Residual current detection method, equipment and product

Abstract

The embodiment of the invention provides a residual current detection method, equipment and a product. The method comprises the following steps: decomposing a residual current signal to obtain modal components at different time points in the residual current signal; according to the sparse characteristic of the modal component, performing reconstruction processing on the modal component to obtain a reconstruction signal; and classifying the reconstructed signals according to the sparse features of the reconstructed signals, and determining signal categories of the signals in the reconstructed signals and fault types corresponding to the signal categories. The method is used for achieving the effect of improving the detection accuracy.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, device, and product for detecting residual current. Background Technology

[0002] In the field of low-voltage power distribution networks, residual current, as the current flowing through grounded or abnormal circuits, can easily cause serious safety accidents such as electric shock and electrical fires if its abnormal conditions are not met. Therefore, residual current detection and protection devices have become a key technology for ensuring the safe operation of the network and play an indispensable role in the entire industry.

[0003] Currently, existing methods for detecting residual current mainly involve monitoring the effective value of the residual current in real time and comparing it with a preset threshold. Once the threshold is exceeded, the device triggers a trip, quickly disconnecting the circuit to ensure personal and equipment safety.

[0004] However, with the increasing diversification of electrical equipment in current power distribution networks and the widespread integration of distributed renewable energy sources, the characteristics of residual current signals have become increasingly complex, with significantly enhanced noise interference and dynamic characteristics. This leads to frequent false alarms or missed alarms in traditional detection methods based on fixed threshold judgments. Summary of the Invention

[0005] This application provides a method, device, and product for detecting residual current, in order to improve the accuracy of detection.

[0006] In a first aspect, embodiments of this application provide a method for detecting residual current, comprising:

[0007] The residual current signal is decomposed to obtain the modal components at different time points in the residual current signal.

[0008] Based on the sparsity characteristics of the modal components, the modal components are reconstructed to obtain the reconstructed signal;

[0009] Based on the sparse characteristics of the reconstructed signals, the reconstructed signals are classified to determine the signal categories and the corresponding fault types.

[0010] In one possible implementation, the residual current signal is decomposed to obtain modal components at different time points in the residual current signal, including:

[0011] The residual current signal is expanded to obtain the expanded signal;

[0012] The extended signal is decomposed to obtain modal candidate functions;

[0013] Based on the preset sparse optimization conditions and residual update conditions, the modal components in the modal candidate function are determined.

[0014] In one possible implementation, the modal components in the modal candidate function are determined based on preset sparse optimization conditions and residual update conditions, including:

[0015] Based on the preset sparse optimization conditions, the modal candidate functions are subjected to sparse optimization processing to obtain the intrinsic modal functions;

[0016] According to the preset residual update conditions, the residual signal is processed to obtain the reconstruction error. The residual signal is determined based on the extended signal and the mode candidate function.

[0017] If the reconstruction error meets the preset error conditions, the modal components are determined based on the inherent mode function.

[0018] In one possible implementation, the modal components are reconstructed based on their sparsity characteristics to obtain a reconstructed signal, including:

[0019] Based on the modal component matrix of the modal components, construct an overcomplete dictionary;

[0020] The overcomplete dictionary is solved by sparse optimization to determine the sparsity coefficients of the overcomplete dictionary;

[0021] Based on the sparsity coefficients of the overcomplete dictionary, the modal components are filtered to determine the effective modal components.

[0022] The reconstructed signal is obtained based on the effective modal components and sparse coefficients.

[0023] In one possible implementation, the reconstructed signal is classified based on its sparsity characteristics to determine the signal category and the corresponding fault type, including:

[0024] Based on the feature extraction network, sparse features are extracted from the reconstructed signal;

[0025] Based on the feature distance between sparse features, the positive and negative samples in the reconstructed signal are determined;

[0026] Based on the positive and negative samples in the reconstructed signal, the feature extraction network is optimized to obtain the target feature network;

[0027] Based on the target feature network, sparse contrast features are extracted from the reconstructed signal;

[0028] The sparse contrast features in the reconstructed signal are input into the recognition model to determine the signal category and the fault type corresponding to the signal category in the reconstructed signal. The recognition model is trained based on the sparse sample features and the signal category labels corresponding to the sparse sample features.

[0029] In one possible implementation, the feature extraction network is optimized based on the positive and negative samples in the reconstructed signal to obtain the target feature network, including:

[0030] The feature distance between positive and negative samples is adjusted to obtain adjusted positive and negative samples.

[0031] The loss function is determined based on the adjustment of positive and negative samples;

[0032] Based on the loss function, the feature extraction network is optimized to determine the target feature network; the loss function satisfies:

[0033]

[0034] Among them, z i ,z j Characterized as positive sample pairs, z i ′,z k The negative sample pairs are represented by τ, which is a temperature parameter used to adjust the sparsity of the distribution. sim(z) i ,z j ) is represented by the cosine similarity of positive sample pairs, sim(z) i ′,z k ) represents the cosine similarity of negative sample pairs, and N represents the number of negative samples.

[0035] In one possible implementation, sparse contrast features in the reconstructed signal are input into a recognition model to determine the signal category of the signal in the reconstructed signal and the fault type corresponding to the signal category, including:

[0036] The sparse contrast features in the reconstructed signal are input into the classification network in the recognition model to determine the signal category of the signal in the reconstructed signal.

[0037] Based on the signal category in the reconstructed signal and the Softmax network in the recognition model, the fault type corresponding to the signal category is determined.

[0038] In one possible implementation, sparse contrastive features from the reconstructed signal are input into a classification network in a recognition model to determine the signal category of the signal in the reconstructed signal, including:

[0039] The sparse contrast features in the reconstructed signal are input into the depthwise separable convolutional units in the classification network to determine the spatial feature information.

