A self-test method and system for power distribution equipment based on intelligent circuit breaker

Abstract

The present invention provides a power distribution equipment self-test method and system based on an intelligent circuit breaker. The method comprises the following steps: calculating the full-phase spectrum of two continuous full-cycle alternating current current data, and obtaining a differential spectrum matrix based on the two full-phase spectrums; taking the full-cycle alternating current current data corresponding to the full-phase spectrum with the largest spectrum energy as the target current data, generating a time-frequency two-dimensional energy distribution diagram of the target current data, and calculating a time-frequency information entropy value based on the time-frequency two-dimensional energy distribution diagram; obtaining a differential spectrum matrix sequence and a time-frequency two-dimensional energy distribution diagram sequence according to a preset sliding window length and step size, and extracting features of the differential spectrum matrix sequence and the time-frequency two-dimensional energy distribution diagram sequence; determining a fusion weight based on the fundamental wave amplitude and total harmonic distortion of the current data within the sliding window, and using the fusion weight to weightedly fuse the features of the differential spectrum matrix sequence and the features of the time-frequency two-dimensional energy distribution diagram sequence, and obtaining a self-test result based on the weighted fusion result.

Description

Technical Field

[0001] The present application relates to the field of equipment self-test, and in particular to a method and system for self-testing power distribution equipment based on an intelligent circuit breaker. Background Art

[0002] Traditional power distribution equipment protection devices, such as fuses and thermal-magnetic circuit breakers, primarily provide passive protection against severe faults such as overloads and short circuits. With the development of smart grid technology, self-testing technologies can elevate equipment protection from passive response to active early warning. Smart circuit breakers, through built-in sensors and microprocessors, can provide 24 / 7, real-time monitoring of key line parameters such as current, voltage, power, and temperature. Not only can they quickly disconnect the circuit when a fault occurs, but they can also analyze subtle changes in electrical data to proactively identify potential safety hazards such as loose connections, aging insulation, and three-phase imbalance, thereby preventing serious incidents such as electrical fires and equipment damage.

[0003] Compared to overload and other faults, accurately detecting series arc faults is a key challenge in the self-test of power distribution equipment for intelligent circuit breakers. Series arcs are caused by poor conductor contact, loose connections, or insulation damage. While the fault current is often insufficient to trigger traditional overcurrent protection, the resulting localized high temperatures are sufficient to quickly ignite surrounding combustibles, making them a leading cause of electrical fires. Arc signals appear as wide-spectrum, high-frequency, non-stationary random noise superimposed on the power frequency current. However, many everyday electrical devices also generate highly similar electrical noise during normal operation. For example, commutation sparks from series-wound motors such as electric drills and vacuum cleaners, high-frequency switching noise from switching power supplies such as calculators and energy-saving lamps, and waveform distortion from phase-controlled loads such as dimmable lamps. This close resemblance between fault signatures and normal interference results in high false alarm and missed detection rates for existing detection methods. Time-domain and frequency-domain analysis methods struggle to distinguish the random, wide-spectrum arc from the normal operating spectrum of certain electrical devices amidst complex interference, making them inadequate for complex and variable real-world operating conditions. Summary of the Invention

[0004] To address the problem that the current arc fault detection method in the self-test of power distribution equipment using intelligent circuit breakers is insufficient in identifying complex interference sources and is prone to false alarms and missed alarms, this application proposes a self-test method for power distribution equipment based on intelligent circuit breakers, comprising:

[0005] Obtain current data of two continuous full-cycle alternating currents monitored by the intelligent circuit breaker, perform full-phase Fourier transform on the current data of the two continuous full-cycle alternating currents to obtain two full-phase spectra, and calculate a differential spectrum matrix based on the two full-phase spectra; calculate the spectral energy of the two full-phase spectra within a preset high-frequency band, use the current data of the full-cycle alternating current corresponding to the full-phase spectrum with the largest spectral energy as the target current data, perform continuous wavelet transform on the target current data to generate a time-frequency two-dimensional energy distribution map, and calculate a time-frequency information entropy value based on the time-frequency two-dimensional energy distribution map;

[0006] A differential spectrum matrix sequence and a time-frequency two-dimensional energy distribution map sequence are obtained according to a preset sliding window length and step size, a time-frequency information entropy value sequence corresponding to the time-frequency two-dimensional energy distribution map sequence is obtained, a channel attention of the differential spectrum matrix sequence is obtained according to the time-frequency information entropy value sequence, and features of the differential spectrum matrix sequence are extracted using the channel attention; features of the time-frequency two-dimensional energy distribution map sequence are extracted based on the stability of the frequency components in the differential spectrum matrix;

[0007] A fusion weight is determined based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window, and the characteristics of the differential spectrum matrix sequence and the characteristics of the time-frequency two-dimensional energy distribution diagram sequence are weightedly fused using the fusion weight. A self-test result is obtained based on the result of the weighted fusion.

