Circuit abnormity identification method and system of three-phase electric meter

Abstract

The invention relates to the technical field of data identification, in particular to a circuit abnormity identification method and system for a three-phase ammeter, and the method comprises the steps: carrying out the Fourier transform of an acquired historical feature sequence, and obtaining a frequency spectrum sequence, calculating a time domain difference and a frequency domain difference according to the historical normal sequence, the abnormal feature sequence and the frequency spectrum sequence corresponding to the three phases in the same load state, and calculating the weights of the time domain and the frequency domain through the time domain difference and the frequency domain difference; calculating the single-phase abnormal degree of the target feature by using the time domain weight and the frequency domain weight of the target feature in each load state; and inputting the single-phase abnormal degree of each target feature on each phase into a control system, and performing circuit abnormality identification of the three-phase electric meter according to a preset abnormal degree threshold. Therefore, the defect that the false alarm rate is high when a single time-frequency domain threshold value and single moment data ignore global change for abnormal recognition is overcome through the data recognition technology.

Description

Technical Field

[0001] This application relates to the field of data recognition technology, and in particular to a method and system for identifying circuit anomalies in a three-phase meter. Background Technology

[0002] Three-phase meters are widely used in industrial, commercial, and residential settings to monitor and record electricity consumption in real time. Traditional methods primarily utilize conventional parameters such as voltage, current, and power factor for circuit monitoring, relying on fixed threshold judgments, waveform analysis, and simple rule engines. Certain thresholds or tolerance ranges are set, and anything exceeding the set value is considered an anomaly.

[0003] However, rule engines based on a single time-frequency domain cannot adapt to dynamic load fluctuations, such as motor start-up and shutdown, and transient photovoltaic power, resulting in a high false alarm rate. For example, the instantaneous current surge during motor startup is often misjudged as overload, while gradual overloads caused by equipment aging are ignored because they do not trigger thresholds. Therefore, processing existing technologies through data recognition techniques, i.e., anomaly identification based on a single time-frequency domain feature threshold, suffers from the drawback of a high false alarm rate. Summary of the Invention

[0004] To address the high false alarm rate associated with using a single time-frequency domain threshold and ignoring global changes in data at a single moment for anomaly identification, this application provides a method and system for circuit anomaly identification of a three-phase meter.

[0005] Firstly, this application provides a method for identifying circuit anomalies in a three-phase electricity meter, employing the following technical solution: A method for identifying circuit anomalies in a three-phase electricity meter includes the following steps: Collect historical normal signals, abnormal characteristic signals, and unidentified characteristic signals for each corresponding three-phase meter under different load conditions; Fourier transform is used to obtain the spectrum sequence of the collected historical feature sequence. The time domain difference and frequency domain difference are calculated based on the three corresponding historical normal sequence, abnormal feature sequence and spectrum sequence under the same load condition. The weights of the time domain and frequency domain are calculated through the time domain difference and the frequency domain difference. The feature sequence to be identified is divided using a sliding window. The time-domain and frequency-domain anomaly levels are calculated based on the time-frequency domain features of the target window and neighboring windows. The single-phase anomaly level of the target feature is calculated using the time-domain and frequency-domain weights of the target feature under each load condition. The single-phase anomaly level of each target feature on each phase is input into the control system, and the circuit anomaly of the three-phase meter is identified according to the preset anomaly level threshold. The calculation of the single-phase anomaly degree includes: using Fourier transform to obtain the spectrum sequence corresponding to each window of the sliding window; calculating the frequency domain anomaly degree based on the spectrum sequences of the window in the previous time step and the neighboring windows; and calculating the single-phase anomaly degree of the target feature based on the time domain anomaly degree and the frequency domain anomaly degree of the target feature, so as to obtain the single-phase anomaly degree of each target feature in each phase.

[0006] Optionally, voltage and current transformers can be used to collect historical normal signals, abnormal characteristic signals, and characteristic signals to be identified for each corresponding three-phase meter under different load conditions.

