signal center frequency and frequency fluctuation range at each measurement point during the interference period, which are then used as frequency characteristics. Energy characteristics of the interference signal are also extracted by calculating the total signal energy at each measurement point during the interference period, which is the sum of the squares of the amplitudes of all sampling points during the interference period, as well as the proportion of energy in the 50Hz-60Hz and 60Hz-70Hz frequency bands to the total energy, which are then used as energy characteristics. Finally, the captured interference start time, end time, amplitude characteristics, frequency characteristics, and energy characteristics at each measurement point are summarized to form sampled suspected interference events. Each suspected interference trigger window corresponds to one sampled suspected interference event.

[0030] In one possible implementation, the multi-measurement point aligned signal set undergoes time-series correlation fluctuation anomaly detection to identify suspected interference trigger points. Step S120 further includes step S121, acquiring time-frequency fluctuation indicators. These indicators include time-domain fluctuation indicators and frequency-domain offset indicators. The time-domain fluctuation indicators include amplitude kurtosis, pulse interval deviation, and short-time energy mutation rate. The frequency-domain offset indicators include center frequency drift, harmonic distortion rate, and frequency band energy proportion change rate. Specifically, the basic data required for calculating the time-frequency fluctuation indicators is acquired. This basic data consists of sampled signal data from each measurement point and each time node in the multi-measurement point aligned signal set, including voltage signal amplitude, current signal amplitude, signal sampling time, and other information. This basic data is directly retrieved from the database of the edge computing gateway. The time-domain fluctuation index is calculated as follows: Amplitude kurtosis is calculated using the fourth-order central moment method. First, the average signal amplitude of 100 consecutive sampling points at a given measurement point is calculated. Then, the fourth power of the difference between the amplitude of each sampling point and the average is calculated. The average of all fourth powers of these differences is then divided by the fourth power of the standard deviation of the amplitude of that group of sampling points to obtain the amplitude kurtosis of that measurement point for that time period. Pulse interval deviation is calculated by first identifying the occurrence time of the pulse signal in the sampled signal, calculating the time interval between two adjacent pulse signals, and then calculating the difference between all time intervals and the average time interval. The average of the absolute values ​​of these differences is taken as the pulse interval deviation. Short-time energy mutation rate is calculated by dividing the sampled signal into short-time windows of 50 sampling points each. The signal energy within each short-time window is calculated as the sum of the squares of the amplitudes of each sampling point within the window. The energy difference between two adjacent short-time windows is then calculated. This difference is divided by the energy of the previous window to obtain the short-time energy mutation rate. Then, the frequency domain offset indices are calculated as follows: The center frequency drift is calculated by performing a Fast Fourier Transform (FFT) algorithm on the sampled signal in the frequency domain. First, the center frequency of the sampled signal under normal operating conditions is calculated, then the center frequency of the current sampled signal is calculated, and the difference between the two is the center frequency drift. The harmonic distortion rate is calculated by extracting the fundamental amplitude and harmonic amplitudes of the sampled signal using the FFT algorithm. The harmonic distortion rate is the square root of the sum of the squares of the harmonic amplitudes, divided by the fundamental amplitude. The frequency band energy ratio change rate is calculated by dividing the frequency domain range of the sampled signal into multiple frequency bands, calculating the proportion of energy in each frequency band to the total signal energy, and then calculating the difference between the current frequency band energy ratio and the frequency band energy ratio under normal conditions. This difference is divided by the frequency band energy ratio under normal conditions to obtain the frequency band energy ratio change rate. The calculated time-domain fluctuation indices and frequency domain offset indices are then summarized to form a set of time-frequency fluctuation indices for each measurement point and each time node.

[0031] Step S122: Analyze the time-frequency fluctuations of the multi-measurement point aligned signal set according to the time-frequency fluctuation indices to obtain the time-frequency fluctuation matrices for each signal. Specifically, determine the dimensions of the time-frequency fluctuation matrix: rows correspond to time nodes of the multi-measurement point aligned signal set, columns correspond to types of time-frequency fluctuation indices, and each matrix element corresponds to a specific value of a certain time node and a certain type of time-frequency fluctuation indices. For each measurement point, extract the values ​​of six time-frequency fluctuation indices corresponding to each time node from the time-frequency fluctuation indices set, and fill them into the corresponding rows of the matrix in order of time node, and into the corresponding columns of the matrix in order of index type. Normalize the values ​​filled into the matrix using the min-max normalization method, mapping all values ​​corresponding to each index to the interval between 0 and 1. Specifically, first find the maximum and minimum values ​​among all values ​​of a certain index, subtract the minimum value from each value of the index, and then divide by the difference between the maximum and minimum values ​​to obtain the normalized values. Each measurement point generates a normalized time-frequency fluctuation matrix. The time-frequency fluctuation matrices of all measurement points together form the time-frequency fluctuation matrix of each signal, which is stored in the signal parsing unit of the edge computing gateway.

[0032] Step S123: Train the interference suspicion detection network based on the historical event set of electricity meter interference. Specifically, obtain the historical event set of electricity meter interference. This event set comes from the electricity meter operation records of the past year in the distribution area and is extracted from the historical database of the edge computing gateway. Each historical interference event contains complete multi-point sampling signal data, time-frequency fluctuation index data, interference trigger point information, and the confirmation result of the interference event, i.e., whether the event is a real interference event. Construct the interference suspicion detection network. This network adopts a three-layer neural network structure, including an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is consistent with the number of columns in the time-frequency fluctuation matrix, i.e., 6 neurons, corresponding to the six time-frequency fluctuation indices. There are two hidden layers: the first hidden layer has 12 neurons and the second hidden layer has 6 neurons. The hidden layers use the ReLU activation function. The output layer has 1 neuron, using the sigmoid activation function. The output result is a value between 0 and 1, used to represent the suspicion that the input signal is an interference signal. The data in the historical event set is preprocessed. The time-frequency fluctuation matrix of each historical event is used as input data, and the confirmation results of interference events are used as labels. The label for real interference events is 1, and the label for non-interference events is 0. The preprocessed data is divided into training and test sets in a 7:3 ratio to train the interference suspicion detection network. The training process is as follows: the input data of the training set is input into the network input layer, and features are extracted through the ReLU activation function of the hidden layer. Then, the suspicion prediction value is output through the sigmoid activation function of the output layer. The mean square error between the prediction value and the label is calculated as the loss function. The gradient descent method is used to update the weights and biases of the network. The learning rate is set to 0.01, and the number of iterations is set to 1000. The accuracy of the training set is calculated after each iteration. When the accuracy reaches above 95% and tends to stabilize, the training is stopped. The trained interference detection network is validated using a test set. The input data of the test set is input into the network, and the matching degree between the output suspected interference prediction value and the label is calculated. If the validation accuracy reaches 93% or higher, the interference detection network is considered to have completed training and can be used to detect suspected interference signals. If the validation accuracy does not reach 93%, the number of hidden layer neurons and the learning rate of the network are adjusted, and the network is retrained until the validation requirements are met.