[0008] Optionally, obtaining the channel attention of the differential spectrum matrix sequence according to the time-frequency information entropy value sequence includes:

[0009] Inputting the time-frequency information entropy value sequence into a multi-layer perceptron, wherein the multi-layer perceptron includes at least one fully connected layer and one activation layer;

[0010] The multilayer perceptron is used to perform a nonlinear transformation on the time-frequency information entropy value sequence to obtain attention weight vectors with the same number as the differential spectrum matrix sequence, and each value in the attention weight vector corresponds to a differential spectrum matrix in the differential spectrum matrix sequence.

[0011] Optionally, obtaining the channel attention of the differential spectrum matrix sequence according to the time-frequency information entropy value sequence includes:

[0012] Normalizing each entropy value in the time-frequency information entropy value sequence to obtain a normalized entropy value sequence;

[0013] Multiplying the normalized entropy value sequence by each matrix in the differential spectrum matrix sequence element by element to obtain a weighted differential spectrum matrix sequence;

[0014] Sum the rows and columns of each matrix in the weighted difference spectrum matrix sequence to obtain the channel attention vector;

[0015] Channel attention is obtained by normalizing the channel attention vector.

[0016] Optionally, the extracting features of the time-frequency two-dimensional energy distribution diagram sequence based on the stability of frequency components in the differential spectrum matrix includes:

[0017] For each time-frequency two-dimensional energy distribution graph in the time-frequency two-dimensional energy distribution graph sequence, obtaining the differential spectrum matrix at the same position of the sliding window;

[0018] Calculating the absolute value of each frequency component in the differential spectrum matrix to obtain a frequency instability vector;

[0019] Normalizing the frequency instability vector to obtain a frequency attention weight vector;

[0020] Multiplying each row of the time-frequency two-dimensional energy distribution map by the corresponding weight value in the frequency attention weight vector to obtain an enhanced time-frequency two-dimensional energy distribution map;

[0021] The enhanced time-frequency two-dimensional energy distribution map sequence is input into a two-dimensional convolutional neural network to obtain the characteristics of the time-frequency two-dimensional energy distribution map sequence.

[0022] Optionally, determining the fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window includes:

[0023] Calculating the fundamental wave amplitude and total harmonic distortion of each current data in the sliding window; normalizing the fundamental wave amplitude and total harmonic distortion to obtain normalized fundamental wave amplitude and normalized total harmonic distortion;

[0024] The normalized fundamental wave amplitude and the normalized total harmonic distortion are input into the weighting function to obtain the fusion weight.

[0025] This application also proposes a power distribution equipment self-test system based on an intelligent circuit breaker, comprising:

[0026] a preprocessing unit for acquiring current data of two continuous full-cycle alternating currents monitored by the intelligent circuit breaker, performing full-phase Fourier transform on the current data of the two continuous full-cycle alternating currents to obtain two full-phase spectra, and calculating a differential spectrum matrix based on the two full-phase spectra; calculating the spectral energy of the two full-phase spectra within a preset high-frequency band, taking the current data of the full-cycle alternating current corresponding to the full-phase spectrum with the largest spectral energy as the target current data, performing continuous wavelet transform on the target current data to generate a time-frequency two-dimensional energy distribution map, and calculating a time-frequency information entropy value based on the time-frequency two-dimensional energy distribution map;

[0027] A feature extraction unit is configured to obtain a differential spectrum matrix sequence and a time-frequency two-dimensional energy distribution map sequence according to a preset sliding window length and step size, obtain a time-frequency information entropy value sequence corresponding to the time-frequency two-dimensional energy distribution map sequence, obtain a channel attention of the differential spectrum matrix sequence based on the time-frequency information entropy value sequence, and extract features of the differential spectrum matrix sequence using the channel attention; and extract features of the time-frequency two-dimensional energy distribution map sequence based on the stability of the frequency components in the differential spectrum matrix.