[0040] Spatial feature information is input into the inverse residual module of the classification network to obtain the feature classification vector;

[0041] Based on the feature classification vector and the fully connected layer in the classification network, the signal category of the reconstructed signal is determined.

[0042] Secondly, embodiments of this application provide a residual current detection device, comprising:

[0043] The decomposition module is used to decompose the residual current signal to obtain the modal components at different time points in the residual current signal.

[0044] The module is used to reconstruct the modal components based on their sparsity characteristics to obtain the reconstructed signal.

[0045] The determination module is used to classify the reconstructed signal based on its sparsity characteristics, and to determine the signal category and the fault type corresponding to the signal category in the reconstructed signal.

[0046] Thirdly, embodiments of this application provide a residual current detection device, including: a memory and a processor;

[0047] The memory stores the instructions that the computer executes;

[0048] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0049] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0050] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0051] This application provides a residual current detection method, device, and product. By decomposing the residual current signal, modal components at different time points are obtained to separate noise and fault characteristics, accurately extracting effective signals and avoiding noise interference. Furthermore, based on the sparsity characteristics of the modal components, noise is further filtered during reconstruction processing to highlight key fault characteristics, making the reconstructed signal purer and more accurately reflecting the residual current. Thus, based on the sparsity characteristics of the reconstructed signal, it is classified to ensure accurate determination of the signal category and the corresponding fault type. This effectively avoids false alarms and missed alarms caused by dynamic signal changes and noise interference, significantly improving detection accuracy. Attached Figure Description

[0052] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0053] Figure 1 A schematic diagram illustrating a scenario for the residual current detection method provided in this application;

[0054] Figure 2 Flowchart of the residual current detection method provided in this application Figure 1 ;

[0055] Figure 3 This is a schematic diagram of the process for processing the original residual current in a new model, provided in an embodiment of this application.

[0056] Figure 4 Flowchart of the residual current detection method provided in this application Figure 2 ;

[0057] Figure 5 A schematic diagram of the structure of the recognition model provided in this application;

[0058] Figure 6 A schematic diagram of the residual current detection device provided in this application;

[0059] Figure 7 This is a schematic diagram of the residual current detection device provided in this application.

[0060] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0061] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0062] First, let me explain the terms used in this application:

[0063] Intrinsic Mode Functions (IMFs) are components with specific properties obtained from analyses such as Empirical Mode Decomposition (EMD) of complex signals. IMFs can be defined as functions that satisfy the following two conditions: First, the number of extreme points (maximums and minimums) must be equal to or at most differ by one from the number of zero-crossings throughout the entire data record; second, at any given time, the average of the upper envelope determined by local maxima and the lower envelope determined by local minima is zero.

[0064] Empirical Mode Decomposition (EMD) is a time-frequency analysis method specifically designed for analyzing nonlinear and non-stationary signals. It does not require pre-defined basis functions and relies entirely on the local characteristics of the signal itself for decomposition.

[0065] An overcomplete dictionary is a concept in sparse representation theory, referring to a dictionary whose number of atoms (i.e., column vectors) exceeds the dimension of the signal space.

[0066] In current technologies, the structure of power distribution networks has undergone significant changes. Electrical equipment is becoming increasingly diversified, with widespread use of nonlinear loads and motor loads, and a large influx of distributed renewable energy sources. This has made the characteristics of residual current signals extremely complex, significantly increasing noise interference and dynamic characteristics. Under these circumstances, traditional residual current detection methods based on fixed threshold judgments exhibit numerous drawbacks, easily leading to false alarms or missed alarms, severely impacting the reliability and safety of protection devices. Furthermore, residual current signals encompass various fault types, such as electric shock from living organisms and non-living organisms, while existing methods lack sufficient ability to identify these fault types, failing to meet the high-precision protection requirements of modern power distribution networks.

[0067] Faced with the challenge of residual current detection and classification, recent research has focused on several directions. On the one hand, it has improved signal processing methods, such as wavelet transform (WT), to extract key features of residual current signals; on the other hand, it has used machine learning algorithms, such as support vector machines (SVM) and dynamic fuzzy neural networks (DFNN), to classify electric shock types, and has also combined deep learning techniques, such as convolutional neural networks (CNN), to improve the accuracy of signal classification.

[0068] However, existing technologies have revealed many obvious shortcomings in practical applications. For example, the signal decomposition accuracy is poor. Traditional empirical mode decomposition is plagued by endpoint effects and mode mixing, resulting in inaccurate decomposition results and reducing the quality of signal feature extraction. Noise interference has a significant impact. The large amount of noise and redundant components in residual current signals make it difficult for existing signal processing methods to retain key information during noise reduction, thus dragging down classification performance. Feature extraction and representation capabilities are weak. Current methods rely heavily on manually designed feature extraction methods, which cannot fully explore the potential features of signals in complex scenarios, thus limiting classification capabilities. The model complexity is too high. Existing deep learning models generally have high computational complexity, making them difficult to implement in embedded devices or real-time detection scenarios, and they lack lightweight design.

[0069] This application provides a residual current detection method that decomposes the residual current signal to obtain modal components at different time points, thereby separating noise and fault features, accurately extracting effective signals, and avoiding noise interference. Furthermore, based on the sparsity characteristics of the modal components, the method further filters noise during reconstruction, highlighting key fault features and making the reconstructed signal purer, more accurately reflecting the residual current. Based on the sparsity characteristics of the reconstructed signal, it classifies the signal to ensure accurate determination of the signal category and the corresponding fault type. This effectively avoids false alarms and missed alarms caused by dynamic signal changes and noise interference, significantly improving detection accuracy.