[0007] Optionally, wavelet denoising can be used to denoise the historical normal signals, abnormal feature signals, and feature signals to be identified for each corresponding three-phase meter under different load conditions.

[0008] Optionally, the historical normal signals, abnormal characteristic signals, and unidentified characteristic signals of each corresponding three-phase meter under different load conditions can be normalized to eliminate dimensional differences.

[0009] Optionally, calculating the weights in the time and frequency domains includes: Fourier transform is used to obtain the spectral sequence corresponding to each characteristic sequence for the three-phase historical normal and abnormal characteristic sequences under different load conditions; The time-domain and frequency-domain differences of the target characteristics under each load condition are calculated based on the characteristic sequences of the three-phase historical normal and abnormal phases under different load conditions. The weights in the time and frequency domains are calculated based on the time-domain and frequency-domain differences of the target characteristics under the same load condition.

[0010] Secondly, this application provides a circuit anomaly identification system for a three-phase electricity meter, employing the following technical solution: A circuit anomaly identification system for a three-phase electricity meter, employing the aforementioned circuit anomaly identification method for a three-phase electricity meter, includes: The signal acquisition module is used to acquire historical normal signals, abnormal characteristic signals, and characteristic signals to be identified for each corresponding three-phase meter under different load conditions; The weight calculation module is used to obtain the spectrum sequence by Fourier transform of the collected historical feature sequence, calculate the time domain difference and frequency domain difference based on the three corresponding historical normal sequence, abnormal feature sequence and spectrum sequence under the same load state, and calculate the weight of the time domain and frequency domain through the time domain difference and the frequency domain difference; The single-phase anomaly degree calculation module is used to divide the feature sequence to be identified using a sliding window, calculate the time-domain and frequency-domain anomaly degree based on the time-frequency domain features of the target window and neighboring windows, and calculate the single-phase anomaly degree of the target feature using the time-domain and frequency-domain weights of the target feature under each load state. The anomaly identification module is used to input the single-phase anomaly degree of each target feature on each phase into the control system, and to identify circuit anomalies of the three-phase meters according to the preset anomaly degree threshold. The calculation of the single-phase anomaly degree includes: using Fourier transform to obtain the spectrum sequence corresponding to each window of the sliding window; calculating the frequency domain anomaly degree based on the spectrum sequences of the window in the previous time step and the neighboring windows; and calculating the single-phase anomaly degree of the target feature based on the time domain anomaly degree and the frequency domain anomaly degree of the target feature, so as to obtain the single-phase anomaly degree of each target feature in each phase.

[0011] This application achieves the following technical effects: It uses Fourier transform to obtain the spectral sequence of the collected historical feature sequences; calculates the time-domain and frequency-domain differences based on the three-phase historical normal and abnormal feature sequences and spectral sequences under the same load condition; and calculates the weights of the time and frequency domains based on these differences. It divides the feature sequence to be identified using a sliding window; calculates the time-domain and frequency-domain anomaly levels based on the time-frequency domain features of the target window and neighboring windows; and calculates the single-phase anomaly level of the feature using its time-domain and frequency-domain weights under each load condition. Adaptive weight fusion of time-frequency domain features avoids high false alarm rates in single time-frequency domains, and time series analysis avoids false alarms caused by instantaneous changes in features at a single moment. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating steps S1-S4 of a circuit anomaly identification method for a three-phase meter according to this application. Detailed Implementation

[0013] This application discloses a method for identifying circuit anomalies in a three-phase electricity meter, referring to... Figure 1 This includes the following steps: S1: Collect historical normal signals, abnormal characteristic signals, and unidentified characteristic signals for each corresponding three-phase meter under different load conditions.

[0014] It should be noted that the specific scenario addressed in this application may be: real-time identification of circuit abnormalities in three-phase electricity meters by performing time-frequency domain fusion based on the time-frequency domain differences and time-frequency domain weights between the target window at the current time and the adjacent windows.