[0028] A self-test unit is used to determine a fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window, use the fusion weight to weightedly fuse the characteristics of the differential spectrum matrix sequence and the characteristics of the time-frequency two-dimensional energy distribution diagram sequence, and obtain a self-test result based on the result of the weighted fusion.

[0029] Optionally, obtaining the channel attention of the differential spectrum matrix sequence according to the time-frequency information entropy value sequence includes:

[0030] Inputting the time-frequency information entropy value sequence into a multi-layer perceptron, wherein the multi-layer perceptron includes at least one fully connected layer and one activation layer;

[0031] The multilayer perceptron is used to perform a nonlinear transformation on the time-frequency information entropy value sequence to obtain attention weight vectors with the same number as the differential spectrum matrix sequence, and each value in the attention weight vector corresponds to a differential spectrum matrix in the differential spectrum matrix sequence.

[0032] Optionally, obtaining the channel attention of the differential spectrum matrix sequence according to the time-frequency information entropy value sequence includes:

[0033] Normalizing each entropy value in the time-frequency information entropy value sequence to obtain a normalized entropy value sequence;

[0034] Multiplying the normalized entropy value sequence by each matrix in the differential spectrum matrix sequence element by element to obtain a weighted differential spectrum matrix sequence;

[0035] Sum the rows and columns of each matrix in the weighted difference spectrum matrix sequence to obtain the channel attention vector;

[0036] Channel attention is obtained by normalizing the channel attention vector.

[0037] Optionally, the extracting features of the time-frequency two-dimensional energy distribution diagram sequence based on the stability of frequency components in the differential spectrum matrix includes:

[0038] For each time-frequency two-dimensional energy distribution graph in the time-frequency two-dimensional energy distribution graph sequence, obtaining the differential spectrum matrix at the same position of the sliding window;

[0039] Calculating the absolute value of each frequency component in the differential spectrum matrix to obtain a frequency instability vector;

[0040] Normalizing the frequency instability vector to obtain a frequency attention weight vector;

[0041] Multiplying each row of the time-frequency two-dimensional energy distribution map by the corresponding weight value in the frequency attention weight vector to obtain an enhanced time-frequency two-dimensional energy distribution map;

[0042] The enhanced time-frequency two-dimensional energy distribution map sequence is input into a two-dimensional convolutional neural network to obtain the characteristics of the time-frequency two-dimensional energy distribution map sequence.

[0043] Optionally, determining the fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window includes:

[0044] Calculating the fundamental wave amplitude and total harmonic distortion of each current data in the sliding window; normalizing the fundamental wave amplitude and total harmonic distortion to obtain normalized fundamental wave amplitude and normalized total harmonic distortion;

[0045] The normalized fundamental wave amplitude and the normalized total harmonic distortion are input into the weighting function to obtain the fusion weight.

[0046] The present invention captures the instability of the arc by calculating the differential spectrum matrix of continuous cycles, and determines the degree of chaos in combination with the time-frequency information entropy, which greatly enhances the recognition of fault characteristics. In addition, the present invention uses the time-frequency information entropy value to adjust the channel attention so that the characteristics of the differential spectrum matrix sequence are extracted more accurately, and uses the stability of the frequency components in the differential spectrum matrix to extract the characteristics of the time-frequency two-dimensional energy distribution diagram sequence to enhance the key frequencies. The real-time working conditions of the line, that is, the load level represented by the fundamental amplitude and the power quality represented by the total harmonic distortion, are further used to adjust the fusion weight of the multi-dimensional features, thereby enhancing the accuracy and reliability under different load types and electromagnetic environments, thereby improving the accuracy of arc fault detection and reducing missed alarms and false alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of Example 1;

[0048] Figure 2 is a schematic diagram of current data;

[0049] Figure 3 Schematic diagram of the amplitude spectrum and phase spectrum of the full phase spectrum;

[0050] Figure 4 It is a schematic diagram of the time-frequency two-dimensional energy distribution diagram;