[0033] Step S124: Input the time-frequency fluctuation matrices of each signal into the interference suspicion detection network to obtain multiple interference suspicion scores. Specifically, the time-frequency fluctuation matrix corresponding to each measurement point in the time-frequency fluctuation matrix of each signal is sequentially input into the trained interference suspicion detection network. The interference suspicion detection network processes the input time-frequency fluctuation matrices, receives data through the input layer, extracts features through the ReLU activation function of the hidden layer, and passes the extracted features to the output layer. The sigmoid activation function is used to calculate and output the interference suspicion score for each time node corresponding to the measurement point. The suspicion score ranges from 0 to 1. The closer the value is to 1, the greater the probability that the signal at that time node is an interference signal; the closer the value is to 0, the greater the probability that the signal at that time node is a normal signal. The interference suspicion scores output by the network are sorted and stored according to the order of measurement points and time nodes. Each measurement point corresponds to an interference suspicion score sequence, and each value in the sequence corresponds to the interference suspicion score of the time node corresponding to the measurement point. The interference suspicion score sequences of all measurement points are summarized to obtain multiple interference suspicion scores, which are stored in the detection network unit of the edge computing gateway.

[0034] Step S125: Determine suspected interference signals based on the multiple interference suspicion levels, and output the multi-dimensional signal characteristic parameters of the suspected interference signals as the suspected interference trigger points. Specifically, an interference suspicion threshold is set. The suspicion threshold is determined through multiple experiments combined with the verification results of historical interference event sets. For example, the suspicion threshold is set to 0.8, meaning that when the interference suspicion level at a certain time point is greater than or equal to 0.8, the signal corresponding to that time point is recorded as a suspected interference signal; when the interference suspicion level is less than 0.8, the signal corresponding to that time point is recorded as a normal signal. The interference suspicion level sequence of all measurement points is traversed, and it is determined whether the interference suspicion level at each time point reaches the threshold. If the threshold is reached, the time point, the corresponding measurement point, and the interference suspicion level value at that time point are recorded, and it is determined as a suspected interference signal. Then, multidimensional signal feature parameters of the suspected interference signals are extracted, including the signal amplitude, frequency, amplitude kurtosis, pulse interval deviation, short-time energy mutation rate, center frequency drift, harmonic distortion rate, frequency band energy ratio change rate, and timestamp information at that time node. These parameters are directly extracted from the multi-measurement point aligned signal set and the time-frequency fluctuation index set. The extracted multidimensional signal feature parameters are organized, with each suspected interference signal corresponding to a set of multidimensional signal feature parameters. These parameters are used as the output results of the suspected interference trigger point and stored in the result output unit of the edge computing gateway, while simultaneously marking the measurement point and time node of the suspected interference trigger point.

[0035] Step S200: Extract the multi-measurement point signal segments corresponding to the sampled suspected interference events, and perform cross-measurement point event matching and time deviation correction on the multi-measurement point signal segments to obtain a set of signals with the same interference events.

[0036] Specifically, multi-measurement point signal segments corresponding to suspected interference events are extracted. For each suspected interference event, based on its interference start and end times, sampled signal data from all measurement points within that time range are extracted from the multi-measurement point aligned signal set to form multi-measurement point signal segments. The time range of these multi-measurement point signal segments is completely consistent with the interference duration of the suspected interference event. Each measurement point corresponds to one signal segment, which contains all sampled data from that measurement point during the interference period. A cross-measurement point event matching and time deviation correction system is built, integrated into the edge computing gateway. This system consists of a signal segment input unit, a time deviation analysis unit, a time correction unit, a time-frequency similarity analysis unit, and a result filtering unit. The multi-measurement point signal segments are input to the time deviation analysis unit to analyze the time deviation between the signal segments from each measurement point and perform time deviation correction. Then, the time-frequency similarity analysis unit performs cross-measurement point event matching to filter out signal segments belonging to the same interference event, ultimately forming a set of signals from the same interference event.

[0037] In one possible implementation, cross-measurement point event matching and time deviation correction are performed on the multi-measurement point signal segments to obtain a set of interference event signals. Step S200 further includes step S210, which analyzes the time deviation characteristics of the multi-measurement point signal segments based on the sampled suspected interference events to obtain multi-dimensional time deviation characteristics, including clock residual deviation, sampling delay deviation, and communication timestamp deviation. Specifically, the reference time information of the sampled suspected interference events is obtained, and the interference start time of the measurement point with the earliest interference start time in the sampled suspected interference events is used as the reference time, and this measurement point is used as the reference measurement point. The clock residual deviation is calculated, which is the deviation between the clock of each measurement point and the clock of the reference measurement point. Specifically, the timestamp of the interference start time in each measurement point signal segment is extracted and compared with the reference time; the difference between the two is the clock residual deviation. The sampling delay deviation is calculated. First, the sampling delay of each measurement point is obtained through the sampling module parameters of the energy meter, that is, the time from receiving the sampling command to completing the sampling by the sampling module. The sampling delay deviation is the difference between the sampling delay of the sampling module of each measurement point and the sampling delay of the reference measurement point. To calculate the communication timestamp deviation, firstly, the sampling time of the sampled data at each measurement point and the receiving time of the edge computing gateway are recorded. The difference between the two is the communication delay. The communication timestamp deviation is the difference between the communication delay of the sampled data from each measurement point to the edge computing gateway and the communication delay of the reference measurement point. Finally, the clock residual deviation, sampling delay deviation, and communication timestamp deviation of each measurement point are summarized to form the multidimensional time deviation characteristic of that measurement point. The multidimensional time deviation characteristics of all measurement points together form a multidimensional time deviation characteristic set.

[0038] Step S220: Perform time compensation and correction on the multi-measurement point signal segments according to the multi-dimensional time deviation characteristics to obtain a set of corrected signal segments. Specifically, for the multi-measurement point signal segments of each measurement point, calculate the total time deviation based on the multi-dimensional time deviation characteristics of that measurement point. The total time deviation is the sum of clock residual deviation, sampling delay deviation, and communication timestamp deviation. Perform time compensation and correction on the signal segments of that measurement point based on the total time deviation. The specific method is as follows: If the total time deviation is positive, it indicates that the signal segments of that measurement point are lagging behind the reference measurement point. Subtract the total time deviation from all timestamps of the signal segments of that measurement point to achieve time advance correction. If the total time deviation is negative, it indicates that the signal segments of that measurement point are ahead of the reference measurement point. Add the absolute value of the total time deviation to all timestamps of the signal segments of that measurement point to achieve time lag correction. Perform time axis alignment verification on the corrected signal segments. Compare the time of all corrected signal segments of all measurement points with the signal segments of the reference measurement point to ensure that the timestamp error of the signal segments of all measurement points at the same time node does not exceed a preset error threshold. The corrected signal segments from all measuring points are compiled to form a set of corrected signal segments. Each signal segment in the set of corrected signal segments is perfectly aligned with the time axis of the signal segments at the reference measuring point, thus eliminating the time deviation between the measuring points.

