Leakage fault detection method and system for low-voltage distribution network

Abstract

The invention relates to the technical field of fault detection, in particular to a low-voltage distribution network electric leakage fault detection method and system, and the method comprises the steps: obtaining a residual current and zero-sequence current data set according to the three-phase current data and neutral current data of a low-voltage distribution network; performing opening and closing operation on the residual current and zero-sequence current data set by using a mathematical morphology method to obtain current signal local-global feature data; obtaining a leakage current change signal based on the current signal local-global feature data; based on a sparrow optimization algorithm fusing a sine and cosine search strategy and a Cauchy variation mechanism, constructing a low-voltage distribution network electric leakage fault detection model; and based on the current change signal, performing fault identification by using a low-voltage distribution network electric leakage fault detection model to obtain an electric leakage fault identification result. According to the invention, high-precision identification of the electric leakage fault of the low-voltage power distribution network is realized through the sparrow optimization algorithm fusing the sine and cosine search strategy and the Cauchy variation mechanism, and the accuracy and efficiency of electric leakage detection are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a method and system for detecting leakage faults in a low-voltage distribution network. Background Art

[0002] As the terminal energy supply link of the power system, the low-voltage distribution network directly faces the user-side load. Its line topology is complex and has numerous branch nodes. It is exposed to complex environmental conditions for a long time. Problems such as line insulation aging, three-phase load imbalance, and equipment grounding abnormality can easily cause leakage faults. The fault current characteristics have significant overlap with normal load switching, resulting in frequent false operation or refusal to operate of traditional residual current protection devices.

[0003] Traditional leakage detection technology mainly relies on residual current protection devices. Existing residual current protection devices mainly adopt typical technical routes such as amplitude comparison type, current pulse type, amplitude and phase detection type and current separation type. However, these traditional methods have significant limitations in practical applications. Although the amplitude comparison type device has a simple structure and low cost, it has an inherent protection dead zone. When the leakage current amplitude is lower than the setting threshold, it will refuse to operate, and it is easy to malfunction under high-order harmonic interference or transient processes; the current pulse type device improves its sensitivity by capturing the sudden change characteristics of the residual current, but it is easy to cause malfunction when unbalanced loads are switched, seriously affecting the continuity of power supply; the amplitude and phase detection type device introduces a phase identification mechanism based on the current pulse type. Although this technology can partially suppress the interference of pure resistive loads, it faces the problem of phase reference drift in practical applications. When the system voltage phase is offset or non-power frequency components are infiltrated, the phase identification unit will produce detection deviations, resulting in electric shock. The protection sensitivity drops by more than 40%. What is more serious is that when the leakage channel presents capacitive characteristics, the difference between the fault current phase and the normal switching current phase is reduced to an indistinguishable range, forming a new protection blind spot. The current separation device can theoretically separate the fault leakage signal from the total residual current, but in engineering practice, it faces technical obstacles such as the current characteristics of metallic grounding faults and electric shock faults in the initial stage are highly similar, and the existing algorithms are difficult to effectively distinguish. What is particularly critical is that the residual current protection devices currently operating online are usually based on the effective value of the residual current as the only criterion for action. This single indicator and this criterion often cannot effectively distinguish between leakage faults and normal load switching operations, resulting in frequent misoperation or refusal of the protection device, which seriously reduces its action reliability and correct commissioning rate. Therefore, the traditional leakage detection method faces the risk of failure in practical applications and cannot meet the high precision and real-time requirements of low-voltage distribution networks for leakage fault detection.

[0004] In summary, in view of the various shortcomings of the existing technology in leakage fault detection, there is an urgent need to provide a leakage fault detection method with high precision and high reliability to improve the fault detection capability of the low-voltage distribution network and ensure its safe and stable operation. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for detecting leakage faults in a low-voltage distribution network.

[0006] In a first aspect, the present invention provides a method for detecting leakage faults in a low-voltage distribution network, the method comprising the following steps:

[0007] According to the three-phase current data and the neutral line current data of the low-voltage distribution network, a residual current and a zero-sequence current data group are obtained;

[0008] Using mathematical morphology method to perform opening and closing operations on the residual current and zero-sequence current data group to obtain local-global characteristic data of current signal;

[0009] The local-global characteristic data of the current signal is converted into a residual-zero-sequence current mutation ratio signal, and a leakage current change signal is calculated;

[0010] Based on the sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism, a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM is constructed;

[0011] Based on the leakage current change signal, the low-voltage distribution network leakage fault detection model is used to perform fault identification to obtain a leakage fault identification result.

[0012] In a further embodiment, the step of obtaining the residual current and zero-sequence current data set according to the three-phase current data and the neutral current data of the low-voltage distribution network comprises:

[0013] Synchronously collecting three-phase current data and neutral current data of the low-voltage distribution network, and obtaining residual current data according to the vector sum of the three-phase current data and the neutral current data;

[0014] Calculating the root mean square value of the residual current data using a sliding window to obtain a residual current effective value sequence;

[0015] Analyzing the three-phase current data using a symmetrical component method to obtain a zero-sequence current phasor;

[0016] Reconstructing a zero-sequence current time-domain waveform according to the amplitude information and phase information of the zero-sequence current phasor, and performing a wavelet transform on the zero-sequence current time-domain waveform to extract transient characteristics;

[0017] The residual current effective value sequence and the transient characteristics are combined to form a residual current and zero-sequence current data set.

[0018] In a further embodiment, the step of analyzing the three-phase current data using the symmetrical component method to obtain the zero-sequence current phasor comprises:

[0019] Performing Fourier transform on the three-phase current data respectively to extract the fundamental wave component amplitude information and fundamental wave component phase information of each phase;

[0020] Generate three-phase current phasors according to the fundamental component amplitude information and the fundamental component phase information;

[0021] The three-phase current phasor is vector-calculated by using the symmetrical component method to obtain the zero-sequence current phasor.

[0022] In a further embodiment, the step of performing opening and closing operations on the residual current and zero-sequence current data groups using a mathematical morphology method to obtain local-global characteristic data of the current signal comprises:

[0023] Determining linear structural elements at different scales according to the sampling rate and noise frequency of the residual current and zero-sequence current data group;

[0024] Performing corrosion expansion boundary processing on both ends of the residual current and zero-sequence current data group by using a mirror filling method to obtain a residual current and zero-sequence current filling data group;

[0025] Performing an erosion operation on the residual current and zero-sequence current filling data group according to the linear structural element to obtain an erosion result, and performing an expansion operation on the erosion result to obtain a local smoothing feature;

[0026] Performing an expansion operation on the residual current and zero-sequence current filling data group according to the linear structural element to obtain an expansion result, and performing an erosion operation on the expansion result to obtain a global fluctuation feature;

[0027] Determining the fusion weight of the local smooth feature and the global fluctuation feature according to the signal-to-noise ratio of the local smooth feature and the signal-to-noise ratio of the global fluctuation feature at different linear structure element scales;

[0028] The local smoothness feature and the global fluctuation feature are weightedly fused based on the fusion weight to obtain local-global feature data of the current signal.