[0015] In one embodiment of this application, a voltage transformer is used to collect historical normal and abnormal voltage signals from each phase of a three-phase meter under different load conditions, and a current transformer is used to collect historical normal and abnormal current signals from each phase of the three-phase meter under different load conditions. Similarly, the voltage and current signals to be identified are obtained. One voltage transformer and one current transformer can be arranged for each phase. For example, they are placed on the phase lines of each phase. Furthermore, wavelet denoising can be used to denoise the collected signals, and the collected signals can be normalized to eliminate dimensional differences.

[0016] Thus, the characteristic (current, voltage) sequences of the three-phase historical normal and abnormal conditions under different load conditions and the characteristic sequences (current, voltage) to be identified are obtained. For the characteristic sequences to be identified, each phase contains a set of characteristic sequences. In this scheme, there are three sets, each of which includes two characteristic sequences: current and voltage.

[0017] S2: Use Fourier transform to obtain the spectrum sequence of the collected historical feature sequence. Calculate the time-domain difference and frequency-domain difference based on the three corresponding historical normal sequences, abnormal feature sequences and spectrum sequences under the same load condition, and calculate the weights of the time-domain and frequency-domain differences through the time-domain difference and frequency-domain difference.

[0018] In one embodiment of this application, since three-phase circuits in daily life contain various load states, the fluctuations in current or voltage caused by load fluctuations under different load states are different in the time and frequency domains. When a circuit malfunctions, there may be cases where the abnormality of current or voltage is not obvious in a single time or frequency domain. The circuit identification of traditional three-phase meters is mainly based on a single time and frequency domain feature or weighted fusion, which ignores the significance of abnormalities in the time and frequency domains, resulting in untimely identification or misidentification when the three-phase meter circuit performs abnormality detection.

[0019] Therefore, this application uses Fourier transform to obtain the spectral sequence of the collected historical feature sequences, and calculates the time-domain and frequency-domain differences based on the three-phase historical normal sequences, abnormal feature sequences, and spectral sequences under the same load condition. The weights of the time and frequency domains are then calculated based on these differences. Specifically, the process of obtaining the anomaly identification weights is as follows: First, Fourier transform is used to obtain the spectrum sequence corresponding to each feature sequence for the three-phase historical normal and abnormal feature (current, voltage) sequences under different load conditions. The feature sequence contains the value of the feature in the time series, and the spectrum sequence contains the amplitude corresponding to each frequency. Each feature sequence corresponds to a spectrum sequence.

[0020] Then, for target feature 'a' among multiple features, the target feature under each load condition is calculated based on the historical normal and abnormal feature sequences of the three-phase system under different load conditions. The time-domain difference can be calculated similarly based on the spectral sequence to determine the frequency-domain difference. As an example, using target features... Taking the calculation of time-domain differences as an example: in, This indicates the characteristics under this load condition. The time domain difference, This indicates the target features collected under this load condition. The number of normal (abnormal) feature sequences, This indicates the length of each feature sequence. Indicates the The first normal feature sequence One element, Indicates the The first abnormal feature sequence One element, Its function is to normalize and eliminate the dimensional differences between the time and frequency domains, allowing calculations to be performed between these differences. This is because there is a phase difference between different phases. Both normal and abnormal feature sequences are signals acquired on the same phase. Similarly, features are calculated based on the spectral sequence. Frequency domain differences The frequency domain difference is calculated based on the spectral sequence, and the calculation method is the same as that for the time domain difference.

[0021] Finally, regarding target features The time-domain and frequency-domain weights are calculated based on the time-domain and frequency-domain differences of this feature under the same load condition. Taking the calculation of the time-domain weight as an example: in, Representation of features Temporal weights in anomaly detection A larger value indicates the normal or abnormal characteristics of the circuit in the time domain. The greater the difference, This indicates the characteristics under this load condition. The time domain difference, This indicates the characteristics under this load condition. The frequency domain difference. Similarly, to obtain features. Frequency domain weights in anomaly detection The logic for dividing the weights in the time and frequency domains is as follows: the time (frequency) domain difference is calculated by the difference between the normal signal and the abnormal signal in the time and frequency domains. The larger the time (frequency) domain difference, the greater and more significant the difference between the abnormal and normal circuits in the time (frequency) domain, and the more helpful it is for anomaly identification.