[0039] Step S230: Based on the sampled suspected interference events, construct a suspected interference time-frequency vector, and perform time-frequency similarity analysis on the corrected signal segment set based on the suspected interference time-frequency vector to obtain the interference time-frequency similarity distribution. Specifically, the suspected interference time-frequency vector is constructed by extracting the time-frequency feature parameters of the signal segment from the benchmark measurement point in the sampled suspected interference events. These parameters include the average signal amplitude, average center frequency, average harmonic distortion rate, average short-time energy mutation rate, and average frequency band energy proportion, totaling five parameters. These five parameters are arranged in order to form the suspected interference time-frequency vector, with a vector dimension of 1 row and 5 columns. The time-frequency feature parameters of each signal segment in the corrected signal segment set are extracted using the same method as the extraction method for the benchmark measurement point's time-frequency feature parameters. Each signal segment corresponds to a 1 row and 5 column time-frequency feature vector. The time-frequency similarity between the time-frequency feature vector of each signal segment and the suspected interference time-frequency vector is calculated using a cosine similarity algorithm. Specifically, the dot product of the two vectors is calculated first, followed by the magnitude of each vector. The dot product is then divided by the product of the magnitudes to obtain the cosine similarity. The cosine similarity value ranges from -1 to 1. A value closer to 1 indicates greater similarity in the time-frequency features of the two vectors, meaning the corresponding signal segments are more likely to belong to the same interference event. Conversely, a value closer to -1 indicates greater difference in the time-frequency features of the two vectors, meaning the corresponding signal segments are less likely to belong to the same interference event. The time-frequency similarity values ​​of all signal segments are arranged in order of measurement points to form an interference time-frequency similarity distribution. This distribution, with the measurement points on the horizontal axis and the time-frequency similarity value on the vertical axis, represents the degree of similarity between each measurement point's signal segment and the suspected interference time-frequency vector.

[0040] Step S240: Based on the interference time-frequency similarity distribution, the set of corrected signal segments is filtered to generate the set of signals from the same interference event. Specifically, a time-frequency similarity threshold is set based on the matching experience of historical interference events. For example, if the time-frequency similarity threshold is set to 0.8, then when the time-frequency similarity is greater than or equal to 0.8, the corrected signal segment of the measurement point is considered similar to the suspected interference time-frequency vector and belongs to the same interference event; when the time-frequency similarity is less than 0.8, the corrected signal segment of the measurement point is considered to have a large difference from the suspected interference time-frequency vector and does not belong to the same interference event. The time-frequency similarity of each measurement point in the interference time-frequency similarity distribution is traversed, and it is determined whether the threshold is reached. If the threshold is reached, the corrected signal segment of the measurement point is filtered out and included in the candidate signal segment set; if the threshold is not reached, the corrected signal segment of the measurement point is removed and not included in the candidate signal segment set. The candidate signal segment set is further validated as follows: The time-frequency similarity between all signal segments in the candidate signal segment set is calculated. If the time-frequency similarity between any two signal segments is greater than or equal to 0.7, then all signal segments in the candidate signal segment set are confirmed to belong to the same interference event. If the time-frequency similarity between two signal segments is less than 0.7, the signal segments with lower similarity are removed until the time-frequency similarity between all signal segments in the candidate signal segment set is greater than or equal to 0.7. The validated candidate signal segment set is then determined as the signal set of the same interference event. All signal segments in the signal set of the same interference event correspond to the same interference event, and the time axes of each signal segment are aligned and their time-frequency characteristics are similar.

[0041] Step S300: Perform joint time-frequency analysis on the set of signals of the same interference event to construct a joint time-frequency feature set that reflects the time-frequency correlation between multiple measurement points.

[0042] Specifically, a joint time-frequency analysis system is constructed and integrated into an edge computing gateway. This system comprises a co-interference event signal input unit, an independent time-frequency transformation unit, a synchronization analysis unit, a mutual information calculation unit, a time-frequency interaction modeling unit, and a feature fusion unit. The set of co-interference event signals is input into the joint time-frequency analysis system. The independent time-frequency transformation unit performs independent time-frequency transformation on each signal segment to extract the time-frequency characteristic information of each signal. The synchronization analysis unit analyzes the synchronization of the time-frequency characteristics of each signal to obtain synchronized time-frequency features. The mutual information calculation unit calculates the mutual information between the time-frequency characteristics of each signal to obtain the time-frequency mutual information distribution. The time-frequency interaction modeling unit constructs a time-frequency interaction model to fuse the time-frequency characteristic information of each signal. Finally, the feature fusion unit generates a joint time-frequency feature set, which comprehensively reflects the time-frequency correlation between multiple measurement points.

[0043] In one possible implementation, joint time-frequency analysis is performed on the set of signals from the same interference events to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measurement points. Step S300 further includes step S310, which involves performing independent time-frequency transformation on each signal in the set of signals from the same interference events to obtain time-frequency feature information of multiple signals. Specifically, a wavelet transform algorithm is selected as the independent time-frequency transformation algorithm. The wavelet transform algorithm uses the db4 wavelet basis, and the decomposition level is set to 5 levels. The wavelet transform processing module built into the edge computing gateway performs independent time-frequency transformation on each signal segment in the set of signals from the same interference events. For each signal segment, the specific transformation process is as follows: The sampled data of the signal segment is input into the wavelet transform processing module, and a 5-level wavelet decomposition is performed to obtain 5 levels of low-frequency components and 5 levels of high-frequency components. Threshold denoising is then performed on the decomposed low-frequency and high-frequency components, with the threshold set to 1.5 times the standard deviation of each component level to remove noise interference from the signal. An inverse wavelet transform is then performed on the denoised low-frequency and high-frequency components to reconstruct the denoised signal segment. Next, the time-frequency feature information of the reconstructed signal segment is extracted, including time-domain and frequency-domain features. Time-domain features include the maximum, minimum, average, variance, kurtosis, skewness, and average short-time energy mutation rate of the signal amplitude. Frequency-domain features include the center frequency, frequency variance, harmonic distortion rate, amplitude of each harmonic, frequency band energy proportion, and center frequency drift. The time-frequency feature information of each signal segment is then organized to form the corresponding signal time-frequency feature information. The time-frequency characteristic information of all signal segments in the same interference event signal set is summarized to obtain multiple signal time-frequency characteristic information, which is stored in an independent time-frequency transformation unit.

[0044] Step S320: Perform synchronization analysis based on the multiple signal time-frequency characteristic information to obtain synchronization time-frequency characteristics. Specifically, determine the characteristic parameters for synchronization analysis, selecting three key parameters from the signal time-frequency characteristic information: center frequency, average signal amplitude, and average short-time energy mutation rate, as the objects of synchronization analysis. These three parameters can accurately reflect the time-frequency synchronization characteristics of the interference signal. For each characteristic parameter, calculate the time series of that parameter for all signal segments in the signal set of the same interference event. The length of the time series is consistent with the number of time nodes of the signal segments. The synchronization coefficients of the time series corresponding to the characteristic parameters of any two signal segments are calculated using the Pearson correlation coefficient. Specifically, the average of the two time series is calculated first. Then, the product of the differences between the two values ​​at each time point and their respective averages is calculated. The sum of all products is then divided by the product of the standard deviations of the two time series and the length of the time series to obtain the Pearson correlation coefficient. The correlation coefficient ranges from -1 to 1. The closer the value is to 1, the stronger the positive synchronization between the two time series; the closer the value is to -1, the stronger the negative synchronization; and a value close to 0 indicates no obvious linear synchronization relationship between the two time series. Next, the average synchronization coefficient of the characteristic parameters corresponding to all signal segments is calculated. The average synchronization coefficient is the average of the synchronization coefficients of all pairwise signal segments. The average synchronization coefficient of the three characteristic parameters, as well as all pairwise synchronization coefficients corresponding to each characteristic parameter, are used as the synchronization time-frequency characteristics. The synchronization time-frequency characteristics also include the maximum value, minimum value, and variance of the synchronization coefficient, used to comprehensively reflect the synchronization of the time-frequency characteristics of signals from multiple measurement points. The synchronization time-frequency characteristics are organized to form a complete set of synchronization time-frequency characteristics, which are then stored in the synchronization analysis unit.