[0070] Figure 1 A schematic diagram illustrating a scenario for the residual current detection method provided in this application, such as... Figure 1 As shown, the specific application scenario of this application is a residual current detection system. The execution entity of this system can be a server, which can be a mobile phone, tablet, computer, or other device. In this embodiment, the execution entity is not limited, as long as it can perform decomposition processing of the residual current signal to obtain the modal components at different time points in the residual current signal; reconstruct the modal components according to their sparsity characteristics to obtain the reconstructed signal; and classify the reconstructed signal according to its sparsity characteristics to determine the signal category and the fault type corresponding to the signal category.

[0071] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0072] Figure 2 Flowchart of the residual current detection method provided in this application Figure 1 ,like Figure 2 As shown, the method includes:

[0073] S201. Decompose the residual current signal to obtain the modal components at different time points in the residual current signal.

[0074] The residual current signal refers to the electrical signal generated during the operation of a low-voltage power distribution system when current flows through a path other than the normal path (such as grounding or through a non-normal conductor like a human body) due to reasons such as line insulation damage, equipment leakage, or grounding faults. The residual current signal can include operating status information of electrical equipment. In this embodiment, the residual current signal is an electrical signal that needs to be detected.

[0075] Decomposing a residual current signal can refer to the process of decomposing the signal into a set of intrinsic mode functions. In some embodiments, when analyzing a residual current signal, techniques such as empirical mode decomposition can be used to decompose the residual current signal to obtain the mode components at different time points in the residual current signal.

[0076] Modal components refer to the components of different frequencies or time scales separated from the original complex signal through specific signal decomposition methods, such as empirical mode decomposition (EMD) and wavelet decomposition. Taking empirical mode decomposition as an example, the obtained modal components are intrinsic mode functions (IMFs) or intrinsic modal components.

[0077] In the field of residual current detection described in this application, when analyzing residual current signals, intrinsic mode functions (IMFs) obtained through techniques such as empirical mode decomposition (EMD) can decompose complex residual current signals into multiple intrinsic mode components (i.e., mode components) with different characteristic scales based on their inherent time-scale characteristics. IMFs can reflect the variations of residual current signals across different frequency bands and time scales, facilitating a more detailed analysis of the characteristics of the residual current signal. This provides crucial information for accurately detecting residual current anomalies and identifying fault types, overcoming the shortcomings of traditional detection methods when dealing with complex signals.

[0078] In this embodiment of the application, the residual current signal is decomposed to obtain the modal components at different time points in the residual current signal, including:

[0079] The residual current signal is expanded to obtain the expanded signal;

[0080] The extended signal is decomposed to obtain modal candidate functions;

[0081] Based on the preset sparse optimization conditions and residual update conditions, the modal components in the modal candidate function are determined.

[0082] The extended signal refers to the signal obtained by extending the two ends of the residual current signal. This is done to provide more sufficient boundary information for subsequent signal processing. In signal processing, operations such as initial decomposition and sparse optimization are easily affected by the data at both ends of the signal. Insufficient boundary information may lead to inaccurate processing results. Extending the two ends of the signal through methods such as mirror extension, periodic extension, or zero extension can effectively reduce boundary effects, making the signal processing algorithm more accurate when processing the data at both ends, and ensuring that the generated modal candidate functions and intrinsic modal functions better match the true characteristics of the signal.

[0083] The modal candidate function h(k) can refer to the intermediate result obtained after preprocessing and initial decomposition of the original residual current signal x(t). The modal candidate function can be the product of preliminary feature extraction of the residual current signal, which includes the potential feature information of the residual current signal at different scales and frequencies.

[0084] Sparse optimization conditions refer to imposing constraints on the modal candidate function h(k) to generate sparse intrinsic mode functions IMFk(t). Sparse optimization conditions can remove redundant components in the signal, making subsequent analysis more efficient, highlighting key features, focusing signal energy on a few important parts, facilitating the identification of essential differences in signals under different states, and enhancing anti-interference capabilities while reducing the impact of noise.

[0085] The residual update condition is a key factor in ensuring the continuous progress of the decomposition process and its approximation of the true signal characteristics. After completing a round of sparse optimization of the mode candidate function h(k) to obtain the intrinsic mode function IMFk(t), the residual between the currently decomposed signal and the residual current signal x(t) is calculated. The residual update condition is used to adjust subsequent decomposition processes based on the residual. For example, if the residual exceeds a preset threshold, it indicates that the current decomposition has not fully extracted the signal characteristics, and the previous decomposition steps need to be repeated on the residual signal to update the mode candidate function, etc., until the residual meets the set requirements, thereby ensuring that the decomposed intrinsic mode function can accurately characterize the intrinsic properties of the original signal as much as possible.

[0086] Modal components can serve as the basic units for constructing modal candidate functions. These components characterize specific vibration modes or feature components of the residual current signal at different frequencies and scales. Each modal component possesses unique attributes such as frequency, amplitude, and phase, reflecting the intrinsic characteristics of the residual current signal in a specific aspect. The superposition and combination of modal components can form modal candidate functions, which carry rich information about the residual current signal.

[0087] In this embodiment of the application, determining the modal components in the modal candidate function based on preset sparse optimization conditions and residual update conditions includes:

[0088] Based on the preset sparse optimization conditions, the modal candidate functions are subjected to sparse optimization processing to obtain the intrinsic modal functions;

[0089] According to the preset residual update conditions, the residual signal is processed to obtain the reconstruction error. The residual signal is determined based on the extended signal and the mode candidate function.