[0022] Thus, the time-domain weight and frequency-domain weight of each feature under each load condition are obtained.

[0023] S3: Divide the feature sequence to be identified using a sliding window, calculate the time-domain and frequency-domain anomaly degree based on the time-frequency domain features of the target window and neighboring windows, and calculate the single-phase anomaly degree of the target feature using the time-domain and frequency-domain weights of the target feature under each load condition.

[0024] In one embodiment of this application, since the load of a three-phase circuit fluctuates instantaneously during use, the instantaneous fluctuation of the load causes the characteristics to produce abrupt values ​​in the time and frequency domain. For example, the harmonic content changes instantaneously when the motor starts and stops. Traditional anomaly identification is usually based on static fixed characteristic (current, voltage) thresholds, which cannot distinguish between normal transient impacts and real faults, harmonic transients caused by load fluctuations, etc., resulting in misidentification or untimely identification when identifying circuit anomalies of three-phase meters.

[0025] Therefore, this application uses a sliding window to divide the feature sequence to be identified, calculates the time-domain and frequency-domain anomaly levels based on the time-frequency domain features of the target window and neighboring windows, and calculates the single-phase anomaly level of the feature using the time-domain and frequency-domain weights of the feature under each load condition. The calculation of the single-phase anomaly level specifically includes the following steps: (1) For a target feature and a phase The feature sequence to be identified is divided into sliding windows, with an example window length of 5. The temporal anomaly level is calculated based on the sliding window. in, Indicates phase upper features The degree of temporal anomaly, Indicates the number of adjacent windows. Indicates the length of the sliding window. Indicates the first in the target window One element, Indicates the The first neighboring window One element, Indicates the first [number] in the target window The slope of each element, Indicates the Within the nearest window The slope of each element, It is a preset feature value whose function is to eliminate dimensions. The calculation logic of the degree of abnormality is as follows: when the circuit is abnormal, the feature value that will produce the abnormality at the current time is significantly different from that at the nearby time. Therefore, the greater the difference between the target window and the nearby windows, the greater the probability of an abnormality occurring at the current time.

[0026] (2) For a target feature and a phase For each window divided in process (1), Fourier transform is used to obtain the spectrum sequence corresponding to each window. The frequency domain anomaly is calculated based on the spectrum sequences of the window in the previous time step and the neighboring windows: in, Indicates phase upper features The degree of frequency domain anomaly, Indicates the number of adjacent windows. Indicates the length of the spectral sequence. The first spectral sequence of the target window represents the... One element, Indicates the The spectral sequence of the nearest neighbor window Each element. The calculation logic for the degree of abnormality is as follows: when the circuit is abnormal, the feature will generate unique harmonics in the frequency domain. This change can be captured by the difference in amplitude changes at various frequencies between the target window and the adjacent windows in the frequency domain. The larger the difference, the greater the probability of generating harmonics. The greater the probability of amplitude changes at different frequencies, the greater the probability of circuit abnormality.

[0027] (3) For a target feature and a phase Based on phase The calculation of the time-domain anomaly degree and frequency-domain anomaly degree of the above features is similar. upper features The degree of monophasic anomaly: in, Indicates in phase upper features The degree of single-phase anomaly, This indicates features under the same load state as the feature sequence to be identified. Temporal weights in anomaly detection Similarly, Indicates phase upper features The degree of temporal anomaly, Same thing.

[0028] (4) Similarly, the degree of single-phase anomaly of each feature in each phase is obtained.

[0029] S4: Input the single-phase anomaly level of each target feature on each phase into the control system, and identify the circuit anomaly of the three-phase meter according to the preset anomaly level threshold.

[0030] In one embodiment of this application, since a three-phase circuit fault typically manifests first in a single phase, an alarm signal is issued to the system when the degree of single-phase fault in any phase exceeds a preset fault threshold. The preset fault threshold is a pre-set standard for determining the numerical value corresponding to the degree of fault. For example, the preset fault threshold is 0.7.