[0045] Step S330: Calculate the mutual information of the multiple signal time-frequency feature information to obtain the time-frequency mutual information distribution. Specifically, determine the feature parameters for mutual information calculation, selecting three key parameters consistent with the synchronization analysis. Each parameter corresponds to a feature variable, and the mutual information is calculated for each feature variable. For each feature variable, calculate the mutual information between the corresponding feature variables of all pairs of signal segments in the signal set of the same interference event. The mutual information is calculated using the histogram estimation method. Specifically, divide the feature variable values ​​of two signal segments into 20 intervals, construct a two-dimensional histogram, count the number of samples in each interval, calculate the joint probability and marginal probability of each interval, and then, according to the mutual information formula, take the logarithm of the ratio of the joint probability to the marginal probability of each interval, multiply it by the joint probability, and sum them to obtain the mutual information between the two feature variables. The value range of the mutual information is non-negative; the larger the value, the stronger the correlation between the two feature variables; the smaller the value, the weaker the correlation between the two feature variables. For each feature variable, the mutual information between all pairs of signal segments is arranged in the order of their combination, forming a mutual information matrix for that feature variable. The mutual information matrix has an n-row, n-column dimension, where n is the number of signal segments in the set of signals from the same interference event, and the matrix elements represent the mutual information between corresponding two signal segments. The mutual information matrices of the three feature variables are then summarized to form a time-frequency mutual information distribution, which reflects the degree of correlation between different feature variables and different signal segments. The time-frequency mutual information distribution is then normalized, mapping all mutual information values ​​to the interval between 0 and 1. The normalization method used is min-max normalization, consistent with the normalization method in step S122. After processing, the normalized time-frequency mutual information distribution is stored in the mutual information calculation unit.

[0046] Step S340: Based on the synchronization time-frequency characteristics and the time-frequency mutual information distribution, perform time-frequency interaction modeling on the set of signals from the same interference events to obtain a time-frequency interaction model. Specifically, construct the framework of the time-frequency interaction model. This model adopts a graph neural network-based model structure. The nodes of the model correspond to each signal segment in the set of signals from the same interference events, i.e., each measurement point. The edges of the model correspond to the time-frequency correlation strength between pairs of signal segments. The correlation strength is jointly determined by the mutual information value in the time-frequency mutual information distribution and the synchronization coefficient in the synchronization time-frequency characteristics. The input parameters of the model include the average synchronization coefficient, the maximum synchronization coefficient, the minimum synchronization coefficient, and the variance of the synchronization coefficient in the synchronization time-frequency characteristics, as well as the normalized mutual information matrix in the time-frequency mutual information distribution. These parameters are organized into the input vector of the model, and the dimension of the input vector matches the number of signal segments and the number of feature parameters. The training process of the model is as follows: Input parameters are fed into the input layer of the graph neural network. Features of nodes and edges are extracted through graph convolutional layers. The kernel size of the graph convolutional layers is set to 3×3, and the activation function is ReLU. Dimensionality reduction of the extracted features is performed through pooling layers, using max pooling with a pooling window size of 2×2. The dimensionality-reduced features are fused through fully connected layers. Two fully connected layers are used: the first has 32 neurons, and the second has 16 neurons. Finally, the parameters of the time-frequency interaction model are output through the output layer. The model parameters include node weights, edge weights, and bias terms. Node weights are determined by synchronous time-frequency features, and edge weights are determined by time-frequency mutual information. The model is trained using the cross-entropy loss function with a learning rate of 0.005 and 500 iterations. The fitting error of the model is calculated after each iteration. Training stops when the fitting error is less than 0.01 and tends to stabilize. After training, a time-frequency interaction model is obtained. This model can accurately describe the time-frequency interaction relationship between signals from multiple measurement points. Based on the time-frequency characteristic information of the input signal, it can output the time-frequency correlation strength between each measurement point and store it in the time-frequency interaction modeling unit.

[0047] Step S350: Based on the time-frequency interaction model, the time-frequency feature information of the multiple signals is fused to generate the joint time-frequency feature set. Specifically, the time-frequency feature information of the multiple signals is input into the trained time-frequency interaction model. The model assigns a corresponding weight to each signal time-frequency feature information according to the time-frequency correlation strength between each measurement point. The higher the correlation strength, the larger the weight; the lower the correlation strength, the smaller the weight. The weight value ranges from 0 to 1, and the sum of the weights of all signal segments is 1. For the feature parameters in each signal time-frequency feature information, a weighted sum is performed. The method is as follows: the final value of each feature parameter is equal to the value of that feature parameter of all signal segments multiplied by the corresponding weight, and then all products are summed. Then, the weighted summation results of the feature parameters are organized to form the basic joint time-frequency features. The average synchronization coefficient, maximum synchronization coefficient, minimum synchronization coefficient, and variance of the synchronization time-frequency features, as well as the average mutual information, maximum mutual information, and minimum mutual information in the time-frequency mutual information distribution, are used as supplementary joint time-frequency features. The basic joint time-frequency features and the supplementary joint time-frequency features are summarized to form a joint time-frequency feature set, which comprehensively reflects the time-frequency correlation between multiple measurement points and is stored in the feature fusion unit.

[0048] Step S400: Extract the cross-measurement point propagation features of the interference signal based on the joint time-frequency feature set, and perform reverse tracing compensation mining based on the cross-measurement point propagation features of the interference signal to obtain the interference tracing compensation vector.

[0049] Specifically, a system for extracting cross-measurement point propagation features and performing reverse tracing compensation for interference signals is constructed. This system is integrated into an edge computing gateway and consists of a joint time-frequency feature input unit, a propagation feature extraction unit, a propagation graph construction unit, a reverse optimization traversal unit, and a compensation parameter configuration unit. The joint time-frequency feature set is input to the propagation feature extraction unit to extract the cross-measurement point propagation features of the interference signal; the propagation features are input to the propagation graph construction unit to construct a weighted directed graph of interference propagation; the weighted directed graph of interference propagation is input to the reverse optimization traversal unit to determine the interference path tracing result; the interference path tracing result is input to the compensation parameter configuration unit to generate an interference tracing compensation vector, which reflects the propagation path of the interference signal and the compensation requirements.