[0090] If the reconstruction error meets the preset error conditions, the modal components are determined based on the inherent mode function.

[0091] Among them, the intrinsic mode function (IMF) can refer to the function obtained by processing the candidate mode function according to the sparse optimization condition. The intrinsic mode function can more accurately reflect the intrinsic vibration mode and characteristics of the residual current signal and is an important component for the effective analysis of the residual current signal.

[0092] The residual signal can refer to the signal determined based on the expanded residual current signal and the mode candidate function. The residual signal can represent the part of the residual current signal that is not fully represented by the mode candidate function at the current decomposition stage.

[0093] Reconstruction error refers to the result obtained after processing the residual signal through residual update. It is used to measure the accuracy of the reconstructed signal. Reconstruction error can reflect the difference between the current decomposition result and the residual current signal.

[0094] Error conditions refer to pre-defined criteria used to determine whether the reconstruction error meets the requirements. When the reconstruction error meets this condition, the current decomposition result is considered sufficiently accurate, and the modal components can be determined based on the intrinsic mode functions. For example, if the error condition is set to a reconstruction error of less than 0.05, and the reconstruction error obtained when processing the residual current signal is 0.03, which is less than the pre-defined 0.05, then the decomposition of the residual current signal has achieved the desired accuracy. In this case, the modal components can be accurately determined based on the obtained intrinsic mode functions.

[0095] In the embodiments of this application, sparse optimization processing and residual update processing can refer to: Sparse Constrained Empirical Mode Decomposition (SC-EMD) introduces sparsity constraints on the basis of EMD, and improves the decomposition accuracy by optimizing the sparse representation of intrinsic mode functions to reduce redundant components and noise interference in the signal.

[0096] The optimization objective of Empirical Mode Decomposition (EMD) is to improve the decomposition process of IMFs through sparsity regularization constraints. Let the original residual current signal be x(t), which can be decomposed into several IMFs as follows:

[0097]

[0098] Among them, IMF i r(t) is the i-th intrinsic mode function, and r(t) is the residual of the decomposition.

[0099] To improve the sparsity and denoising effect of the decomposition, sparse constraint optimization is added, and the objective function can be expressed as:

[0100]

[0101] Among them, ||IMF i ||1 represents the sparsity constraint term, indicating the sparsity of IMFs in the time domain. Minimizing the L1 norm reduces the non-zero components of the IMFs, thus suppressing noise and redundant information. The content after st represents the error constraint, limiting the difference between the decomposition result and the original signal to ensure the accuracy of the decomposition result. ∈ represents the allowed reconstruction error.

[0102] Figure 3 This is a schematic diagram of the process for processing the original residual current in an embodiment of this application, as shown below. Figure 3 As shown, the process includes: Input: the original residual current signal x(t); Signal preprocessing: expanding both ends of the signal; Initial decomposition: extracting the modal candidate function h(k); Sparse optimization: performing sparsity constraint optimization on h(k) to generate the IMF. k (t); Update residual: r k (t)=x extendde (t)-IMF k (t) Whether the reconstruction error is satisfied; if not, perform the initial decomposition again; if yes, output: remove the noise mode and output the valid IMFs.

[0103] S202. Based on the sparsity characteristics of the modal components, the modal components are reconstructed to obtain the reconstructed signal.

[0104] The sparsity characteristic refers to a signal where, under a certain representation, only a few coefficients are non-zero, while most coefficients are zero or close to zero. In sparse representation techniques, a signal can be represented as a linear combination of sparse coefficients using an overcomplete dictionary. In the embodiments of this application, the modal components obtained from the decomposition of the residual current signal often contain redundant information such as noise. However, based on the sparsity characteristic of the signal, it is known that the truly critical signal components have non-zero or relatively large coefficients under a specific representation, while the coefficients of noise and other interference components tend to be zero or relatively small. Utilizing this characteristic, when reconstructing the modal components, key signal components can be selected based on the magnitude of the coefficients, while noise and other interference can be suppressed. The reconstructed signal thus removes redundancy, highlights essential features, improves signal quality, and provides a more reliable foundation for subsequent accurate feature characterization and analysis, thereby more effectively extracting useful information from complex signals.

[0105] Reconstruction processing refers to the process of using filtered effective information to recombine and restore a new signal. In the embodiments of this application, based on the sparsity characteristics of modal components, specific algorithms and techniques are used to filter the modal components, retain key signal components, remove noise interference, and then linearly combine these filtered effective modal components to obtain a higher quality reconstructed signal.

[0106] A reconstructed signal is a new signal obtained after reconstruction processing. It is the result of decomposing, filtering, and recombining the original signal. Compared with the original signal, the reconstructed signal removes noise and redundant information while retaining key signal components, thus resulting in higher quality.

[0107] In this embodiment of the application, based on the sparsity characteristics of the modal components, the modal components are reconstructed to obtain a reconstructed signal, including:

[0108] Based on the modal component matrix of the modal components, construct an overcomplete dictionary;

[0109] The overcomplete dictionary is solved by sparse optimization to determine the sparsity coefficients of the overcomplete dictionary;

[0110] Based on the sparsity coefficients of the overcomplete dictionary, the modal components are filtered to determine the effective modal components.

[0111] The reconstructed signal is obtained based on the effective modal components and sparse coefficients.

[0112] The modal component matrix (IMF) refers to a matrix composed of modal components. These modal components are arranged in a specific way to facilitate mathematical operations and processing. The IMF matrix contains information about each modal component after decomposition of the original signal. Each row or column can represent a modal component, and the elements in the matrix represent the values ​​of that modal component at different times or different feature dimensions. For example, the decomposed IMF matrix might be IMF = [IMF1, IMF2, ..., IMF...]. N ].