[0031] This application also discloses a circuit anomaly identification system for a three-phase electricity meter, which employs the aforementioned circuit anomaly identification method for a three-phase electricity meter, specifically including: The signal acquisition module is used to acquire historical normal signals, abnormal characteristic signals, and characteristic signals to be identified for each corresponding three-phase meter under different load conditions; The weight calculation module is used to obtain the spectrum sequence by Fourier transform of the collected historical feature sequence, calculate the time domain difference and frequency domain difference based on the three corresponding historical normal sequence, abnormal feature sequence and spectrum sequence under the same load state, and calculate the weights in the time domain and frequency domain through the time domain difference and frequency domain difference. The single-phase anomaly degree calculation module is used to divide the feature sequence to be identified using a sliding window, calculate the time-domain and frequency-domain anomaly degree based on the time-frequency domain features of the target window and neighboring windows, and calculate the single-phase anomaly degree of the target feature using the time-domain and frequency-domain weights of the target feature under each load state. The anomaly identification module is used to input the single-phase anomaly degree of each target feature on each phase into the control system, and to identify circuit anomalies of the three-phase meters according to the preset anomaly degree threshold. The calculation of the single-phase anomaly degree includes: using Fourier transform to obtain the spectrum sequence corresponding to each window of the sliding window; calculating the frequency domain anomaly degree based on the spectrum sequence of the window in the previous time step and the neighboring windows; and calculating the single-phase anomaly degree of the target feature based on the time domain anomaly degree and the frequency domain anomaly degree of the target feature, so as to obtain the single-phase anomaly degree of each target feature in each phase.

[0032] In summary, this application uses Fourier transform to obtain the spectral sequence of the collected historical feature sequences. Based on the three-phase historical normal and abnormal feature sequences and spectral sequences under the same load condition, time-domain and frequency-domain differences are calculated, and the weights of the time and frequency domains are calculated using these differences. A sliding window is used to divide the feature sequence to be identified. The time-domain and frequency-domain anomaly levels are calculated based on the time-frequency domain features of the target window and neighboring windows. The single-phase anomaly level of the feature is calculated using the time-domain and frequency-domain weights of the feature under each load condition. Adaptive weight fusion of time-frequency domain features avoids high false alarm rates from single time-frequency domain features, and time series analysis avoids false alarms caused by instantaneous changes in features at a single moment.