[0050] In one possible implementation, reverse tracing compensation mining is performed based on the cross-measurement point propagation characteristics of the interference signal to obtain the interference tracing compensation vector. Step S400 further includes step S410, where the cross-measurement point propagation characteristics of the interference signal include interference arrival time sequence characteristics, interference energy propagation attenuation characteristics, interference frequency offset consistency, and interference signal correlation strength matrix. Specifically, the interference arrival time sequence characteristics are extracted, which are the time order and time interval of the interference signal arriving at each measurement point. Specifically, from the set of signals of the same interference event, the interference start time of each signal segment is extracted, and the interference start times of each measurement point are arranged in ascending order to obtain the interference arrival time sequence. The difference between the interference start times of two adjacent measurement points is calculated to obtain the time interval. The time sequence and time interval are summarized to form the interference arrival time sequence characteristics. The interference energy propagation attenuation characteristics are extracted. These characteristics are the ratio of the interference signal energy at each measuring point to the interference signal energy at the reference measuring point, as well as the energy attenuation rate. Specifically, the measuring point with the earliest interference arrival time is used as the reference measuring point. The total energy of the interference signal at the reference measuring point is extracted, and then the total energy of the interference signals at other measuring points is extracted. The ratio of the total energy at each measuring point to the total energy at the reference measuring point is calculated. This ratio is the energy attenuation coefficient. The closer the ratio is to 1, the smaller the interference energy attenuation at that measuring point. The smaller the ratio, the greater the attenuation. The energy attenuation rate is calculated based on the interference arrival time sequence. The difference in the energy attenuation coefficient between two adjacent measuring points is calculated and then divided by the interference arrival time interval between the two adjacent measuring points to obtain the energy attenuation rate per unit time. The interference frequency offset consistency is extracted to reflect the degree of synchronization of the interference signal frequency offset at each measurement point. The method is as follows: extract the center frequency drift during the interference period from the time-frequency characteristic information of each signal segment, calculate the standard deviation of the center frequency drift of all measurement points. The smaller the standard deviation, the stronger the frequency offset consistency of each measurement point; the larger the standard deviation, the weaker the consistency. At the same time, calculate the deviation between the center frequency drift of each measurement point and the average center frequency drift of all measurement points. The smaller the absolute value of the deviation, the more consistent the frequency offset of that measurement point is with the overall offset trend. The interference signal correlation strength matrix is ​​extracted and constructed based on the time-frequency mutual information distribution. The method involves using the normalized mutual information matrix from the time-frequency mutual information distribution as a foundation, and combining it with the synchronization coefficient from the synchronization time-frequency characteristics to perform a weighted correction. The correction formula is: Correlation Strength = 0.6 × Normalized Mutual Information + 0.4 × Synchronization Coefficient, where 0.6 and 0.4 are weighting coefficients, determined through historical data verification to ensure that the correlation strength simultaneously reflects the correlation degree and synchronization of the time-frequency characteristics. The corrected interference signal correlation strength matrix is ​​obtained; larger element values ​​indicate a stronger correlation between the interference signals at the corresponding two measurement points.The arrival time characteristics of interference, the attenuation characteristics of interference energy propagation, the consistency of interference frequency offset, and the correlation strength matrix of interference signal are summarized to form a complete cross-measurement point propagation characteristic of interference signal, which is stored in the propagation characteristic extraction unit.

[0051] Step S420: Using the installation location topology of the multiple energy meters as the node reference, the interference arrival time sequence characteristics as the edge weight reference, and the interference energy propagation attenuation characteristics as the node weight reference, a weighted directed graph for interference propagation is constructed. Specifically, the node reference of the weighted directed graph for interference propagation is determined. Taking the installation location topology of the multiple energy meters as the core, the actual installation coordinates of all measuring points (energy meters) in the power distribution area are obtained. Combined with the power distribution line routing and equipment layout, a measuring point topology structure is constructed. Each measuring point is a node in the graph. The node identifier corresponds one-to-one with the energy meter measuring point number. The node attributes include the installation coordinates of the measuring point, the equipment model, and the corresponding interference energy propagation attenuation characteristics. The node weight directly adopts the energy attenuation coefficient of the measuring point. The closer the attenuation coefficient is to 1, the larger the node weight, indicating that the interference energy attenuation of the measuring point is smaller and closer to the source of interference propagation. The edges and edge weights of the graph are determined based on the interference arrival time sequence characteristics. Combined with the topology of the measurement points, the propagation direction of the interference signal is determined: if the interference arrival time at measurement point A is earlier than that at measurement point B, and the two points are directly connected by a power distribution line in the topology, a directed edge is constructed from measurement point A to measurement point B, representing the interference signal propagating from A to B. The edge weights are based on the time interval in the interference arrival time sequence characteristics and are normalized using the same method as steps S122 and S330, mapping to the 0-1 interval. The smaller the time interval, the larger the edge weight, indicating that the interference signal propagates faster and has a stronger correlation along that path. The node weights and edge weights are integrated to optimize the constructed directed graph: weakly correlated edges with weights less than a preset threshold are removed to avoid invalid path interference; the energy attenuation rate information of the nodes is added as an auxiliary attribute. The constructed interference propagation weighted directed graph is stored in the propagation graph construction unit. This graph presents the topological relationships of the measurement points, the direction of interference propagation, the energy attenuation characteristics of the nodes, and the propagation correlation of the edges.

[0052] Step S430: A reverse optimization traversal is performed on the interference propagation weighted directed graph based on the interference frequency offset consistency and the interference signal correlation strength matrix to determine the interference path tracing result. Specifically, based on the interference propagation weighted directed graph, and combined with the interference frequency offset consistency and the interference signal correlation strength matrix, bidirectional constraints are constructed to ensure the accuracy of the tracing path. The starting node for reverse optimization is set to the measurement point with the latest interference arrival time, i.e., the end node of interference propagation. If there are multiple latest arrival measurement points, the measurement point with the smallest node weight is selected as the starting node, and the tracing proceeds backward from the end node to the source of interference propagation. During the traversal, edge weight, interference frequency offset consistency deviation, and interference signal correlation strength are used as comprehensive evaluation indicators, with evaluation weight percentages set, for example, interference signal correlation strength 50%, edge weight 30%, and frequency offset consistency deviation 20%. The higher the comprehensive score, the stronger the rationality of the reverse path. The path with the highest comprehensive score is selected for tracing until a node without a preceding propagation path is found, initially forming candidate results for interference path tracing. Further filtering and optimization are then performed in conjunction with steps S431-S433.

[0053] Step S440: Based on the interference path tracing results, compensation parameters are configured to generate the interference tracing compensation vector. Specifically, taking the interference path tracing results as the core, and combining the interference energy propagation attenuation characteristics and interference frequency offset consistency extracted in step S410, as well as the time-frequency characteristic information of each measurement point signal extracted in step S310, targeted compensation parameters are configured. The compensation parameters mainly include three categories: time compensation parameters, energy compensation parameters, and frequency compensation parameters, which correspond to the time deviation, energy attenuation, and frequency offset generated during interference propagation, respectively, to ensure that the timing of each measurement point signal is aligned and the characteristics are consistent after compensation. The various compensation parameters are configured as follows: Time compensation parameters are based on the time interval in the interference arrival time sequence characteristics, using the interference start time at the interference propagation origin as the benchmark. The time compensation amount for each measurement point is equal to the difference between the interference arrival time at that measurement point and the origin. Energy compensation parameters are based on the interference energy propagation attenuation characteristics, using the total interference energy at the origin as the benchmark. The energy compensation coefficient for each measurement point is equal to 1 divided by the energy attenuation coefficient at that measurement point. During compensation, the amplitude of the interference signal at that measurement point is amplified according to the compensation coefficient to offset the energy attenuation. Frequency compensation parameters are based on the consistency of interference frequency offset, using the average value of the center frequency drift of all measurement points as the benchmark. The frequency compensation amount for each measurement point is equal to the difference between the center frequency drift at that measurement point and the average value. During compensation, the center frequency of the interference signal at that measurement point is calibrated to the average value to eliminate the frequency offset. These three types of compensation parameters are integrated to form a compensation vector for each measurement point. The compensation vector has a dimension of 1 row and 3 columns, corresponding sequentially to the time compensation parameter, energy compensation parameter, and frequency compensation parameter. The compensation vectors for all measurement points are arranged in order of measurement point number, forming an interference tracing compensation vector stored in the compensation parameter configuration unit.