[0113] Sparse coefficients refer to the coefficients of the basis vectors when a signal is represented using an overcomplete dictionary. The magnitude of the sparse coefficients reflects the contribution of each basis vector to the signal representation, and is used to determine which basis vectors play a key role in the signal representation.

[0114] Effective modal components refer to those modal components selected from the original modal components that are of significant importance and value to the characteristics and subsequent analysis of the signal. These components contain key information of the original signal, represent the main characteristics and variation patterns of the signal, and remove any noise or redundant modal components.

[0115] In this embodiment, an orthogonal matching pursuit algorithm is used to learn an overcomplete dictionary D from the modal component matrix; wherein the overcomplete dictionary D satisfies:

[0116]

[0117] Where, α∈R k It is a sparse coefficient vector.

[0118] After determining the complete dictionary D, the sparse coefficient vector α * satisfy:

[0119]

[0120] Determine the sparse coefficient vector α * After determining the amplitude, the components in the IMFs are filtered. The filtering process is as follows:

[0121]

[0122] Among them, E i Let α be the effective component of the i-th signal in the sparse coefficient vector, used to measure the correlation between the signal and the corresponding basis vector in the dictionary. i Let be the sparse coefficient vector of the i-th signal, where i and j typically represent the indices of the signal and the dictionary basis vectors. That is, i is the index of the signal, and j is the index of the basis vector in the dictionary, which is used to represent the relationship between the signal and the basis vector involved in calculating the sparse coefficients.

[0123] Where, if E i If E is less than a set threshold δ, the corresponding mode is discarded. In this embodiment, the set threshold can be a preset constant used to control the precision of the sparse representation. For example, the set threshold can be 0.3. i If the value is less than 0.3, it will be removed.

[0124] The signal is reconstructed using the filtered sparse coefficients, where the reconstructed signal satisfies:

[0125]

[0126] S203. Based on the sparse characteristics of the reconstructed signal, classify the reconstructed signal to determine the signal category of the signal in the reconstructed signal and the fault type corresponding to the signal category.

[0127] Extracting sparse features from the reconstructed signal involves using sparse representation algorithms to decompose the signal into a linear combination of non-zero coefficients and dictionary basis vectors to obtain features that reflect the signal's inherent structure. Next, a classification model, such as a support vector machine or neural network, is constructed and trained using a large amount of sample data with known signal categories and corresponding fault types. This allows the model to learn the mapping relationship between different sparse features and signal categories / fault types. Then, the sparse features of the reconstructed signal to be classified are input into the trained classification model. The model classifies the signal based on the learned patterns and outputs its corresponding signal category. Finally, based on a pre-established correspondence table between signal categories and fault types, the specific fault type corresponding to that signal category is determined.

[0128] The residual current detection method provided in this application successfully overcomes the endpoint effect and mode aliasing problems of traditional empirical mode decomposition by introducing the sparse-constrained empirical mode decomposition (SC-EMD) method. It can accurately extract intrinsic mode functions (IMFs) from the residual current signal. The sparse regularization constraint also reduces redundant information in the decomposition process, providing high-quality signal components for subsequent feature extraction. In terms of enhancing signal denoising and reconstruction capabilities, the sparse representation technique is used to screen and reconstruct the IMFs obtained by SC-EMD decomposition. The sparse optimization model is used to remove noise interference and invalid components, retaining only key signals, thereby generating a high-quality reconstructed signal, which greatly improves signal quality and feature expression capabilities, and provides more accurate input for classification models.

[0129] Figure 4 Flowchart of the residual current detection method provided in this application Figure 2 ,like Figure 4 As shown, in this embodiment... Figure 2Based on the embodiments, the process of classifying reconstructed signals according to their sparsity characteristics, determining the signal categories and corresponding fault types within the reconstructed signals, is described in detail. This method includes:

[0130] S301. Extract sparse features from the reconstructed signal using the feature extraction network;

[0131] S302. Determine the positive and negative samples in the reconstructed signal based on the feature distance between sparse features;

[0132] S303. Based on the positive and negative samples in the reconstructed signal, optimize the feature extraction network to obtain the target feature network;

[0133] S304. Based on the target feature network, extract the sparse contrast features in the reconstructed signal.

[0134] Wherein, the sparse features satisfy the formula:

[0135]

[0136] in, To reconstruct the signal, f θ This is a feature extraction network, where parameter θ is the model parameter in the feature extraction network.

[0137] When determining positive and negative samples in the reconstructed signal, a suitable feature distance metric such as Euclidean distance or cosine distance is first selected. Then, the feature distance between each sparse feature in the reconstructed signal is calculated. Based on a pre-set threshold, the reconstructed signal with a feature distance less than the threshold is determined as a positive sample, and the reconstructed signal with a feature distance greater than the threshold is determined as a negative sample.

[0138] After determining the positive and negative samples, the feature extraction network can be trained through contrastive learning. That is, by narrowing the feature distance between positive sample pairs (signals of the same class) and widening the feature distance between negative sample pairs (signals of different classes), the feature extraction network is trained based on the positive and negative samples to obtain the target feature network. After obtaining the target feature network, sparse contrastive features in the reconstructed signal can be extracted more accurately.

[0139] In other words, sparse contrastive features refer to those obtained after processing the reconstructed signal using a target feature network. Because the target feature network is optimized using both positive and negative samples, it can extract features from the reconstructed signal more accurately. The extracted sparse contrastive features are a set of features that highlight the differences between the reconstructed signal and the previously extracted ordinary sparse features. It not only preserves the characteristic of the reconstructed signal having an uneven coefficient distribution in a specific transform domain, with non-zero coefficients corresponding to key components, but also, through comparison, highlights the differences between different reconstructed signals at the sparse feature level.