[0033] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A method for identifying circuit anomalies in a three-phase electricity meter, characterized in that, Includes the following steps: Collect historical normal signals, abnormal characteristic signals, and unidentified characteristic signals for each corresponding three-phase meter under different load conditions; Fourier transform is used to obtain the spectrum sequence of the collected historical feature sequence. The time domain difference and frequency domain difference are calculated based on the three corresponding historical normal sequence, abnormal feature sequence and spectrum sequence under the same load condition. The weights of the time domain and frequency domain are calculated through the time domain difference and the frequency domain difference. The feature sequence to be identified is divided using a sliding window. The time-domain and frequency-domain anomaly levels are calculated based on the time-frequency domain features of the target window and neighboring windows. The single-phase anomaly level of the target feature is calculated using the time-domain and frequency-domain weights of the target feature under each load condition. The single-phase anomaly level of each target feature on each phase is input into the control system, and the circuit anomaly of the three-phase meter is identified according to the preset anomaly level threshold. The calculation of the single-phase anomaly degree includes: using Fourier transform to obtain the spectrum sequence corresponding to each window of the sliding window; calculating the frequency domain anomaly degree based on the spectrum sequence of the window in the previous time step and the neighboring windows; and calculating the single-phase anomaly degree of the target feature based on the time domain anomaly degree and the frequency domain anomaly degree of the target feature, so as to obtain the single-phase anomaly degree of each target feature in each phase. The calculation of the degree of single-phase anomaly specifically includes the following steps: (1) For a target feature and a phase The target feature sequence is divided into a sliding window, and the temporal anomaly degree is calculated based on the window after sliding division. in, Indicates phase upper features The degree of temporal anomaly, Indicates the number of adjacent windows. Indicates the length of the sliding window. Indicates the first in the target window One element, Indicates the first The first neighboring window One element, Indicates the first [number] in the target window The slope of each element, Indicates the first Within the nearest window The slope of each element, It is a preset feature value whose function is to eliminate dimensions. The calculation logic of the degree of abnormality is as follows: when the circuit is abnormal, the feature value that will produce the abnormality at the current moment is significantly different from that at the nearest moment. Therefore, the greater the difference between the target window and the nearest window, the greater the probability of an abnormality occurring at the current moment. (2) For a target feature and a phase For each window divided in process (1), Fourier transform is used to obtain the spectrum sequence corresponding to each window. The frequency domain anomaly is calculated based on the spectrum sequences of the window in the previous time step and the neighboring windows: in, Indicates phase upper features The degree of frequency domain anomaly, Indicates the number of adjacent windows. Indicates the length of the spectral sequence. The first spectral sequence of the target window represents the... One element, Indicates the first The spectral sequence of the nearest neighbor window Each element. The calculation logic for the degree of abnormality is as follows: when the circuit is abnormal, the feature will generate unique harmonics in the frequency domain. This change can be captured by the difference in amplitude changes at various frequencies between the target window and the adjacent windows in the frequency domain. The larger the difference, the greater the probability of generating harmonics. The greater the probability of amplitude changes at different frequencies, the greater the probability of circuit abnormality. (3) For a target feature and a phase Based on phase The calculation of the time-domain anomaly degree and frequency-domain anomaly degree of the above features is similar. upper features The degree of monophasic anomaly: in, Indicates in phase upper features The degree of single-phase anomaly, This indicates features under the same load state as the feature sequence to be identified. Temporal weights in anomaly detection Similarly, Indicates phase upper features The degree of temporal anomaly, Similarly; (4) Similarly, the degree of single-phase anomaly of each feature in each phase is obtained; Calculating the weights in the time and frequency domains includes: Fourier transform is used to obtain the spectral sequence corresponding to each characteristic sequence for the three-phase historical normal and abnormal characteristic sequences under different load conditions; The time-domain and frequency-domain differences of the target characteristics under each load condition are calculated based on the characteristic sequences of the three-phase historical normal and abnormal phases under different load conditions. The weights in the time and frequency domains are calculated based on the time-domain and frequency-domain differences of the target characteristics under the same load condition.

2. The method according to claim 1, characterized in that, Voltage and current transformers are used to collect historical normal signals, abnormal characteristic signals, and unidentified characteristic signals for each corresponding three-phase meter under different load conditions.

3. The method according to claim 2, characterized in that, Wavelet denoising is used to denoise the historical normal signals, abnormal feature signals, and feature signals to be identified for each corresponding three-phase meter under different load conditions.

4. The method according to claim 2, characterized in that, Normalize the historical normal signals, abnormal characteristic signals, and unidentified characteristic signals of each corresponding three-phase meter under different load conditions to eliminate dimensional differences.