[0051] Figure 5Schematic diagram of the differential spectrum matrix sequence and the time-frequency two-dimensional energy distribution diagram sequence. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0053] Specific embodiments, such as Figure 1 As shown, the present application proposes a self-test method for power distribution equipment based on an intelligent circuit breaker, comprising:

[0054] S1, obtaining current data of two continuous full-cycle alternating currents monitored by the intelligent circuit breaker, performing full-phase Fourier transform on the current data of the two continuous full-cycle alternating currents to obtain two full-phase spectra, and calculating a differential spectrum matrix based on the two full-phase spectra; calculating the spectral energy of the two full-phase spectra within a preset high-frequency band, taking the current data of the full-cycle alternating current corresponding to the full-phase spectrum with the largest spectral energy as the target current data, performing continuous wavelet transform on the target current data to generate a time-frequency two-dimensional energy distribution map, and calculating a time-frequency information entropy value based on the time-frequency two-dimensional energy distribution map;

[0055] Smart circuit breakers continuously monitor the current flowing through the line, such as Figure 2 As shown, the current data of two closely connected complete cycles are obtained from it. Taking 50hz alternating current as an example, the current waveform data with a total duration of 40 milliseconds is obtained, and the two 20-millisecond current waveforms are subjected to full-phase Fourier transform respectively to obtain two spectra such as spectrum A and spectrum B. The value of each frequency point in spectrum B is subtracted from the value of the corresponding frequency point in spectrum A to obtain the differential spectrum. The differential spectrum highlights the frequency components that change between the two cycles, while the stable background harmonics will be offset. Spectra A and spectrum B are both complex sequences, assuming they are respectively 、 , after subtraction, , separate the real and imaginary parts of the one-dimensional complex difference spectrum, such as Figure 3 As shown in Figure 1, the two-dimensional matrix is ​​arranged into an N×2 matrix, which is the differential spectrum matrix. The first column of the differential spectrum matrix is ​​the real part of the differential complex number at all frequency points, and the second column is the imaginary part of the differential complex number at all frequency points. The differential spectrum matrix contains the real part difference representing the change of the waveform on the cosine component and the imaginary part difference representing the change of the waveform on the sine component.

[0056] Set a high-frequency band, for example, 2 kHz to 100 kHz, and calculate the total energy within this band for both spectrum A and spectrum B. If the high-frequency energy of spectrum B is significantly greater than that of spectrum A, a dangerous, persistent arc will inevitably cause an abnormal increase in the high-frequency spectrum energy of the line. Without this increase in high-frequency energy, a true arc cannot occur. Therefore, the second cycle is more likely to be the arcing period, and the current data from this second cycle is selected as the target current data. A continuous wavelet transform is performed on the target current data to generate a two-dimensional time-frequency energy distribution graph. The horizontal axis of the time-frequency energy distribution graph is time, and the vertical axis is frequency. The brightness of each point in the graph represents the energy intensity of that frequency at that moment. The overall information entropy of the time-frequency energy distribution graph is further calculated to obtain the time-frequency information entropy value. The time-frequency information entropy value indicates the degree of disorder in the energy distribution on the graph; a highly disordered distribution will result in a high entropy value.

[0057] S2, obtaining a differential spectrum matrix sequence and a time-frequency two-dimensional energy distribution map sequence according to a preset sliding window length and step size, obtaining a time-frequency information entropy value sequence corresponding to the time-frequency two-dimensional energy distribution map sequence, obtaining a channel attention of the differential spectrum matrix sequence according to the time-frequency information entropy value sequence, and extracting features of the differential spectrum matrix sequence using the channel attention; extracting features of the time-frequency two-dimensional energy distribution map sequence based on the stability of the frequency components in the differential spectrum matrix;

[0058] Get the preset sliding window, assuming the length is 100ms and the step size is 20ms. Each window position contains a sequence consisting of 5 groups of initial features, namely a sequence of 5 differential spectrum matrices and a sequence of 5 time-frequency two-dimensional energy distribution maps, as well as a sequence of 5 corresponding time-frequency information entropy values, such as Figure 5 shown.

[0059] To extract features from the differential spectrum matrix sequence, we use the time-frequency information entropy value sequence to obtain channel attention. For example, we use an activation function to map the five time-frequency information entropy values ​​into five weights between 0 and 1. Then, we extract preliminary features from each of the five differential spectrum matrices. These five preliminary features are weighted and summed according to the weights to obtain a feature vector, which serves as the feature of the differential spectrum matrix sequence.