[0054] In one possible implementation, the interference propagation weighted directed graph is traversed in reverse optimization based on the interference frequency offset consistency and the interference signal correlation strength matrix to determine the interference path tracing result. Step S430 further includes step S431, which involves performing a reverse search based on the interference propagation weighted directed graph to obtain a set of searched interference paths. Specifically, the measurement point with the latest interference arrival time is taken as the starting node for the reverse search. The search direction is opposite to the edge direction of the interference propagation weighted directed graph, i.e., searching from the end node to the starting node. Starting from the starting node, all directed edges pointing to that node are traversed, and the predecessor node corresponding to each edge is recorded, i.e., the node before which the interference propagated to the current node. The process is repeated until a candidate starting point without a predecessor node is reached. For each end starting node, there may be multiple reverse propagation paths, and each one is searched and recorded. All the reverse paths obtained from the search are sorted out, and duplicate and invalid paths are removed. The remaining valid reverse paths are summarized to form a set of search interference paths. Each path contains information such as the node order, the energy attenuation coefficient of each node, the weight of each edge and the propagation time interval, which are stored in the reverse optimization traversal unit.

[0055] Step S432: Perform frequency consistency matching on the search interference path set based on the interference frequency offset consistency to obtain a candidate interference path set. Specifically, extract the center frequency drift of all nodes (measurement points) of each path in the search interference path set. This data is directly retrieved from the interference frequency offset consistency features extracted in step S410. Set a frequency consistency matching threshold. For example, based on historical interference data verification, set the standard deviation threshold of the center frequency drift of all measurement points to 0.2Hz, and set a single node deviation threshold, such as the absolute value of the deviation between the center frequency drift of each measurement point and the average value ≤ 0.1Hz. If both thresholds are met, the frequency consistency is considered to be up to standard. Perform frequency consistency verification on each search interference path: calculate the standard deviation of the center frequency drift of all nodes in the path. If the standard deviation is ≤ 0.2 Hz, and the absolute value of the deviation between the center frequency drift of each node and the average value of the drift of all measurement points is ≤ 0.1 Hz, then the path passes the frequency consistency matching; if any one of them is not met, the path is removed. All paths that pass frequency consistency matching are aggregated to form a candidate interference path set. This ensures that the interference frequency offsets of each measurement point in the candidate path have strong synchronization, and eliminates erroneous paths caused by excessive frequency offset differences.

[0056] Step S433: The candidate interference path set is weighted and filtered based on the time-frequency characteristic intensity of the interference signal correlation strength matrix to obtain the interference path tracing results. Specifically, the correlation strength values ​​(matrix elements) corresponding to each adjacent node of each path in the candidate interference path set are extracted from the interference signal correlation strength matrix. The larger the correlation strength value, the stronger the correlation of the interference signals between adjacent nodes, and the higher the rationality of the path. A weighted screening evaluation system is constructed, using the average correlation strength of all adjacent nodes in the path as the core evaluation index. Combined with the comprehensive score set in step S430, weighting coefficients are assigned to both indicators. The weighting coefficients are determined through verification using historical interference data to ensure that the screening results both conform to the time-frequency characteristic correlation strength and take into account the overall rationality of the path. A weighted screening threshold is also set: the weighted comprehensive score ≥ the weighted comprehensive score threshold, and the average correlation strength ≥ the average correlation strength threshold. A path is considered qualified if both conditions are met simultaneously. Weighted screening is performed on each path in the candidate interference path set: first, the average correlation strength of adjacent nodes of each path is calculated, and then, combined with the comprehensive score of the path, a weighted comprehensive score is calculated using a weighting formula. If multiple qualified paths exist, the path with the highest weighted comprehensive score and the highest average correlation strength is selected as the optimal path; if only one qualified path exists, that path is the optimal path. The selected optimal path is determined as the interference path tracing result, which includes the interference propagation starting point, the complete propagation path, the energy attenuation characteristics of each node, the correlation strength of each adjacent node, and the propagation time interval, and is stored in the reverse optimization traversal unit.

[0057] Step S500: Perform anti-interference reconstruction on the set of signals of the same interference event according to the interference tracing compensation vector.

[0058] Specifically, an anti-interference reconstruction system is built, integrated into the edge computing gateway, and linked with the time correction unit in step S200 and the joint time-frequency analysis system in step S300. This system consists of a compensation vector input unit, a signal reconstruction unit, and a reconstruction verification unit. It can receive interference tracing compensation vectors and sets of signals from the same interference event, and complete targeted anti-interference reconstruction. The anti-interference reconstruction process is as follows: The interference tracing compensation vector is input to the compensation vector input unit. The signal reconstruction unit calls each signal segment from the set of signals from the same interference event. Based on the time compensation parameters of the corresponding measurement points in the compensation vector, all timestamps of the signal segments are calibrated to ensure that the time axis of all measurement point signals is completely aligned with the time axis of the interference propagation starting signal, eliminating time deviations during propagation. Based on the energy compensation parameters, the amplitude of each sampling point of the signal segment is amplified, and the total signal energy is adjusted according to the compensation coefficient to offset the energy attenuation during interference propagation. Based on the frequency compensation parameters, the center frequency of the signal segment is calibrated to eliminate frequency offsets and ensure that the frequency of the interference signals at each measurement point is synchronized. Then, the reconstructed signal set is verified, including time synchronization error, energy consistency, frequency consistency, and signal time-frequency characteristic similarity. If all verification indicators meet the requirements, the anti-interference reconstruction is considered complete; if any indicators are not met, the process returns to step S440 to adjust the compensation parameters and reconstruct and verify. Finally, the reconstructed multi-measurement point synchronization signal set is stored in the database of the edge computing gateway, and the reconstruction results are transmitted to the power distribution monitoring center for subsequent interference prevention, equipment operation and maintenance, and other tasks.

[0059] This application embodiment acquires sampling signal groups from multiple energy meters within a power distribution area. Based on a clock synchronization mechanism, it performs time-series correlation-based anomaly detection to identify suspected interference events. It extracts multi-measurement point signal segments corresponding to these events, and obtains a set of signals for the same interference event through cross-measurement point event matching and time deviation correction. Joint time-frequency analysis is performed on this signal set to construct a joint time-frequency feature set reflecting the time-frequency correlation of multiple measurement points. Based on this feature set, the cross-measurement point propagation characteristics of the interference signal are extracted. An interference tracing compensation vector is obtained through reverse tracing compensation mining. This compensation vector is then used to reconstruct the signal set for the same interference event, completing the processing and optimization of the interference signal. These techniques solve the technical problems of low interference identification accuracy and insufficient source tracing capability in existing energy meter interference signal analysis and processing, achieving the technical effects of improving interference signal identification accuracy and realizing accurate tracing of interference sources and propagation paths.

[0060] In the above text, refer to Figure 1 This paper describes in detail a method for extracting the time-frequency features of interference signals using electricity meter clock synchronization according to an embodiment of the present invention. Next, reference will be made to... Figure 2 A system for extracting time-frequency features of interference signals using electricity meter clock synchronization is described according to an embodiment of the present invention.

[0061] The interference signal time-frequency feature extraction system utilizing electricity meter clock synchronization according to embodiments of the present invention addresses the technical problems of low interference identification accuracy and insufficient source tracing capability in existing electricity meter interference signal analysis and processing, achieving the technical effect of improving interference signal identification accuracy and realizing accurate source tracing of interference sources and propagation paths. The interference signal time-frequency feature extraction system utilizing electricity meter clock synchronization includes: an abnormal trigger detection module 10, a cross-measuring point event matching module 20, a joint time-frequency analysis module 30, a reverse tracing compensation mining module 40, and an anti-interference reconstruction module 50.