[0140] In this embodiment of the application, the feature extraction network is optimized based on the positive and negative samples in the reconstructed signal to obtain the target feature network, including:

[0141] The feature distance between positive and negative samples is adjusted to obtain adjusted positive and negative samples.

[0142] The loss function is determined based on the adjustment of positive and negative samples;

[0143] Based on the loss function, the feature extraction network is optimized to determine the target feature network; the loss function satisfies:

[0144]

[0145] Among them, z i ,z j Characterized as positive sample pairs, z i ′,z k The negative sample pairs are represented by τ, which is a temperature parameter used to adjust the sparsity of the distribution. sim(z) i ,z j ) is represented by the cosine similarity of positive sample pairs, sim(z) i ′,z k ) represents the cosine similarity of negative sample pairs, and N represents the number of negative samples.

[0146] Here, adjusting positive samples and adjusting negative samples can refer to obtaining samples by narrowing the feature distance between positive sample pairs (signals of the same class) and widening the feature distance between negative sample pairs (signals of different classes). In this embodiment, the feature distance between positive sample pairs and the feature distance between negative sample pairs can be achieved through metric learning methods.

[0147] Therefore, by minimizing the loss function of contrastive learning, the parameters θ of the feature extraction network are optimized, so that the features of signals of the same type are clustered in the sparse feature space, while the features of signals of different types are far apart.

[0148] S305. Input the sparse contrast features in the reconstructed signal into the recognition model to determine the signal category of the signal in the reconstructed signal and the fault type corresponding to the signal category. The recognition model is trained based on the sparse sample features and the signal category label corresponding to the sparse sample features.

[0149] The identification model can include a lightweight deep learning model and a Softmax network. The lightweight deep learning model is used to classify sparse contrast features, and convolutional layers are used to extract high-level features. Combined with the Softmax network, the fault type is output, thereby achieving efficient detection and classification of residual current signals.

[0150] In this embodiment, the sparse contrast features in the reconstructed signal are input into the recognition model to determine the signal category and the fault type corresponding to the signal category in the reconstructed signal, including:

[0151] The sparse contrast features in the reconstructed signal are input into the classification network in the recognition model to determine the signal category of the signal in the reconstructed signal.

[0152] Based on the signal category in the reconstructed signal and the Softmax network in the recognition model, the fault type corresponding to the signal category is determined.

[0153] In this embodiment of the application, the sparse contrastive features in the reconstructed signal are input into the classification network in the recognition model to determine the signal category of the signal in the reconstructed signal, including:

[0154] The sparse contrast features in the reconstructed signal are input into the depthwise separable convolutional units in the classification network to determine the spatial feature information.

[0155] Spatial feature information is input into the inverse residual module of the classification network to obtain the feature classification vector;

[0156] Based on the feature classification vector and the fully connected layer in the classification network, the signal category of the reconstructed signal is determined.

[0157] Among them, the classification network is a lightweight deep learning model, which can be MobileNetV2, to classify sparse contrastive features.

[0158] Figure 5 A schematic diagram of the structure of the recognition model provided in this application is shown below. Figure 5As shown, the MobileNetV2 model employs a depthwise separable convolution module to optimize standard convolution operations. This module decomposes a standard convolution into two steps: first, it extracts the spatial features of each input channel using a single-channel convolution kernel through depthwise convolution; then, it fuses these spatial features using pointwise convolution (1×1 convolution) to enhance the interaction between channels. Furthermore, MobileNetV2 introduces an inverse residual module to improve feature representation capabilities, thereby reducing the model's computational cost and enhancing feature representation.

[0159] Thanks to the application of depthwise separable convolutional modules and inverse residual modules, the MobileNetV2 model can achieve efficient extraction and classification of sparse features with low computational cost, thus adapting to applications in resource-constrained scenarios. Therefore, in the embodiments of this application, this architecture is used for fast and efficient processing tasks in edge computing or power grid embedded devices, ensuring strong performance support with low resource consumption.

[0160] The processing of the recognition model can be divided into two stages: a training stage and a classification stage, specifically including:

[0161] I. Training Phase:

[0162] The specific process for training a feature classification model to identify the signal category and fault type corresponding to sparse contrastive features is as follows:

[0163] (1) Input sparse contrast features: Input the vector z of sparse contrast features into the MobileNetV2 model.

[0164] (2) Construct the MobileNetV2 model and extract features:

[0165] The core of MobileNetV2 is a depthwise separable convolutional structure, consisting of two parts:

[0166] Depthwise convolution: Performs convolution operation independently on each channel to extract the spatial features of each channel, as shown in the following formula:

[0167]

[0168] Among them, X i W represents the input of the i-th channel; i The value represents the corresponding depthwise convolution kernel, and * represents the convolution operation. This is the output after convolution of the i-th channel.

[0169] Pointwise Convolution: This method uses 1×1 convolutions to fuse the results from each channel, outputting a fused feature.

[0170]

[0171] in, It is a pointwise convolution kernel, and Y represents the fused spatial feature map.

[0172] (3) Use inverse residual structures to enhance feature representation:

[0173] The extracted intermediate feature Y is input into the inverse residual module to enhance the network's nonlinear expressive power and output a high-order semantic feature vector. The module structure satisfies:

[0174] Y res =X + F(X);

[0175] Where X is the input feature, F is the nonlinear mapping of the inverse residual module, and Y... res This is the enhanced feature vector.