[0029] In a further embodiment, the step of converting the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal and calculating the leakage current change signal comprises:

[0030] Extracting residual current local-global characteristic data and zero-sequence current local-global characteristic data in a normal state from the current signal local-global characteristic data;

[0031] Eliminate the first power frequency cycle data in the residual current local-global characteristic data and the zero-sequence current local-global characteristic data in the normal state to obtain a residual-zero-sequence current normal data group;

[0032] The normal residual-zero-sequence current data group is processed by using a sliding time window to calculate the residual current average value and the zero-sequence current average value;

[0033] Obtaining a mutation ratio constant according to a ratio between the residual current average value and the zero-sequence current average value;

[0034] Calculating the residual current characteristic change and the zero-sequence current characteristic change of each sampling point according to the local-global characteristic data of the current signal;

[0035] According to the residual current characteristic change and the zero-sequence current characteristic change, the leakage current change value is obtained by weighted calculation through mutation ratio constant.

[0036] In a further implementation scheme, the steps of constructing a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on the sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism include:

[0037] The convolutional neural network and the bidirectional long short-term memory network are cascaded to construct the CNN-BiLSTM sequence prediction model;

[0038] The sparrow optimization algorithm that integrates the sine and cosine search strategy and the Cauchy mutation mechanism is used to globally optimize the multi-dimensional parameter space of the CNN-BiLSTM sequence prediction model to obtain the feature optimization parameters;

[0039] The feature optimization parameters are mapped to the CNN-BiLSTM sequence prediction model to construct a low-voltage distribution network leakage fault detection model;

[0040] The historical leakage current change values ​​are converted into a binary image matrix, and the low-voltage distribution network leakage fault detection model is trained using the binary image matrix to obtain a trained low-voltage distribution network leakage fault detection model.

[0041] In a further embodiment, the feature optimization parameters include convolutional neural network structure parameters, model hyperparameters, and the number of hidden layer neurons in the bidirectional long short-term memory network; the model hyperparameters include learning rate and L2 regularization coefficient.

[0042] In a further embodiment, the step of performing fault identification based on the leakage current change signal and using the low-voltage distribution network leakage fault detection model to obtain a leakage fault identification result comprises:

[0043] Converting the leakage current change signal into a leakage current change binary image, and inputting the leakage current change binary image into the low-voltage distribution network leakage fault detection model for fault identification, to obtain a leakage fault identification result;

[0044] If the leakage fault identification result is that the line is in normal operation or load switching occurs, continue to monitor the leakage current change signal;

[0045] If the leakage fault identification result is that a leakage fault occurs in the line, a leakage fault alarm mechanism is triggered.

[0046] In a further embodiment, the step of converting the leakage current change signal into a leakage current change binary image comprises:

[0047] Converting the leakage current change signal into an image format to obtain an original leakage current change image;

[0048] Performing a cropping process on the original leakage current change image to extract a cropped image containing a leakage current change value curve;

[0049] The cropped image is converted into a grayscale image, and the grayscale image is globally binarized to obtain a leakage current change binary image.

[0050] In a second aspect, the present invention provides a low-voltage distribution network leakage fault detection system, the system comprising:

[0051] A data acquisition module is used to obtain a residual current and a zero-sequence current data group according to the three-phase current data and the neutral current data of the low-voltage distribution network;

[0052] A data processing module, used for performing opening and closing operations on the residual current and zero-sequence current data groups using a mathematical morphology method to obtain local-global characteristic data of the current signal;

[0053] A mutation analysis module, used for converting the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal, and calculating a leakage current change signal;

[0054] Model building module, which is used to build a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on the sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism;

[0055] The fault identification module is used to perform fault identification based on the leakage current change signal and utilize the low-voltage distribution network leakage fault detection model to obtain a leakage fault identification result.

[0056] The present invention provides a low-voltage distribution network leakage fault detection method and system. The method obtains residual current and zero-sequence current data groups according to three-phase current data and neutral line current data of the low-voltage distribution network; performs opening and closing operations on the residual current and zero-sequence current data groups using a mathematical morphology method to obtain local-global characteristic data of current signals; converts the local-global characteristic data of current signals into residual-zero-sequence current mutation ratio signals, and calculates leakage current change signals; constructs a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on a sparrow optimization algorithm that integrates a sine-cosine search strategy and a Cauchy mutation mechanism; and performs fault identification using a low-voltage distribution network leakage fault detection model based on the leakage current change signal to obtain leakage fault identification results. Compared with the prior art, the method comprehensively utilizes residual current signals and zero-sequence current signals, and realizes high-precision identification of low-voltage distribution network leakage faults through a sparrow optimization algorithm that integrates a sine-cosine search strategy and a Cauchy mutation mechanism, effectively improving the accuracy and efficiency of leakage detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the flow of a method for detecting leakage faults in a low-voltage distribution network provided by an embodiment of the present invention;

[0058] Figure 2 It is a schematic diagram of a low-voltage distribution network leakage fault detection process provided by an embodiment of the present invention;

[0059] Figure 3 This is an example diagram of a 400V distribution network architecture provided by an embodiment of the present invention;

[0060] Figure 4 It is a schematic diagram of the confusion matrix of the SCSSA-CNN-BiLSTM model test set provided by an embodiment of the present invention;

[0061] Figure 5 Schematic diagram of the confusion matrix of the CNN-BiLSTM model test set provided by an embodiment of the present invention;

[0062] Figure 6 The present invention provides a low-voltage distribution network leakage fault detection system block diagram. DETAILED DESCRIPTION

[0063] The following specifically illustrates the implementation mode of the present invention in conjunction with the accompanying drawings. The embodiments are provided for illustrative purposes only and cannot be understood as limiting the present invention. The accompanying drawings are provided for reference and illustration only and do not constitute a limitation on the scope of patent protection of the present invention, because many changes can be made to the present invention without departing from the spirit and scope of the present invention.