[0062] An anomaly trigger detection module 10 is used to acquire multiple sampling signal groups from multiple energy meters within the power distribution area, and perform anomaly trigger detection on the multiple sampling signal groups under time-series correlation according to the clock synchronization mechanism to determine suspected interference events. A cross-measuring point event matching module 20 is used to extract multi-measuring point signal segments corresponding to the suspected interference events, and perform cross-measuring point event matching and time deviation correction on the multi-measuring point signal segments to obtain a set of signals with the same interference events. A joint time-frequency analysis module 30 is used to perform joint time-frequency analysis on the set of signals with the same interference events to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measuring points. A reverse tracing compensation mining module 40 is used to extract cross-measuring point propagation features of interference signals based on the joint time-frequency feature set, and perform reverse tracing compensation mining based on the cross-measuring point propagation features of interference signals to obtain an interference tracing compensation vector. An anti-interference reconstruction module 50 is used to perform anti-interference reconstruction on the set of signals with the same interference events based on the interference tracing compensation vector.

[0063] The specific configuration of the anomaly trigger detection module 10 is described in detail below: As mentioned above, the anomaly trigger detection module 10 performs time-related anomaly trigger detection on the multiple sampled signal groups according to the clock synchronization mechanism to determine suspected interference events. The anomaly trigger detection module 10 may further include: a unified time alignment unit for performing unified time alignment on the multiple sampled signal groups according to the clock synchronization mechanism to obtain a multi-measurement point aligned signal set; a time-related fluctuation anomaly detection unit for performing time-related fluctuation anomaly detection on the multi-measurement point aligned signal set to determine suspected interference trigger points; a front and back node expansion unit for expanding the suspected interference trigger points with front and back nodes according to the multi-measurement point aligned signal set to establish a suspected interference trigger window; and an interference event feature capture unit for capturing interference event features on the multi-measurement point aligned signal set according to the suspected interference trigger window to generate the sampled suspected interference event.

[0064] The process of detecting temporal correlation fluctuation anomalies in the multi-measurement point aligned signal set to identify suspected interference trigger points can be further comprised of: a time-frequency fluctuation index acquisition subunit for acquiring time-frequency fluctuation indices, including time-domain fluctuation indices and frequency-domain offset indices; a time-frequency fluctuation analysis subunit for performing time-frequency fluctuation analysis on the multi-measurement point aligned signal set based on the time-frequency fluctuation indices to acquire time-frequency fluctuation matrices for each signal; an interference suspected detection network training subunit for training an interference suspected detection network based on a set of historical interference events from the energy meter; an interference suspected detection subunit for inputting the time-frequency fluctuation matrices of each signal into the interference suspected detection network to acquire multiple interference suspected values; and a suspected interference signal determination subunit for determining suspected interference signals based on the multiple interference suspected values ​​and outputting the multi-dimensional signal feature parameters of the suspected interference signals as the suspected interference trigger points.

[0065] The detailed description of the cross-measurement point event matching module 20 is as follows: As mentioned above, the cross-measurement point event matching module 20 performs cross-measurement point event matching and time deviation correction on the multi-measurement point signal segments to obtain a set of signals with the same interference event. The cross-measurement point event matching module 20 may further include: a time deviation characteristic analysis unit for performing time deviation characteristic analysis on the multi-measurement point signal segments based on the sampled suspected interference events to obtain multi-dimensional time deviation characteristics, including clock residual deviation, sampling delay deviation, and communication timestamp deviation; a time compensation and correction unit for performing time compensation and correction on the multi-dimensional time deviation characteristics to obtain a set of corrected signal segments; a time-frequency similarity analysis unit for constructing a suspected interference time-frequency vector based on the sampled suspected interference events, and performing time-frequency similarity analysis on the set of corrected signal segments based on the suspected interference time-frequency vector to obtain an interference time-frequency similarity distribution; and a filtering unit for filtering the set of corrected signal segments based on the interference time-frequency similarity distribution to generate the set of signals with the same interference event.

[0066] The detailed description of the specific configuration of the joint time-frequency analysis module 30 is explained as follows: As described above, joint time-frequency analysis is performed on the set of signals from the same interference event to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measurement points. The joint time-frequency analysis module 30 may further include: a signal-independent time-frequency transformation unit for performing independent time-frequency transformation on the set of signals from the same interference event to obtain time-frequency feature information of multiple signals; a synchronization analysis unit for performing synchronization analysis based on the time-frequency feature information of multiple signals to obtain synchronization time-frequency features; a mutual information calculation unit for calculating mutual information on the time-frequency feature information of multiple signals to obtain the time-frequency mutual information distribution; a time-frequency interaction modeling unit for performing time-frequency interaction modeling on the set of signals from the same interference event based on the synchronization time-frequency features and the time-frequency mutual information distribution to obtain a time-frequency interaction model; and a fusion unit for fusing the time-frequency feature information of multiple signals based on the time-frequency interaction model to generate the joint time-frequency feature set.

[0067] The detailed description of the specific configuration of the reverse tracing compensation mining module 40 is as follows: As mentioned above, the reverse tracing compensation mining is performed based on the cross-measurement point propagation characteristics of the interference signal to obtain the interference tracing compensation vector. The reverse tracing compensation mining module 40 may further include: an interference signal cross-measurement point propagation characteristic construction unit for constructing interference signal cross-measurement point propagation characteristics, which include interference arrival time sequence characteristics, interference energy propagation attenuation characteristics, interference frequency offset consistency, and interference signal correlation strength matrix; an interference propagation weighted directed graph construction unit for constructing an interference propagation weighted directed graph based on the installation location topology of the multiple energy meters as node references, the interference arrival time sequence characteristics as edge weight references, and the interference energy propagation attenuation characteristics as node weight references; a reverse optimization traversal unit for performing a reverse optimization traversal on the interference propagation weighted directed graph based on the interference frequency offset consistency and the interference signal correlation strength matrix to determine the interference path tracing result; and a compensation parameter configuration unit for configuring compensation parameters based on the interference path tracing result to generate the interference tracing compensation vector.

[0068] Specifically, the interference path tracing result is determined by performing a reverse optimization traversal on the interference propagation weighted directed graph based on the interference frequency offset consistency and the interference signal correlation strength matrix. The reverse optimization traversal unit may further include: a reverse search subunit for performing a reverse search on the interference propagation weighted directed graph to obtain a set of search interference paths; a frequency consistency matching subunit for performing frequency consistency matching on the set of search interference paths based on the interference frequency offset consistency to obtain a set of candidate interference paths; and a time-frequency feature strength weighted filtering subunit for performing time-frequency feature strength weighted filtering on the set of candidate interference paths based on the interference signal correlation strength matrix to obtain the interference path tracing result.

[0069] The abnormal trigger detection module 10 may further include: the clock synchronization mechanism includes synchronizing the clocks of the multiple energy meters to a standard time source through a network time protocol.

[0070] The time-frequency fluctuation index acquisition subunit may further include: the time-domain fluctuation index includes amplitude kurtosis, pulse interval deviation, and short-time energy mutation rate; the frequency-domain offset index includes center frequency drift, harmonic distortion rate, and frequency band energy ratio change rate.