[0176] (4) Classification mapping and Softmax output:

[0177] Enhanced feature vector Y res The data is compressed to a classification dimension c using a global average pooling layer and a fully connected classification layer, outputting a logits vector:

[0178] z = [z1, z2, ..., z c ]

[0179] Among them, z c This represents the characteristic response value corresponding to the c-th type of signal.

[0180] Next, the logits vector is normalized using the Softmax function to obtain the predicted probability distribution for each signal class:

[0181]

[0182] in, This is the predicted probability for the c-th classification dimension, where C is the number of all categories, and z p This is the output of logits for class p.

[0183] (5) Model optimization training:

[0184] The training process aims to minimize the cross-entropy loss function, which is:

[0185]

[0186] Among them, y c It's a real label. This is the predicted probability for class c.

[0187] The model was trained using the Adam optimizer with an initial learning rate of 0.001, which was decayed to 0.1 every 10 epochs.

[0188] II. Classification Stage:

[0189] After training, the model can be directly used for fault type identification based on sparse contrastive features. The specific process is as follows:

[0190] (1) Input sparse contrast features: Use the sparse contrast features of the signal to be detected as input.

[0191] (2) Model forward inference: The input features pass through the depthwise separable convolutional module, the inverse residual module structure, the fully connected classification layer and the Softmax layer in sequence, and output the probability distribution of each category. The Softmax layer is connected to the output of the fully connected classification layer and is used to convert the output of the fully connected classification layer into a probability distribution.

[0192] (3) Signal category determination and fault identification: Select the index corresponding to the category with the highest probability in the output of the Softmax layer as the identification category of the signal, and combine it with the preset category-fault type mapping table to determine the specific fault type corresponding to the signal (such as biological electric shock, non-biological leakage, normal signal, etc.).

[0193] The residual current detection method provided in this application optimizes the feature distribution in a sparse feature space by combining sparse contrastive learning technology. It uses contrastive learning to narrow the feature distance between signals of the same type while widening the feature distance between signals of different types, thereby extracting highly discriminative key features. This significantly improves the accuracy and robustness of signal classification. Furthermore, it employs the lightweight deep learning model MobileNetV2 to classify the sparse contrastive features. By utilizing depthwise separable convolutions and inverse residual structures, it effectively reduces the number of model parameters and computational complexity. Combined with the Softmax output of the probability distribution of fault types, it achieves efficient detection and real-time classification of residual current signals, perfectly meeting the deployment requirements of embedded devices and resource-constrained environments.

[0194] Figure 6 A schematic diagram of the residual current detection device provided in this application is shown below. Figure 6 As shown, the residual current detection device 40 provided in this embodiment includes:

[0195] The decomposition module 401 is used to decompose the residual current signal to obtain the modal components at different time points in the residual current signal.

[0196] Module 402 is used to reconstruct the modal components based on their sparsity characteristics to obtain a reconstructed signal.

[0197] The determination module 403 is used to classify the reconstructed signal based on the sparse characteristics of the reconstructed signal, and to determine the signal category of the signal in the reconstructed signal, as well as the fault type corresponding to the signal category.

[0198] In one possible implementation, the decomposition module 401 can be specifically used for:

[0199] The residual current signal is expanded to obtain the expanded signal;

[0200] The extended signal is decomposed to obtain modal candidate functions;

[0201] Based on the preset sparse optimization conditions and residual update conditions, the modal components in the modal candidate function are determined.

[0202] In one possible implementation, the decomposition module 401 can be specifically used for:

[0203] Based on the preset sparse optimization conditions, the modal candidate functions are subjected to sparse optimization processing to obtain the intrinsic modal functions;

[0204] According to the preset residual update conditions, the residual signal is processed to obtain the reconstruction error. The residual signal is determined based on the extended signal and the mode candidate function.

[0205] If the reconstruction error meets the preset error conditions, the modal components are determined based on the inherent mode function.

[0206] In one possible implementation, module 402 can be specifically used for:

[0207] Based on the modal component matrix of the modal components, construct an overcomplete dictionary;

[0208] The overcomplete dictionary is solved by sparse optimization to determine the sparsity coefficients of the overcomplete dictionary;

[0209] Based on the sparsity coefficients of the overcomplete dictionary, the modal components are filtered to determine the effective modal components.

[0210] The reconstructed signal is obtained based on the effective modal components and sparse coefficients.

[0211] In one possible implementation, the determining module 403 can be specifically used for:

[0212] Based on the feature extraction network, sparse features are extracted from the reconstructed signal;

[0213] Based on the feature distance between sparse features, the positive and negative samples in the reconstructed signal are determined;

[0214] Based on the positive and negative samples in the reconstructed signal, the feature extraction network is optimized to obtain the target feature network;

[0215] Based on the target feature network, sparse contrast features are extracted from the reconstructed signal;

[0216] The sparse contrast features in the reconstructed signal are input into the recognition model to determine the signal category and the fault type corresponding to the signal category in the reconstructed signal. The recognition model is trained based on the sparse sample features and the signal category labels corresponding to the sparse sample features.

[0217] In one possible implementation, the determining module 403 can be specifically used for:

[0218] The feature distance between positive and negative samples is adjusted to obtain adjusted positive and negative samples.

[0219] The loss function is determined based on the adjustment of positive and negative samples;

[0220] Based on the loss function, the feature extraction network is optimized to determine the target feature network; the loss function satisfies:

[0221]

[0222] Among them, z i ,z j Characterized as positive sample pairs, z i ′,z k The negative sample pairs are represented by τ, which is a temperature parameter used to adjust the sparsity of the distribution. sim(z) i ,z j ) is represented by the cosine similarity of positive sample pairs, sim(z) i ′,z k ) represents the cosine similarity of negative sample pairs, and N represents the number of negative samples.