[0064] refer to Figure 1 The embodiment of the present invention provides a method for detecting leakage faults in a low-voltage distribution network. Figure 1 As shown, the method comprises the following steps:

[0065] S1. Obtain a residual current and zero-sequence current data set based on the three-phase current data and neutral line current data of the low-voltage distribution network.

[0066] In some embodiments, the step of obtaining the residual current and zero-sequence current data set according to the three-phase current data and the neutral current data of the low-voltage distribution network includes:

[0067] Synchronously collecting three-phase current data and neutral current data of the low-voltage distribution network, and obtaining residual current data according to the vector sum of the three-phase current data and the neutral current data;

[0068] Calculating the root mean square value of the residual current data using a sliding window to obtain a residual current effective value sequence;

[0069] Analyzing the three-phase current data using a symmetrical component method to obtain a zero-sequence current phasor;

[0070] Reconstructing a zero-sequence current time-domain waveform according to the amplitude information and phase information of the zero-sequence current phasor, and performing a wavelet transform on the zero-sequence current time-domain waveform to extract transient characteristics;

[0071] The residual current effective value sequence and the transient characteristics are combined to form a residual current and zero-sequence current data set.

[0072] Specifically, in this embodiment, high-precision current transformers are installed at key nodes of the low-voltage distribution network. The current transformers are used to collect three-phase current data of phase A, phase B, and phase C and neutral line (N line) current data in real time. In this embodiment, synchronous sampling technology is used to ensure that the three-phase current and neutral line current data are collected at the same time point to avoid data errors caused by time differences. The collected current data is uploaded to the main station or data processing center in real time through a communication network (such as optical fiber, wireless, etc.) through a feeder terminal unit (FTU) in the distribution network. Then, the uploaded three-phase current data and neutral line current data are preprocessed, including denoising, filtering, etc., to improve data quality. According to the preprocessed three-phase current data and neutral line current data, the A-phase, B-phase, C-phase current data and the neutral line current data are vector-added to obtain residual current data. Since the three-phase currents cancel each other out under ideal conditions, the residual current mainly reflects the sum of the neutral line current and the three-phase unbalanced current.

[0073] At the same time, this embodiment sets a sliding window of a fixed length, and the window length can be determined according to actual needs, such as one power frequency cycle (50Hz is 0.02 seconds). In each sliding window, the residual current data is calculated for the root mean square value to obtain the residual current effective value in the window. After each calculation, the window slides forward by one sampling point, and continues to calculate the root mean square value of the residual current in the next window to form a residual current effective value sequence. Then, the three-phase current data is converted into a symmetrical component form, that is, a positive sequence, a negative sequence and a zero sequence component. Since the zero sequence current phasor reflects the degree of asymmetry of the three-phase current, this embodiment extracts the zero sequence current phasor from the symmetrical component. In some embodiments, the step of analyzing the three-phase current data using the symmetrical component method to obtain the zero sequence current phasor includes:

[0074] Performing Fourier transform on the three-phase current data respectively to extract the fundamental wave component amplitude information and fundamental wave component phase information of each phase;

[0075] Generate three-phase current phasors according to the fundamental component amplitude information and the fundamental component phase information;

[0076] The three-phase current phasor is vector-calculated by using the symmetrical component method to obtain the zero-sequence current phasor.

[0077] In this embodiment, the three-phase current data of the low-voltage distribution network collected by the current transformer (CT) is stored in the form of a discrete time series, and the collected three-phase current data are respectively subjected to fast Fourier transform (FFT). FFT can convert the time domain signal into a frequency domain signal, thereby extracting the amplitude and phase information of each frequency component, and finding the amplitude and phase information corresponding to the fundamental frequency in the frequency domain. These information represent the fundamental component of the three-phase current, and the amplitude and phase information of the fundamental component extracted from the FFT are used to construct the phasor representation of the three-phase current. The transformation formula of the three-phase current phasor is used to calculate the zero-sequence current phasor through the symmetrical component method, and then the time domain waveform of the zero-sequence current is reconstructed according to the amplitude and phase information of the zero-sequence current phasor. The mathematical expression of the time domain waveform of the zero-sequence current is:

[0078]

[0079] In the formula, I 零序 (t) is the zero-sequence current time domain waveform at time t; |I0| is the amplitude of the zero-sequence current phasor; w is the angular frequency; is the phase of the zero-sequence current phasor.

[0080] In this embodiment, wavelet transform is performed on the reconstructed zero-sequence current time domain waveform to extract transient features. Wavelet transform can well capture local changes of the signal, thereby extracting transient signal features, and standardizing the extracted transient features to improve the stability and reliability of subsequent analysis. In this embodiment, the calculated residual current effective value sequence and the extracted zero-sequence current transient features are combined to form a complete residual current and zero-sequence current data group. The time span of the data group should be consistent, which is an integer multiple of the power frequency period and contains a complete transient process. The combined data group is stored in a database for subsequent analysis and processing. In the TN-C system, since the neutral line (N line) and the protective grounding line (PE line) are combined into one line (PEN line), therefore, in practical applications, when collecting power grid signals, it is necessary to comprehensively consider a variety of working conditions to ensure the applicability of subsequent fault identification models. Specifically, signal collection under normal power grid conditions must take into account changes in load intensity, the number of repeated grounding points, and differences in collection lines; when a leakage fault occurs in the power grid, special attention should be paid to transient signal characteristics such as the resistance of the transition resistor and the distance between the leakage fault point and the signal collection point; and during the load switching process, attention should be paid to transient information such as the size of the switched load and the distance between the switching point and the signal collection point. To ensure the validity and comparability of the data, the time spans of these three types of signal groups should be consistent, for example, they should preferably be integer multiples of the power frequency cycle, and should include a complete transient process.

[0081] S2. Using mathematical morphology methods, open and close operations are performed on the residual current and zero-sequence current data groups to obtain local-global characteristic data of the current signal.

[0082] In some embodiments, the step of performing opening and closing operations on the residual current and zero-sequence current data group using a mathematical morphology method to obtain local-global characteristic data of the current signal includes:

[0083] Determining linear structural elements at different scales according to the sampling rate and noise frequency of the residual current and zero-sequence current data group;

[0084] Performing corrosion expansion boundary processing on both ends of the residual current and zero-sequence current data group by using a mirror filling method to obtain a residual current and zero-sequence current filling data group;

[0085] Performing an erosion operation on the residual current and zero-sequence current filling data group according to the linear structural element to obtain an erosion result, and performing an expansion operation on the erosion result to obtain a local smoothing feature;

[0086] Performing an expansion operation on the residual current and zero-sequence current filling data group according to the linear structural element to obtain an expansion result, and performing an erosion operation on the expansion result to obtain a global fluctuation feature;

[0087] Determining the fusion weight of the local smooth feature and the global fluctuation feature according to the signal-to-noise ratio of the local smooth feature and the signal-to-noise ratio of the global fluctuation feature at different linear structure element scales;

[0088] The local smoothness feature and the global fluctuation feature are weightedly fused based on the fusion weight to obtain local-global feature data of the current signal.