[0060] For feature extraction of a time-frequency two-dimensional energy distribution graph sequence, the information of the differential spectrum matrix sequence is used to enhance the extraction process. For example, when analyzing the third time-frequency two-dimensional energy distribution graph, the third differential spectrum matrix is ​​obtained, and calculations show that the values ​​of the two frequency points of 15 kHz and 40 kHz in the third differential spectrum matrix are particularly high, indicating that these two frequencies are the most unstable. Then, two horizontal rows of data representing 15 kHz and 40 kHz are obtained from the third time-frequency two-dimensional energy distribution graph to obtain a compressed time-frequency graph. The compressed time-frequency graph is input into a recurrent neural network, for example, to obtain the features of the third time-frequency two-dimensional energy distribution graph. The above operation is performed on all five time-frequency graphs in the sequence to obtain the features of the time-frequency two-dimensional energy distribution graph sequence.

[0061] S3, determining a fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window, using the fusion weight to weightedly fuse the features of the differential spectrum matrix sequence and the features of the time-frequency two-dimensional energy distribution diagram sequence, and obtaining a self-test result based on the result of the weighted fusion.

[0062] Based on the current data within the current sliding window, two operating condition indicators—the fundamental amplitude and total harmonic distortion—are calculated. Assume that these two indicators indicate that the current line is under high load and has severe harmonic pollution. These two operating condition indicators are input into a pre-trained decision tree model. The decision tree model outputs an optimal set of fusion weights based on the input operating conditions. For example, for high load and high harmonic pollution, the decision tree outputs are 0.3 and 0.7. The differential spectrum matrix sequence features obtained in the second step are multiplied by 0.3, and the time-frequency two-dimensional energy distribution map sequence features are multiplied by 0.7. These two are then added together to generate a fused feature vector. This fused feature vector is input into a logistic regression classifier. If the confidence score output by the classifier remains above a preset fault threshold, such as 0.95, for a sustained period of time, an arc fault is determined to exist in the current distribution equipment, and an alarm signal is generated.

[0063] In an optional embodiment, obtaining the channel attention of the differential spectrum matrix sequence according to the time-frequency information entropy value sequence includes:

[0064] Inputting the time-frequency information entropy value sequence into a multi-layer perceptron, wherein the multi-layer perceptron includes at least one fully connected layer and one activation layer;

[0065] The multilayer perceptron is used to perform a nonlinear transformation on the time-frequency information entropy value sequence to obtain attention weight vectors with the same number as the differential spectrum matrix sequence, and each value in the attention weight vector corresponds to a differential spectrum matrix in the differential spectrum matrix sequence.

[0066] Specifically, suppose that within a sliding window, the sequence of five time-frequency information entropy values ​​is [1.2, 1.5, 3.8, 1.4, 1.1]. This five-digit sequence is input into a multilayer perceptron. The fully connected layer of the multilayer perceptron operates on these five input values, and the subsequent activation layer performs nonlinear processing. After the nonlinear transformation of the multilayer perceptron, a new five-dimensional vector, or attention weight vector, is output. For example, [0.05, 0.10, 0.70, 0.10, 0.05], the weight corresponding to the information of the fifth differential spectrum matrix is ​​0.05, while the weight corresponding to the information of the third matrix is ​​0.70. Since the time-frequency information entropy value measured at the third time point is the highest, indicating that the signal is the most chaotic, the main attention is focused on the third differential spectrum matrix, thereby reducing the attention to the matrix information at other more stable time points.

[0067] In an optional embodiment, obtaining the channel attention of the differential spectrum matrix sequence according to the time-frequency information entropy value sequence includes:

[0068] Normalizing each entropy value in the time-frequency information entropy value sequence to obtain a normalized entropy value sequence;

[0069] Multiplying the normalized entropy value sequence by each matrix in the differential spectrum matrix sequence element by element to obtain a weighted differential spectrum matrix sequence;

[0070] Sum the rows and columns of each matrix in the weighted difference spectrum matrix sequence to obtain the channel attention vector;

[0071] Channel attention is obtained by normalizing the channel attention vector.