[0223] In one possible implementation, the determining module 403 can be specifically used for:

[0224] The sparse contrast features in the reconstructed signal are input into the classification network in the recognition model to determine the signal category of the signal in the reconstructed signal.

[0225] Based on the signal category in the reconstructed signal and the Softmax network in the recognition model, the fault type corresponding to the signal category is determined.

[0226] In one possible implementation, the determining module 403 can be specifically used for:

[0227] The sparse contrast features in the reconstructed signal are input into the depthwise separable convolutional units in the classification network to determine the spatial feature information.

[0228] Spatial feature information is input into the inverse residual module of the classification network to obtain the feature classification vector;

[0229] Based on the feature classification vector and the fully connected layer in the classification network, the signal category of the reconstructed signal is determined.

[0230] The residual current detection device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0231] Figure 7 This is a schematic diagram of the residual current detection device provided in this application. Figure 7 As shown, the residual current detection device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0232] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0233] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0234] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0235] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0236] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0237] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0238] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0239] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0240] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0241] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0242] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0243] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0244] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0245] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0246] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for detecting residual current, characterized in that, include: The residual current signal is decomposed to obtain the modal components at different time points in the residual current signal. Based on the sparsity characteristics of the modal components, the modal components are reconstructed to obtain a reconstructed signal; Based on the sparse characteristics of the reconstructed signal, the reconstructed signal is classified to determine the signal category of the signal in the reconstructed signal and the fault type corresponding to the signal category.

2. The method according to claim 1, characterized in that, The process of decomposing the residual current signal to obtain the modal components at different time points in the residual current signal includes: The residual current signal is extended to obtain an extended signal; The extended signal is decomposed to obtain mode candidate functions; Based on preset sparse optimization conditions and residual update conditions, the modal components in the modal candidate function are determined.

3. The method according to claim 2, characterized in that, The step of determining the modal components in the modal candidate function based on preset sparse optimization conditions and residual update conditions includes: Based on preset sparse optimization conditions, the modal candidate functions are subjected to sparse optimization processing to obtain intrinsic modal functions; According to the preset residual update conditions, the residual signal is processed to obtain the reconstruction error, wherein the residual signal is determined based on the extended signal and the mode candidate function; If the reconstruction error meets the preset error conditions, then the modal components are determined according to the inherent mode function.

4. The method according to claim 2, characterized in that, The step of reconstructing the modal components based on their sparsity characteristics to obtain a reconstructed signal includes: Based on the modal component matrix of the modal components, an overcomplete dictionary is constructed; The overcomplete dictionary is subjected to sparse optimization to determine the sparsity coefficients of the overcomplete dictionary; Based on the sparsity coefficients of the overcomplete dictionary, the modal components are filtered to determine the effective modal components. The reconstructed signal is obtained based on the effective modal components and the sparse coefficients.

5. The method according to claim 1, characterized in that, The step of classifying the reconstructed signal based on its sparse characteristics to determine the signal category and the corresponding fault type within the reconstructed signal includes: Based on the feature extraction network, sparse features are extracted from the reconstructed signal; Based on the feature distance between the sparse features, the positive and negative samples in the reconstructed signal are determined; Based on the positive and negative samples in the reconstructed signal, the feature extraction network is optimized to obtain the target feature network; Based on the target feature network, sparse contrast features are extracted from the reconstructed signal; The sparse contrast features in the reconstructed signal are input into the recognition model to determine the signal category of the signal in the reconstructed signal and the fault type corresponding to the signal category. The recognition model is trained based on the sparse sample features and the signal category label corresponding to the sparse sample features.

6. The method according to claim 5, characterized in that, The step of optimizing the feature extraction network based on the positive and negative samples in the reconstructed signal to obtain the target feature network includes: The feature distance between the positive sample and the negative sample is adjusted to obtain the adjusted positive sample and the adjusted negative sample. The loss function is determined based on the adjusted positive samples and the adjusted negative samples; The feature extraction network is optimized based on the loss function to determine the target feature network; the loss function satisfies: Among them, z i ,z j Characterized as positive sample pairs, z i ′ ,z k The negative sample pairs are represented by τ, which is a temperature parameter used to adjust the sparsity of the distribution. sim(z) i ,z j ) is represented by the cosine similarity of positive sample pairs, sim(z) i ′ ,z k ) represents the cosine similarity of negative sample pairs, and N represents the number of negative samples.

7. The method according to claim 5, characterized in that, The step of inputting the sparse contrast features in the reconstructed signal into the recognition model to determine the signal category of the signal in the reconstructed signal and the fault type corresponding to the signal category includes: The sparse contrast features in the reconstructed signal are input into the classification network in the recognition model to determine the signal category of the signal in the reconstructed signal. Based on the signal category in the reconstructed signal and the Softmax network in the recognition model, the fault type corresponding to the signal category is determined.

8. The method according to claim 7, characterized in that, The step of inputting the sparse contrastive features in the reconstructed signal into the classification network of the recognition model to determine the signal category in the reconstructed signal includes: The sparse contrast features in the reconstructed signal are input into the depthwise separable convolutional unit in the classification network to determine spatial feature information; The spatial feature information is input into the inverse residual module of the classification network to obtain the feature classification vector; The signal category of the reconstructed signal is determined based on the feature classification vector and the fully connected layer in the classification network.

9. A residual current detection device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8.

10. A computer program product comprising a computer program that, when executed by a processor, implements the method of any one of claims 1-8.

Images