[0089] Specifically, this embodiment analyzes the sampling rate and noise frequency of the residual current and zero-sequence current data groups, wherein the sampling rate determines the time resolution of the signal, and the noise frequency affects the design rate of the structural element. According to the sampling rate and the noise frequency, the length of the linear structural element used at different scales is determined, so that the linear structural element is obtained according to the length of the linear structural element. These elements are at a zero degree angle to the horizontal direction, and the lengths are from short to long to adapt to the noise and signal characteristics of different scales. It should be noted that the length and direction of the structural element should match the characteristics of the signal. For example, for high-frequency noise, this embodiment can select a shorter structural element; for low-frequency noise, this embodiment can select a longer structural element. The direction of the structural element is usually consistent with the main direction of the signal, such as the horizontal direction. The length calculation formula of the linear structural element is:

[0090]

[0091] Where L is the length of the linear structure element; round is the rounding operation; f sis the sampling rate; f noise The main frequency of the noise is determined by performing FFT spectrum analysis on the residual current and zero-sequence current data groups.

[0092] At the same time, in order to deal with the boundary effects at both ends of the signal, this embodiment uses a mirror filling method to mirror fill the residual current and zero-sequence current data groups to obtain a filled data group, so that there will be no boundary problems when performing subsequent corrosion and expansion operations. The mirror filling method extends the two ends of the signal outward respectively, and the extended part generates a filled data group by mirroring the part inside the signal. Then, this embodiment uses the selected linear structure element to perform corrosion operation on the filled residual current and zero-sequence current data groups. The corrosion operation can smooth the signal and suppress peak noise, and the corrosion result is expanded to restore some details of the signal, while removing the concave part introduced by the corrosion operation. After the opening operation (corrosion + expansion), local smooth features are obtained. These features reflect the smooth changes of the signal at a smaller scale; the corrosion operation is defined as:

[0093]

[0094] Wherein, f is the residual current and zero-sequence current filling data group; b is the linear structure element; J is the total number of linear structure elements; m is the mth linear structure element.

[0095] The dilation operation is defined as:

[0096]

[0097] Similarly, this embodiment uses a linear structural element to perform an expansion operation on the filled data group. The expansion operation can suppress the signal trough noise, and the expansion result is corroded to remove the convex part introduced by the expansion operation, while retaining the main fluctuation trend of the signal. After the closing operation (expansion + corrosion), the global fluctuation characteristics are obtained. These characteristics reflect the overall change trend of the signal at a larger scale. This embodiment calculates the signal-to-noise ratio (SNR) of the local smoothness characteristics and the global fluctuation characteristics at different scales. SNR is an important indicator for measuring signal quality. A high SNR means that there is more useful information and less noise in the signal. According to the SNR of the local smoothness characteristics and the global fluctuation characteristics, the fusion weight is determined. This embodiment assigns a greater weight to the feature with a higher SNR, and based on the determined fusion weight, the local smoothness characteristics and the global fluctuation characteristics are weightedly fused to obtain the local-global feature data of the current signal. The fused feature data not only contains the local details of the signal, but also reflects the overall trend of the signal. The local-global feature data finally obtained can be used for subsequent fault detection, classification and analysis.

[0098] S3. Convert the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal, and calculate the leakage current change signal.

[0099] In some embodiments, the step of converting the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal and calculating the leakage current change signal comprises:

[0100] Extracting residual current local-global characteristic data and zero-sequence current local-global characteristic data in a normal state from the current signal local-global characteristic data;

[0101] Eliminate the first power frequency cycle data in the residual current local-global characteristic data and the zero-sequence current local-global characteristic data in the normal state to obtain a residual-zero-sequence current normal data group;

[0102] The normal residual-zero-sequence current data group is processed by using a sliding time window to calculate the residual current average value and the zero-sequence current average value;

[0103] Obtaining a mutation ratio constant according to a ratio between the residual current average value and the zero-sequence current average value;

[0104] Calculating the residual current characteristic change and the zero-sequence current characteristic change of each sampling point according to the local-global characteristic data of the current signal;

[0105] According to the residual current characteristic change and the zero-sequence current characteristic change, the leakage current change value is obtained by weighted calculation of the mutation ratio constant; wherein, the calculation formula of the leakage current change value is:

[0106] ΔI LD =ΔI r -R c ΔI0

[0107] In the formula, ΔI LD is the leakage current change value; ΔI r is the residual current characteristic change; R c is the mutation ratio constant; ΔI0 is the change in zero-sequence current characteristics.

[0108] Specifically, this embodiment selects the residual current and zero-sequence current data in a normal state from the local-global characteristic data of the current signal. Since the data of the first power frequency cycle may contain unstable factors in the startup process, the number of sampling points contained in the first power frequency cycle is calculated, and these points are removed from the residual current normal data set and the zero-sequence current normal data set to ensure the accuracy of subsequent calculations. Then, the data after the first power frequency cycle is removed is processed using a sliding time window. For the data in each time window, this embodiment calculates the average values ​​of the residual current and the zero-sequence current to form two new Datasets: residual current average value data set and zero-sequence current average value data set. For the data in each time window, this embodiment calculates the ratio of the residual current average value to the zero-sequence current average value to obtain the mutation ratio constant. These constants will be used for the subsequent calculation of the leakage current change value. Taking the TN-C line without branches and repeated grounding at the end as an example, the constant (mutation ratio constant) is mainly significantly affected by the zero line impedance of the low-voltage station area, the impedance of the neutral point of the station transformer to the ground, and the impedance of the repeated grounding point to the ground. This embodiment calculates the constant R according to the zero line impedance, the impedance of the neutral point of the station transformer to the ground, and the impedance of the repeated grounding point to the ground. c , so it is possible to determine whether a leakage fault occurs in the line by setting the value. Specifically, in a short period of time, these impedances will not change suddenly, so R c It can be regarded as a constant. When the line is in a normal state, the residual current characteristic change is 0, the zero-sequence current characteristic change is 0, so the leakage current change is close to 0; when the line load is switched, the ratio of the residual current characteristic change to the zero-sequence current characteristic change is approximately equal to the mutation ratio constant R c , so the leakage current change value is also close to 0 (because the mutation ratio is approximately equal to the fixed value); when a leakage fault occurs in the line, the ratio of the residual current characteristic change to the zero-sequence current characteristic change (residual-zero-sequence current mutation ratio) is not equal to the mutation ratio fixed value R c , so the leakage current change value will significantly deviate from 0. Therefore, this embodiment can introduce the residual-zero sequence current mutation ratio to avoid the interference of load switching on the judgment of leakage. Taking the TN-C line without branches and repeated grounding at the end as an example, the mutation ratio constant R c The calculation formula is:

[0109]

[0110] In the formula, Z n Zero line impedance; Z sg is the impedance of the transformer neutral point to ground; Z rg is the impedance of the repeated grounding point to ground.