[0072] Specifically, assume that within a time window, the sequence of five time-frequency information entropy values ​​is [1.2, 1.5, 3.8, 1.4, 1.1]. Normalization converts these values ​​to the range [0, 1], resulting in a sequence of normalized entropy values. The first normalized entropy value is multiplied by each element in the first matrix, the second normalized entropy value is multiplied by each element in the second matrix, and so on. Since the third entropy value, 3.8, has the largest value, its corresponding normalized value is also the largest. Therefore, all values ​​in the third difference spectrum matrix are proportionally amplified, while the values ​​in the other matrices are amplified less. This results in a sequence of weighted difference spectrum matrices, where the overall value of the third matrix is ​​enhanced. For each matrix in the weighted difference spectrum matrix sequence, the values ​​of all its elements are summed, compressing the entire matrix into a single value. The third matrix is ​​amplified the most and has the largest value, resulting in a new vector consisting of five values, which is the channel attention vector. The channel attention vector is normalized again so that the sum of all elements is equal to 1 to obtain the channel attention, which is exemplarily [0.1, 0.15, 0.5, 0.15, 0.1].

[0073] In an optional embodiment, the extracting features of the time-frequency two-dimensional energy distribution diagram sequence based on the stability of the frequency components in the differential spectrum matrix includes:

[0074] For each time-frequency two-dimensional energy distribution graph in the time-frequency two-dimensional energy distribution graph sequence, obtaining the differential spectrum matrix at the same position of the sliding window;

[0075] Calculating the absolute value of each frequency component in the differential spectrum matrix to obtain a frequency instability vector;

[0076] Normalizing the frequency instability vector to obtain a frequency attention weight vector;

[0077] Multiplying each row of the time-frequency two-dimensional energy distribution map by the corresponding weight value in the frequency attention weight vector to obtain an enhanced time-frequency two-dimensional energy distribution map;

[0078] The enhanced time-frequency two-dimensional energy distribution map sequence is input into a two-dimensional convolutional neural network to obtain the characteristics of the time-frequency two-dimensional energy distribution map sequence.

[0079] Specifically, the two-dimensional time-frequency energy distribution map and the differential spectrum matrix correspond in time. If the differential spectrum matrix corresponding to the two-dimensional time-frequency energy distribution map is calculated, a frequency instability vector containing values ​​for four frequency bands is obtained, for example, [0.1, 0.2, 4.5, 0.2]. The instability vector shows that the signal in the third frequency band varies most dramatically, with an instability value of 4.5, while the other frequency bands are relatively stable. Normalizing the instability vector generates a frequency attention weight vector, which becomes [0.02, 0.04, 0.9, 0.04]. The time-frequency map consists of many rows and columns, with each row representing a specific frequency band. Multiplying the first row of data by a weight of 0.02, the second row by 0.04, the third row by 0.9, and the fourth row by 0.04 amplifies all features in the third frequency band, while features in other frequency bands are suppressed. This time-frequency graph with key frequency information is input into a two-dimensional convolutional neural network for feature extraction to obtain features that reflect the time-frequency characteristics of the signal.

[0080] In an optional embodiment, determining the fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window includes:

[0081] Calculating the fundamental wave amplitude and total harmonic distortion of each current data in the sliding window; normalizing the fundamental wave amplitude and total harmonic distortion to obtain normalized fundamental wave amplitude and normalized total harmonic distortion;

[0082] The normalized fundamental wave amplitude and the normalized total harmonic distortion are input into the weighting function to obtain the fusion weight.

[0083] Specifically, assume that the calculated average fundamental amplitude within the window is 10A, indicating a medium load level, and the average total harmonic distortion is 30%, indicating significant harmonic contamination in the line. These values ​​are then mapped to a normalized range of zero to one. Assuming the normal range of fundamental amplitude is 0 to 20A and the range of total harmonic distortion is 0 to 50%, the normalized fundamental amplitude is 0.5, and the normalized total harmonic distortion is 0.6. These normalized fundamental amplitude and total harmonic distortion are then input into a pre-defined weighting function, which calculates the fusion weights, for example, 0.25 and 0.75. This indicates that in the current medium load and severe harmonic contamination environment, the reliability of the differential spectrum analysis channel is low, while the results of the time-frequency analysis channel are more reliable. In one embodiment, the weighting function is implemented using a piecewise function. First, rules are established. For example, a rule might be: If total harmonic distortion (THD) is high, it indicates high line background noise. The reliability of the differential spectrum features decreases, and the time-frequency analysis features should be trusted more. Fusion weights are then determined based on these rules. In another embodiment, the weighting function is derived from historical data using a fitting function such as multivariate polynomial regression or support vector regression.