5. A circuit anomaly identification system for a three-phase electricity meter, employing the aforementioned circuit anomaly identification method for a three-phase electricity meter, characterized in that, include: The signal acquisition module is used to acquire historical normal signals, abnormal characteristic signals, and characteristic signals to be identified for each corresponding three-phase meter under different load conditions; The weight calculation module is used to obtain the spectrum sequence by Fourier transform of the collected historical feature sequence, calculate the time domain difference and frequency domain difference based on the three corresponding historical normal sequence, abnormal feature sequence and spectrum sequence under the same load state, and calculate the weight of the time domain and frequency domain through the time domain difference and the frequency domain difference; The single-phase anomaly degree calculation module is used to divide the feature sequence to be identified using a sliding window, calculate the time-domain and frequency-domain anomaly degree based on the time-frequency domain features of the target window and neighboring windows, and calculate the single-phase anomaly degree of the target feature using the time-domain and frequency-domain weights of the target feature under each load state. The anomaly identification module is used to input the single-phase anomaly degree of each target feature on each phase into the control system, and to identify circuit anomalies of the three-phase meters according to the preset anomaly degree threshold. The calculation of the single-phase anomaly degree includes: using Fourier transform to obtain the spectrum sequence corresponding to each window of the sliding window; calculating the frequency domain anomaly degree based on the spectrum sequence of the window in the previous time step and the neighboring windows; and calculating the single-phase anomaly degree of the target feature based on the time domain anomaly degree and the frequency domain anomaly degree of the target feature, so as to obtain the single-phase anomaly degree of each target feature in each phase. The calculation of the degree of single-phase anomaly specifically includes the following steps: (1) For a target feature and a phase The target feature sequence is divided into a sliding window, and the temporal anomaly degree is calculated based on the window after sliding division. in, Indicates phase upper features The degree of temporal anomaly, Indicates the number of adjacent windows. Indicates the length of the sliding window. Indicates the first in the target window One element, Indicates the first The first neighboring window One element, Indicates the first [number] in the target window The slope of each element, Indicates the first Within the nearest window The slope of each element, It is a preset feature value whose function is to eliminate dimensions. The calculation logic of the degree of abnormality is as follows: when the circuit is abnormal, the feature value that will produce the abnormality at the current moment is significantly different from that at the nearest moment. Therefore, the greater the difference between the target window and the nearest window, the greater the probability of an abnormality occurring at the current moment. (2) For a target feature and a phase For each window divided in process (1), Fourier transform is used to obtain the spectrum sequence corresponding to each window. The frequency domain anomaly is calculated based on the spectrum sequences of the window in the previous time step and the neighboring windows: in, Indicates phase upper features The degree of frequency domain anomaly, Indicates the number of adjacent windows. Indicates the length of the spectral sequence. The first spectral sequence of the target window represents the... One element, Indicates the first The spectral sequence of the nearest neighbor window Each element. The calculation logic for the degree of abnormality is as follows: when the circuit is abnormal, the feature will generate unique harmonics in the frequency domain. This change can be captured by the difference in amplitude changes at various frequencies between the target window and the adjacent windows in the frequency domain. The larger the difference, the greater the probability of generating harmonics. The greater the probability of amplitude changes at different frequencies, the greater the probability of circuit abnormality. (3) For a target feature and a phase Based on phase The calculation of the time-domain anomaly degree and frequency-domain anomaly degree of the above features is similar. upper features The degree of monophasic anomaly: in, Indicates in phase upper features The degree of single-phase anomaly, This indicates features under the same load state as the feature sequence to be identified. Temporal weights in anomaly detection Similarly, Indicates phase upper features The degree of temporal anomaly, Similarly; (4) Similarly, the degree of single-phase anomaly of each feature in each phase is obtained; The weight calculation module also includes: The transformation module is used to obtain the spectrum sequence corresponding to each characteristic sequence by applying Fourier transform to the characteristic sequences of three-phase historical normal and abnormal under different load conditions; The difference calculation module is used to calculate the time-domain and frequency-domain differences of the target characteristics under different load conditions based on the characteristic sequences of the three-phase historical normal and abnormal features under different load conditions. The target weight calculation module is used to calculate the time-domain and frequency-domain weights based on the time-domain and frequency-domain differences of the target characteristics under the same load condition.

6. The system according to claim 5, characterized in that, The signal acquisition module includes: The target signal acquisition module is used to acquire historical normal signals, abnormal characteristic signals and unidentified characteristic signals of each corresponding three-phase meter under different load conditions using voltage and current transformers. The denoising module is used to perform denoising processing on the historical normal signals, abnormal feature signals and feature signals to be identified for each corresponding three-phase meter under different load conditions using wavelet denoising. The normalization module is used to normalize the historical normal signals, abnormal characteristic signals, and unidentified characteristic signals of each corresponding three-phase meter under different load conditions to eliminate dimensional differences.

Images