[0111] At the same time, for each sampling point, this embodiment calculates the difference between the residual current and the zero-sequence current of each sampling point and the previous sampling point according to the local-global characteristic data of the current signal, and obtains the residual current characteristic change amount and the zero-sequence current characteristic change amount. For each sampling point, this embodiment uses its corresponding mutation ratio constant to perform weighted summation on the residual current characteristic change amount and the zero-sequence current characteristic change amount to obtain a leakage current change value, which will be used to determine whether a leakage fault occurs, thereby converting the residual current and zero-sequence current data groups into leakage current change signals in batches.

[0112] S4. Based on the sparrow optimization algorithm that integrates the sine-cosine search strategy and the Cauchy mutation mechanism, a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM is constructed.

[0113] In some embodiments, the step of constructing a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on the sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism includes:

[0114] The convolutional neural network and the bidirectional long short-term memory network are cascaded to construct the CNN-BiLSTM sequence prediction model;

[0115] The sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism is used to globally optimize the multidimensional parameter space of the CNN-BiLSTM sequence prediction model to obtain feature optimization parameters; the feature optimization parameters include convolutional neural network structure parameters, model hyperparameters, and the number of neurons in the hidden layer of the bidirectional long short-term memory network; the model hyperparameters include learning rate and L2 regularization coefficient;

[0116] The feature optimization parameters are mapped to the CNN-BiLSTM sequence prediction model to construct a low-voltage distribution network leakage fault detection model;

[0117] The historical leakage current change values ​​are converted into a binary image matrix, and the low-voltage distribution network leakage fault detection model is trained using the binary image matrix to obtain a trained low-voltage distribution network leakage fault detection model.

[0118] Specifically, this embodiment collects the historical data of leakage current of the low-voltage distribution network, and pre-processes the historical data of leakage current, such as denoising and normalization, to ensure the quality and consistency of the data. At the same time, a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM) structure are set. The structure of the convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, etc. The convolutional neural network is used to extract the spatial features of the input data, such as Figure 2 As shown in the figure, the construction process of the convolutional neural network includes:

[0119] In this embodiment, the image label is first converted into a vector form, wherein the label of the normal state and the load switching is represented by [1,0], and the label of the leakage fault is represented by [0,1]. The label of the entire data set constitutes a digital matrix of [i,2]. Then, the one-dimensional data set of the binary image of the leakage current change value finally obtained is input into the convolution layer for feature extraction. The output of the convolution layer is then nonlinearly mapped through the sigmoid activation function to map the input value to the interval of [0,1]. The output of the activation function is then input into the pooling layer to reduce the dimension of the image data set. This process includes dividing the data input into the pooling layer into multiple matrices. shaped area, and output the maximum value from each rectangular area, so as to reduce the number of parameters and the amount of calculation, thereby achieving dimensionality reduction. Finally, this embodiment stacks the convolution layer and the pooling layer, and outputs the integrated feature information to the dropout layer. When constructing the dropout layer, the image data set after dimensionality reduction processing is input to the flat layer for integration, and then the integrated image information is output to the dropout layer. In the forward propagation process of the dropout layer, this embodiment sets the dropout parameter to 0.5, which means that a set proportion of feature detectors will be randomly ignored in each iteration, so as to reduce the interaction between detectors, thereby reducing the impact of overfitting.

[0120] At the same time, the present embodiment can set a bidirectional long short-term memory network including an input layer, a BiLSTM layer, a fully connected layer, etc. The BiLSTM layer is used to process the feature sequence extracted by the CNN layer and capture the time dependency in the sequence data. The present embodiment uses the output of CNN as the input of BiLSTM, that is, the convolutional neural network and the bidirectional long short-term memory network are cascaded to form a CNN-BiLSTM sequence prediction model. This combination can simultaneously utilize the spatial feature extraction capability of CNN and the time dependency capture capability of BiLSTM. In the bidirectional long short-term memory network, the present embodiment selectively allows feature information to pass through the neural layer of the sigmoid activation function and a point-by-point multiplication operation to realize the construction of the forget gate. The construction of the forget gate includes:

[0121] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0122] In the formula, f t is the forget gate; σ is the sigmoid activation function; W f is the weight coefficient of the forget gate; h t-1 is the input at time (t-1); b f is the bias constant of the forget gate; xt The input at the current moment.

[0123] The construction of the input gate structure includes:

[0124] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0125] c t =tanh(W c ·[h t-1 ,x t ]+b c )

[0126] y t =f t *y t-1 +i t *c t

[0127] In the formula, i t is the output of the sigmoid input gate unit activation function; W i is the weight coefficient of the input gate structure; b i is the bias constant of the input gate; c t is the output of the tan function of the input gate unit; W c is the output weight coefficient of the tan function of the input gate unit; tanh is the activation function of tanh; b c is the output bias constant of the input gate unit tan function; y t is the updated cell state; y t-1 This is the cell state when not updated.

[0128] The construction process of the output gate structure is:

[0129] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0130] h t =o t *tanh(y t )

[0131] In the formula, o t is the output of the sigmoid function of the input unit; W o is the weight coefficient of the output gate structure; b o is the bias constant of the output gate; h tis the total output value of the bidirectional long short-term memory neural network structure.

[0132] The total output value of the bidirectional long short-term memory neural network structure at time t is the sum of the outputs of the forward long short-term memory neural network and the backward long short-term memory neural network. The calculation of the total output value includes:

[0133]

[0134] In the formula, represents the output of the forward long short-term memory neural network; Represents the output of the backward long short-term memory neural network; Represents a vector sum operation.