[0084] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. In addition, the various different implementations of the embodiments of the present invention can also be arbitrarily combined, as long as they do not violate the ideas of the embodiments of the present invention, and they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A self-test method for power distribution equipment based on an intelligent circuit breaker, characterized in that: include: Obtain current data of two continuous full-cycle alternating currents monitored by the intelligent circuit breaker, perform full-phase Fourier transform on the current data of the two continuous full-cycle alternating currents to obtain two full-phase spectra, and calculate a differential spectrum matrix based on the two full-phase spectra; calculate the spectral energy of the two full-phase spectra within a preset high-frequency band, use the current data of the full-cycle alternating current corresponding to the full-phase spectrum with the largest spectral energy as the target current data, perform continuous wavelet transform on the target current data to generate a time-frequency two-dimensional energy distribution map, and calculate a time-frequency information entropy value based on the time-frequency two-dimensional energy distribution map; A differential spectrum matrix sequence and a time-frequency two-dimensional energy distribution map sequence are obtained according to a preset sliding window length and step size, a time-frequency information entropy value sequence corresponding to the time-frequency two-dimensional energy distribution map sequence is obtained, a channel attention of the differential spectrum matrix sequence is obtained according to the time-frequency information entropy value sequence, and features of the differential spectrum matrix sequence are extracted using the channel attention; The characteristics of the time-frequency two-dimensional energy distribution sequence are extracted based on the stability of the frequency components in the differential spectrum matrix; A fusion weight is determined based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window, and the characteristics of the differential spectrum matrix sequence and the characteristics of the time-frequency two-dimensional energy distribution diagram sequence are weightedly fused using the fusion weight. A self-test result is obtained based on the result of the weighted fusion.

2. The method according to claim 1, characterized in that The channel attention of the differential spectrum matrix sequence is obtained according to the time-frequency information entropy value sequence, including: Inputting the time-frequency information entropy value sequence into a multi-layer perceptron, wherein the multi-layer perceptron includes at least one fully connected layer and one activation layer; The multilayer perceptron is used to perform a nonlinear transformation on the time-frequency information entropy value sequence to obtain attention weight vectors with the same number as the differential spectrum matrix sequence, and each value in the attention weight vector corresponds to a differential spectrum matrix in the differential spectrum matrix sequence.

3. The method according to claim 1, characterized in that The channel attention of the differential spectrum matrix sequence is obtained according to the time-frequency information entropy value sequence, including: Normalizing each entropy value in the time-frequency information entropy value sequence to obtain a normalized entropy value sequence; Multiplying the normalized entropy value sequence by each matrix in the differential spectrum matrix sequence element by element to obtain a weighted differential spectrum matrix sequence; Sum the rows and columns of each matrix in the weighted difference spectrum matrix sequence to obtain the channel attention vector; Channel attention is obtained by normalizing the channel attention vector.

4. The method according to claim 1, wherein The feature extraction of the time-frequency two-dimensional energy distribution diagram sequence based on the stability of the frequency components in the differential spectrum matrix includes: For each time-frequency two-dimensional energy distribution graph in the time-frequency two-dimensional energy distribution graph sequence, obtaining the differential spectrum matrix at the same position of the sliding window; Calculating the absolute value of each frequency component in the differential spectrum matrix to obtain a frequency instability vector; Normalizing the frequency instability vector to obtain a frequency attention weight vector; Multiplying each row of the time-frequency two-dimensional energy distribution map by the corresponding weight value in the frequency attention weight vector to obtain an enhanced time-frequency two-dimensional energy distribution map; The enhanced time-frequency two-dimensional energy distribution map sequence is input into a two-dimensional convolutional neural network to obtain the characteristics of the time-frequency two-dimensional energy distribution map sequence.

5. The method according to claim 1, wherein The determining of the fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window includes: Calculating the fundamental wave amplitude and total harmonic distortion of each current data in the sliding window; normalizing the fundamental wave amplitude and total harmonic distortion to obtain normalized fundamental wave amplitude and normalized total harmonic distortion; The normalized fundamental wave amplitude and the normalized total harmonic distortion are input into the weighting function to obtain the fusion weight.