[0071] The interference signal time-frequency feature extraction system using electricity meter clock synchronization provided in this embodiment of the invention can execute the interference signal time-frequency feature extraction method using electricity meter clock synchronization provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0072] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0073] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for extracting the time-frequency features of interference signals from electricity meter clock synchronization, characterized in that, The method includes: Multiple sampling signal groups from multiple energy meters within the power distribution area are acquired, and abnormal trigger detection under time-series correlation is performed on the multiple sampling signal groups according to the clock synchronization mechanism to determine suspected sampling interference events; Extract the multi-measurement point signal segments corresponding to the suspected interference events, and perform cross-measurement point event matching and time deviation correction on the multi-measurement point signal segments to obtain the signal set of the same interference event; Joint time-frequency analysis is performed on the set of signals from the same interference events to construct a joint time-frequency feature set that reflects the time-frequency correlation between multiple measurement points; Based on the joint time-frequency feature set, the cross-measurement point propagation features of the interference signal are extracted, and the reverse tracing compensation mining is performed based on the cross-measurement point propagation features of the interference signal to obtain the interference tracing compensation vector. The interference tracing compensation vector is used to reconstruct the set of signals from the same interference event in an anti-interference manner.

2. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 1, characterized in that, Based on the clock synchronization mechanism, anomaly trigger detection is performed on the multiple sampling signal groups under time-correlation to determine suspected sampling interference events, including: The multiple sampling signal groups are aligned in a unified time according to the clock synchronization mechanism to obtain a multi-measurement point aligned signal set. The time-series correlation fluctuation anomaly detection is performed on the multi-measurement point aligned signal set to identify suspected interference trigger points; Based on the multi-measurement point aligned signal set, the suspected interference trigger point is expanded with front and rear nodes to establish a suspected interference trigger window; Based on the suspected interference trigger window, the interference event features of the multi-measurement point aligned signal set are captured to generate the sampled suspected interference event.

3. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 2, characterized in that, The time-series correlation fluctuation anomaly detection is performed on the multi-measurement point aligned signal set to identify suspected interference trigger points, including: Obtain time-frequency fluctuation indicators, which include time-domain fluctuation indicators and frequency-domain offset indicators; Based on the time-frequency fluctuation index, the time-frequency fluctuation of the multi-measurement point aligned signal set is analyzed to obtain the time-frequency fluctuation matrix of each signal. Train a suspected interference detection network based on the historical event set of interference from electricity meters; The time-frequency fluctuation matrices of each signal are input into the interference suspicion detection network to obtain multiple interference suspicion scores. Based on the multiple interference suspicion levels, a suspected interference signal is determined, and the multidimensional signal characteristic parameters of the suspected interference signal are output as the suspected interference trigger point.

4. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 1, characterized in that, Perform cross-measurement point event matching and time deviation correction on the multi-measurement point signal segments to obtain a set of co-interference event signals, including: Based on the suspected interference events in the sampling, the time deviation characteristics of the multi-measurement point signal segments are analyzed to obtain multi-dimensional time deviation characteristics, which include clock residual deviation, sampling delay deviation, and communication timestamp deviation. Based on the multidimensional time deviation characteristics, time compensation and correction are performed on the multi-measurement point signal segments to obtain a set of corrected signal segments; Based on the sampled suspected interference events, a suspected interference time-frequency vector is constructed, and a time-frequency similarity analysis is performed on the corrected signal segment set based on the suspected interference time-frequency vector to obtain the interference time-frequency similarity distribution; The set of corrected signal segments is selected based on the time-frequency similarity distribution of the interference to generate the set of signals of the same interference event.

5. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 1, characterized in that, Joint time-frequency analysis is performed on the set of signals from the same interference events to construct a joint time-frequency feature set reflecting the time-frequency correlation between multiple measurement points, including: Perform independent time-frequency transformation on each signal in the set of signals from the same interference event to obtain time-frequency characteristic information of multiple signals; Synchronization analysis is performed based on the time-frequency characteristic information of the multiple signals to obtain synchronization time-frequency characteristics; The mutual information of the time-frequency characteristics of the multiple signals is calculated to obtain the time-frequency mutual information distribution; Based on the synchronous time-frequency characteristics and the time-frequency mutual information distribution, the time-frequency interaction model of the same interference event signal set is performed to obtain the time-frequency interaction model. The joint time-frequency feature set is generated by fusing the time-frequency feature information of the multiple signals according to the time-frequency interaction model.

6. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 1, characterized in that, Based on the cross-measurement point propagation characteristics of the interference signal, reverse tracing compensation mining is performed to obtain the interference tracing compensation vector, including: The cross-measurement point propagation characteristics of the interference signal include interference arrival timing characteristics, interference energy propagation attenuation characteristics, interference frequency offset consistency, and interference signal correlation strength matrix. Using the installation location topology of the multiple energy meters as the node reference, the arrival time sequence characteristics of the interference as the edge weight reference, and the attenuation characteristics of the interference energy propagation as the node weight reference, a weighted directed graph of interference propagation is constructed. Based on the consistency of the interference frequency offset and the correlation strength matrix of the interference signal, a reverse optimization traversal is performed on the weighted directed graph of interference propagation to determine the interference path tracing result; Based on the interference path tracing results, compensation parameters are configured to generate the interference tracing compensation vector.

7. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 6, characterized in that, Based on the interference frequency offset consistency and the interference signal correlation strength matrix, a reverse optimization traversal is performed on the interference propagation weighted directed graph to determine the interference path tracing result, including: A reverse search is performed based on the weighted directed graph of interference propagation to obtain the set of interference paths. Based on the frequency offset consistency, the search interference path set is matched for frequency consistency to obtain a candidate interference path set; The candidate interference path set is filtered by time-frequency feature intensity weighting based on the interference signal correlation strength matrix to obtain the interference path tracing result.

8. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 1, characterized in that, The clock synchronization mechanism includes synchronizing the clocks of the multiple energy meters to a standard time source via a network time protocol.

9. The method for extracting time-frequency features of interference signals using electricity meter clock synchronization as described in claim 3, characterized in that, The time-domain fluctuation indicators include amplitude kurtosis, pulse interval deviation, and short-time energy mutation rate; The frequency domain offset indicators include center frequency drift, harmonic distortion rate, and frequency band energy percentage change rate.

10. A system for extracting the time-frequency features of interference signals synchronized with an electricity meter clock, characterized in that, The system is used to implement the time-frequency feature extraction method for interference signals using the clock synchronization of an electricity meter as described in any one of claims 1-9, and the system comprises: An abnormal trigger detection module is used to acquire multiple sampling signal groups from multiple energy meters in the power distribution area, and to perform abnormal trigger detection on the multiple sampling signal groups under time-series correlation according to the clock synchronization mechanism to determine suspected sampling interference events. The cross-measurement point event matching module is used to extract the multi-measurement point signal segments corresponding to the sampled suspected interference events, and perform cross-measurement point event matching and time deviation correction on the multi-measurement point signal segments to obtain the signal set of the same interference event. The joint time-frequency analysis module is used to perform joint time-frequency analysis on the set of signals of the same interference event, and construct a joint time-frequency feature set that reflects the time-frequency correlation between multiple measurement points; The reverse tracing compensation mining module is used to extract the cross-measurement point propagation features of the interference signal based on the joint time-frequency feature set, and to perform reverse tracing compensation mining based on the cross-measurement point propagation features of the interference signal to obtain the interference tracing compensation vector. An anti-interference reconstruction module is used to reconstruct the set of signals of the same interference event based on the interference tracing compensation vector.

Images