[0135] This embodiment sets the basic parameters of the sparrow search algorithm (SSA). The basic parameters of the sparrow search algorithm (SSA) may include parameters such as population size and number of iterations. Each individual in the population represents a set of parameters of the CNN-BiLSTM model. At the same time, the sine-cosine search strategy and the Cauchy mutation mechanism are introduced. The sine-cosine function is used to generate the search direction to guide the sparrow to perform a global search in the parameter space. The sine-cosine search strategy can increase the diversity and randomness of the search to enhance the search ability of the algorithm and avoid falling into the local optimum. The formula is:

[0136] S new =S+A×cos(ω×k)

[0137] Where S is the current solution; A is the step size factor; ω is the frequency of the sine and cosine functions; and k is the current number of iterations.

[0138] At the same time, the Cauchy distribution is introduced to generate variable step lengths, and the positions of sparrows are mutated to enhance the local search ability of the algorithm. The Cauchy distribution has a heavy-tailed characteristic, which can increase the randomness and globality of the search. The mutation process can generate a random number through the Cauchy distribution, and adjust the position of the current individual according to the value to increase the diversity of the search process. The probability density function of the Cauchy distribution is:

[0139]

[0140] Where f(x) is the probability density function of the Cauchy distribution; γ is the scale parameter; x0 is the location parameter; and x is the current location.

[0141] In this embodiment, CNN structural parameters (such as convolution kernel size and number, etc.), model hyperparameters (such as learning rate, L2 regularization coefficient, etc.) and the number of BiLSTM hidden layer neurons are encoded as optimization variables of the algorithm, and the multi-dimensional parameter space of the CNN-BiLSTM sequence prediction model is globally optimized using the sparrow optimization algorithm that integrates the sine-cosine search strategy and the Cauchy mutation mechanism. The algorithm updates the individuals in the population through continuous iteration, and finally obtains the optimal feature optimization parameter combination. The optimal feature optimization parameters obtained by the SCSSA algorithm are mapped to the CNN-BiLSTM sequence prediction model, and the structure and hyperparameters of the model are adjusted. According to the mapped parameters, a low-voltage distribution network leakage fault detection model is constructed. The model can effectively process multi-dimensional time series data, capture the spatial characteristics and time dependencies in the data, and thus improve the accuracy and reliability of leakage fault detection. In this embodiment, the historical leakage current change value is converted into a binary image matrix, and the leakage current value is divided into two categories, high and low, and represented by 1 and 0 respectively, so as to form a binary image matrix. The converted binary image matrix is ​​used to train the low-voltage distribution network leakage fault detection model. During the training process, the embodiment continuously adjusts the weights and biases of the model so that the model can better fit the training data and predict leakage faults. After the training is completed, the embodiment uses a test data set to evaluate the model and calculates the accuracy of leakage detection. The accuracy is the ratio of the number of correctly identified leakage faults to the total number of test samples to ensure the performance and reliability of the model, so that a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM can be constructed and trained. The model can use historical leakage current data to predict future leakage conditions and accurately determine whether there is a leakage fault.

[0142] S5. Based on the leakage current change signal, the low-voltage distribution network leakage fault detection model is used to perform fault identification to obtain a leakage fault identification result.

[0143] In some implementations, the step of performing fault identification based on the leakage current change signal using the low-voltage distribution network leakage fault detection model to obtain a leakage fault identification result includes:

[0144] Converting the leakage current change signal into a leakage current change binary image, and inputting the leakage current change binary image into the low-voltage distribution network leakage fault detection model for fault identification, to obtain a leakage fault identification result;

[0145] If the leakage fault identification result is that the line is in normal operation or load switching occurs, continue to monitor the leakage current change signal;

[0146] If the leakage fault identification result is that a leakage fault occurs in the line, a leakage fault alarm mechanism is triggered.

[0147] Wherein, the step of converting the leakage current change signal into a leakage current change binary image comprises:

[0148] Converting the leakage current change signal into an image format to obtain an original leakage current change image;

[0149] Performing a cropping process on the original leakage current change image to extract a cropped image containing a leakage current change value curve;

[0150] The cropped image is converted into a grayscale image, and the grayscale image is globally binarized to obtain a leakage current change binary image.

[0151] Specifically, this embodiment uses Python's image processing library PIL (Pillow) to convert the time series data of the leakage current change signal into an image format. The leakage current change value at each time point corresponds to a pixel value in the image. The converted original leakage current change image is cropped to remove parts that are not related to the leakage current change value curve, such as blank areas or redundant information at the edge of the image, and extract the core area containing the leakage current change value curve. Then, the cropped color image is converted into a grayscale image, the color information of the image is removed, and only the brightness information is retained to simplify subsequent processing steps. This embodiment performs global binarization on the grayscale image, sets the pixel points with grayscale values ​​higher than the threshold to white (or 1), and sets the pixel points with grayscale values ​​lower than the threshold to black (or 0), to obtain a leakage current change binary image, and converts the binary image into a one-dimensional array. Each binary image contains a fixed number of pixels, so the obtained data set is a tensor with a shape of (total number of images, number of pixels), and the value in each one-dimensional array represents the proportion of black (or white) in the corresponding pixel point (in the binary image, the proportion is 0 or 1).

[0152] This embodiment uses a one-dimensional array of a binary image of leakage current changes (i.e., an array representation of the binary image of leakage current changes) as the input of a leakage fault detection model for a low-voltage distribution network. Inside the model, this embodiment first uses a convolutional neural network (CNN) to extract low-level features from the input one-dimensional array, which includes steps such as convolution operations, activation functions (such as sigmoid), and pooling operations to gradually extract and integrate feature information in the image. Specifically, this embodiment converts image labels into vector form, and the labels of normal states and load switching are represented as [1, 0], and the labels of leakage faults are represented as [0, 1]. The data set label is a (M, 2) digital matrix. Next, this embodiment uses a bidirectional long short-term memory network (BiLSTM) to capture the temporal dependency in the leakage current change data set. BiLSTM can process forward and reverse time series information at the same time, thereby improving the model's understanding and prediction capabilities for time series data.

[0153] During the model training process, this embodiment uses the sparrow optimization algorithm (SCSSA) that integrates sine, cosine and Cauchy variations to optimize the feature extraction process of CNN, the temporal dependency capture process of BiLSTM, and the hyperparameters of the model (such as learning rate, regularization parameter, etc.). This helps to enhance the representation ability and generalization performance of the model. Finally, the output of the neural network structure and the feature information are used as the input of the output layer of the leakage detection model to obtain the leakage fault recognition result. The output result may be a label such as the line is in normal operation, load switching occurs, or a leakage fault occurs. According to the leakage fault identification result, it is determined whether the line is in normal operation, load switching occurs, or a leakage fault occurs. If the identification result is that the line is in normal operation or load switching occurs, the leakage current change signal continues to be monitored, and preparations are made for the next fault identification; if the identification result is that a leakage fault occurs in the line, the leakage fault alarm mechanism is triggered, and relevant personnel are notified in time for processing. In summary, this embodiment can perform fault identification based on the leakage current change signal and the low-voltage distribution network leakage fault detection model, and obtain accurate leakage fault identification results, and take corresponding measures according to the results to improve the safety and reliability of the power grid.