6. A power distribution equipment self-test system based on intelligent circuit breaker, characterized in that: include: a preprocessing unit for acquiring current data of two continuous full-cycle alternating currents monitored by the intelligent circuit breaker, performing full-phase Fourier transform on the current data of the two continuous full-cycle alternating currents to obtain two full-phase spectra, and calculating a differential spectrum matrix based on the two full-phase spectra; calculating the spectral energy of the two full-phase spectra within a preset high-frequency band, taking the current data of the full-cycle alternating current corresponding to the full-phase spectrum with the largest spectral energy as the target current data, performing continuous wavelet transform on the target current data to generate a time-frequency two-dimensional energy distribution map, and calculating a time-frequency information entropy value based on the time-frequency two-dimensional energy distribution map; A feature extraction unit is used to obtain a differential spectrum matrix sequence and a time-frequency two-dimensional energy distribution map sequence according to a preset sliding window length and step size, obtain a time-frequency information entropy value sequence corresponding to the time-frequency two-dimensional energy distribution map sequence, obtain a channel attention of the differential spectrum matrix sequence based on the time-frequency information entropy value sequence, and use the channel attention to extract features of the differential spectrum matrix sequence; The characteristics of the time-frequency two-dimensional energy distribution sequence are extracted based on the stability of the frequency components in the differential spectrum matrix; A self-test unit is used to determine a fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window, use the fusion weight to weightedly fuse the characteristics of the differential spectrum matrix sequence and the characteristics of the time-frequency two-dimensional energy distribution diagram sequence, and obtain a self-test result based on the result of the weighted fusion.

7. The system according to claim 6, characterized in that The channel attention of the differential spectrum matrix sequence is obtained according to the time-frequency information entropy value sequence, including: Inputting the time-frequency information entropy value sequence into a multi-layer perceptron, wherein the multi-layer perceptron includes at least one fully connected layer and one activation layer; The multilayer perceptron is used to perform a nonlinear transformation on the time-frequency information entropy value sequence to obtain attention weight vectors with the same number as the differential spectrum matrix sequence, and each value in the attention weight vector corresponds to a differential spectrum matrix in the differential spectrum matrix sequence.

8. The system according to claim 6, wherein: The channel attention of the differential spectrum matrix sequence is obtained according to the time-frequency information entropy value sequence, including: Normalizing each entropy value in the time-frequency information entropy value sequence to obtain a normalized entropy value sequence; Multiplying the normalized entropy value sequence by each matrix in the differential spectrum matrix sequence element by element to obtain a weighted differential spectrum matrix sequence; Sum the rows and columns of each matrix in the weighted difference spectrum matrix sequence to obtain the channel attention vector; Channel attention is obtained by normalizing the channel attention vector.

9. The system according to claim 6, wherein: The feature extraction of the time-frequency two-dimensional energy distribution diagram sequence based on the stability of the frequency components in the differential spectrum matrix includes: For each time-frequency two-dimensional energy distribution graph in the time-frequency two-dimensional energy distribution graph sequence, obtaining the differential spectrum matrix at the same position of the sliding window; Calculating the absolute value of each frequency component in the differential spectrum matrix to obtain a frequency instability vector; Normalizing the frequency instability vector to obtain a frequency attention weight vector; Multiplying each row of the time-frequency two-dimensional energy distribution map by the corresponding weight value in the frequency attention weight vector to obtain an enhanced time-frequency two-dimensional energy distribution map; The enhanced time-frequency two-dimensional energy distribution map sequence is input into a two-dimensional convolutional neural network to obtain the characteristics of the time-frequency two-dimensional energy distribution map sequence.

10. The system according to claim 6, wherein: The determining of the fusion weight based on the fundamental amplitude and total harmonic distortion of the current data in the sliding window includes: Calculating the fundamental wave amplitude and total harmonic distortion of each current data in the sliding window; normalizing the fundamental wave amplitude and total harmonic distortion to obtain normalized fundamental wave amplitude and normalized total harmonic distortion; The normalized fundamental wave amplitude and the normalized total harmonic distortion are input into the weighting function to obtain the fusion weight.

Images