[0154] In order to verify the effectiveness of the low-voltage distribution network leakage fault detection method based on the SCSSA-CNN-Bi LSTM model, this embodiment uses the traditional technical solution to compare with the low-voltage distribution network leakage fault detection method. In the verification test, this embodiment selects the following Figure 3The 400V distribution network shown in the figure is used as the test object. In order to comprehensively evaluate the performance of the method, this embodiment simulates 8 groups of leakage faults and 8 groups of load switching events at different locations of the distribution network. At the same time, the residual current and zero-sequence current signals are synchronously collected from 7 monitoring points. These signals are converted into leakage change signals after preprocessing and further drawn into images. Subsequently, these image data are converted into array form and input into the CNN model for dimensionality reduction. After optimization by the SCSSA algorithm, the data is input into the Bi-LSTM model for training. In this process, this embodiment divides the data set into a training set and a test set in a ratio of 4:3. In order to intuitively demonstrate the advantages of the method proposed in this embodiment, this embodiment is Figure 4 and Figure 5 The confusion matrices of the true value and the predicted value of the SCSSA-CNN-Bi-LSTM model that comprehensively utilizes the residual current signal and the zero-sequence current signal and the CNN-Bi-LSTM model that only uses the residual current signal are shown respectively. By comparison, it can be seen that the CNN-Bi-LSTM model that only relies on the residual current signal has obvious difficulties in distinguishing between leakage faults and load switching, while the SCSSA-CNN-Bi-LSTM model provided in this embodiment can accurately distinguish between the two situations and effectively achieve the purpose of leakage detection.

[0155] In addition, this embodiment also compares the prediction accuracy of the SCSSA-CNN-BI-LSTM model in Table 1 with the accuracy of the traditional single criterion, and comprehensively evaluates the prediction results based on the SCSSA-CNN-Bi-LSTM model, the traditional CNN-Bi-LSTM model and the traditional single criterion. The evaluation results show that the SCSSA-CNN-Bi-LSTM model proposed in this embodiment performs well in terms of prediction accuracy, which fully demonstrates that the architecture based on the SCSSA-CNN-Bi-LSTM model can effectively extract feature information, so that the model can learn the feature quantity that effectively distinguishes leakage faults and load switching, thereby significantly improving the accuracy of the prediction model. This result further verifies the effectiveness of the method of the present invention. Table 1 is shown below:

[0156] Table 1

[0157]

[0158]

[0159] The embodiment of the present invention provides a method for detecting leakage faults in a low-voltage distribution network. The method obtains a residual current and a zero-sequence current data group according to the three-phase current data and the neutral current data of the low-voltage distribution network; performs opening and closing operations on the residual current and the zero-sequence current data group using a mathematical morphology method to obtain local-global characteristic data of the current signal; converts the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal, and calculates a leakage current change signal; constructs a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on a sparrow optimization algorithm that integrates a sine-cosine search strategy and a Cauchy mutation mechanism; and performs fault identification based on the leakage current change signal using the low-voltage distribution network leakage fault detection model to obtain a leakage fault identification result. Compared with the prior art, the method comprehensively utilizes the residual current signal and the zero-sequence current signal, and realizes high-precision identification of leakage faults in the low-voltage distribution network through the sparrow optimization algorithm that integrates a sine-cosine search strategy and a Cauchy mutation mechanism, effectively improving the accuracy and efficiency of leakage detection, thereby ensuring the safe and stable operation of the low-voltage distribution network.

[0160] It should be noted that the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0161] In one embodiment, Figure 6 As shown, an embodiment of the present invention provides a low-voltage distribution network leakage fault detection system, the system comprising:

[0162] The data acquisition module 101 is used to obtain a residual current and a zero-sequence current data group according to the three-phase current data and the neutral current data of the low-voltage distribution network;

[0163] The data processing module 102 is used to perform opening and closing operations on the residual current and zero-sequence current data group using a mathematical morphology method to obtain local-global characteristic data of the current signal;

[0164] The mutation analysis module 103 is used to convert the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal, and calculate the leakage current change signal;

[0165] A model building module 104 is used to build a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on a sparrow optimization algorithm integrating a sine-cosine search strategy and a Cauchy mutation mechanism;

[0166] The fault identification module 105 is used to perform fault identification based on the leakage current change signal and utilize the low-voltage distribution network leakage fault detection model to obtain a leakage fault identification result.

[0167] For the specific definition of a low-voltage distribution network leakage fault detection system, please refer to the above-mentioned definition of a low-voltage distribution network leakage fault detection method, which will not be repeated here. Those of ordinary skill in the art will appreciate that the various modules and steps described in conjunction with the embodiments disclosed in this application can be implemented in hardware, software, or a combination of both. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0168] The embodiment of the present invention provides a low-voltage distribution network leakage fault detection system. The data acquisition module of the system obtains a residual current and a zero-sequence current data group according to three-phase current data and neutral current data of the low-voltage distribution network; the data processing module uses a mathematical morphology method to perform opening and closing operations on the residual current and zero-sequence current data group to obtain local-global characteristic data of the current signal; the mutation analysis module converts the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal, and calculates a leakage current change signal; the model construction module constructs a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on a sparrow optimization algorithm that integrates a sine-cosine search strategy and a Cauchy mutation mechanism; the fault identification module uses the low-voltage distribution network leakage fault detection model to perform fault identification based on the leakage current change signal to obtain a leakage fault identification result. Compared with the existing technology, this system comprehensively utilizes residual current signals and zero-sequence current signals, and realizes high-precision identification of leakage faults in low-voltage distribution networks by integrating sine-cosine search strategies and Cauchy mutation mechanisms. It effectively improves the accuracy and efficiency of leakage detection, thereby ensuring the safe and stable operation of low-voltage distribution networks.

[0169] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A method for detecting leakage faults in a low voltage distribution network, characterized in that: The following steps are involved: According to the three-phase current data and the neutral line current data of the low-voltage distribution network, a residual current and a zero-sequence current data group are obtained; Using mathematical morphology method to perform opening and closing operations on the residual current and zero-sequence current data group to obtain local-global characteristic data of current signal; The local-global characteristic data of the current signal is converted into a residual-zero-sequence current mutation ratio signal, and a leakage current change signal is calculated; Based on the sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism, a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM is constructed; Based on the leakage current change signal, the low-voltage distribution network leakage fault detection model is used to perform fault identification to obtain a leakage fault identification result.

2. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 1, characterized in that: The step of obtaining the residual current and zero-sequence current data group according to the three-phase current data and the neutral current data of the low-voltage distribution network comprises: Synchronously collecting three-phase current data and neutral current data of the low-voltage distribution network, and obtaining residual current data according to the vector sum of the three-phase current data and the neutral current data; Calculating the root mean square value of the residual current data using a sliding window to obtain a residual current effective value sequence; Analyzing the three-phase current data using a symmetrical component method to obtain a zero-sequence current phasor; Reconstructing a zero-sequence current time-domain waveform according to the amplitude information and phase information of the zero-sequence current phasor, and performing a wavelet transform on the zero-sequence current time-domain waveform to extract transient characteristics; The residual current effective value sequence and the transient characteristics are combined to form a residual current and zero-sequence current data set.

3. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 2, characterized in that: The step of analyzing the three-phase current data by using the symmetrical component method to obtain the zero-sequence current phasor comprises: Performing Fourier transform on the three-phase current data respectively to extract the fundamental wave component amplitude information and fundamental wave component phase information of each phase; Generate three-phase current phasors according to the fundamental component amplitude information and the fundamental component phase information; The three-phase current phasor is vector-calculated by using the symmetrical component method to obtain the zero-sequence current phasor.

4. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 1, characterized in that: The step of performing opening and closing operations on the residual current and zero-sequence current data group using a mathematical morphology method to obtain local-global characteristic data of the current signal comprises: Determining linear structural elements at different scales according to the sampling rate and noise frequency of the residual current and zero-sequence current data group; Performing corrosion expansion boundary processing on both ends of the residual current and zero-sequence current data group by using a mirror filling method to obtain a residual current and zero-sequence current filling data group; Performing an erosion operation on the residual current and zero-sequence current filling data group according to the linear structural element to obtain an erosion result, and performing an expansion operation on the erosion result to obtain a local smoothing feature; Performing an expansion operation on the residual current and zero-sequence current filling data group according to the linear structural element to obtain an expansion result, and performing an erosion operation on the expansion result to obtain a global fluctuation feature; Determining the fusion weight of the local smooth feature and the global fluctuation feature according to the signal-to-noise ratio of the local smooth feature and the signal-to-noise ratio of the global fluctuation feature at different linear structure element scales; The local smoothness feature and the global fluctuation feature are weightedly fused based on the fusion weight to obtain local-global feature data of the current signal.

5. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 1, characterized in that: The step of converting the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal and calculating the leakage current change signal comprises: Extracting residual current local-global characteristic data and zero-sequence current local-global characteristic data in a normal state from the current signal local-global characteristic data; Eliminate the first power frequency cycle data in the residual current local-global characteristic data and the zero-sequence current local-global characteristic data in the normal state to obtain a residual-zero-sequence current normal data group; The normal residual-zero-sequence current data group is processed by using a sliding time window to calculate the residual current average value and the zero-sequence current average value; Obtaining a mutation ratio constant according to a ratio between the residual current average value and the zero-sequence current average value; Calculating the residual current characteristic change and the zero-sequence current characteristic change of each sampling point according to the local-global characteristic data of the current signal; According to the residual current characteristic change and the zero-sequence current characteristic change, the leakage current change value is obtained by weighted calculation through mutation ratio constant.

6. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 1, characterized in that: The steps of constructing a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on the sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism include: The convolutional neural network and the bidirectional long short-term memory network are cascaded to construct the CNN-BiLSTM sequence prediction model; The sparrow optimization algorithm that integrates the sine and cosine search strategy and the Cauchy mutation mechanism is used to globally optimize the multi-dimensional parameter space of the CNN-BiLSTM sequence prediction model to obtain the feature optimization parameters; The feature optimization parameters are mapped to the CNN-BiLSTM sequence prediction model to construct a low-voltage distribution network leakage fault detection model; The historical leakage current change values ​​are converted into a binary image matrix, and the low-voltage distribution network leakage fault detection model is trained using the binary image matrix to obtain a trained low-voltage distribution network leakage fault detection model.

7. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 6, characterized in that: The feature optimization parameters include convolutional neural network structure parameters, model hyperparameters, and the number of hidden layer neurons in the bidirectional long short-term memory network; the model hyperparameters include learning rate and L2 regularization coefficient.

8. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 1, characterized in that: The step of performing fault identification based on the leakage current change signal using the low-voltage distribution network leakage fault detection model to obtain a leakage fault identification result comprises: Converting the leakage current change signal into a leakage current change binary image, and inputting the leakage current change binary image into the low-voltage distribution network leakage fault detection model for fault identification, to obtain a leakage fault identification result; If the leakage fault identification result is that the line is in normal operation or load switching occurs, continue to monitor the leakage current change signal; If the leakage fault identification result is that a leakage fault occurs in the line, a leakage fault alarm mechanism is triggered.

9. A method for detecting leakage faults in a low voltage distribution network as claimed in claim 8, characterized in that: The step of converting the leakage current change signal into a leakage current change binary image comprises: Converting the leakage current change signal into an image format to obtain an original leakage current change image; Performing a cropping process on the original leakage current variation image to extract a cropped image including a leakage current variation value curve; The cropped image is converted into a grayscale image, and the grayscale image is globally binarized to obtain a leakage current change binary image.

10. A low voltage distribution network leakage fault detection system, characterized in that: The system comprises: A data acquisition module is used to obtain a residual current and a zero-sequence current data group according to the three-phase current data and the neutral current data of the low-voltage distribution network; A data processing module, used for performing opening and closing operations on the residual current and zero-sequence current data groups using a mathematical morphology method to obtain local-global characteristic data of the current signal; A mutation analysis module, used for converting the local-global characteristic data of the current signal into a residual-zero-sequence current mutation ratio signal, and calculating a leakage current change signal; Model building module, which is used to build a low-voltage distribution network leakage fault detection model based on SCSSA-CNN-BiLSTM based on the sparrow optimization algorithm integrating the sine-cosine search strategy and the Cauchy mutation mechanism; The fault identification module is used to perform fault identification based on the leakage current change signal and utilize the low-voltage distribution network leakage fault detection model to obtain a leakage fault identification result.

Images