A method for recording the waveform of fault current in an AC power grid

Abstract

The present invention discloses a method for recording the waveform of fault current in an AC power grid, which relates to the technical field of power grid fault detection. When waveform distortion or high-resistance fault signs are detected in the power grid, the data acquisition and preprocessing module denoises and wavelet-enhances multi-source recorded and simulated data to generate an initial data set with multi-label high-order features; the multi-task learning module combines domain adaptation and adaptive weighted loss to correct the data distribution deviation, and incorporates physical priors such as arc discharge to identify composite faults such as harmonic distortion, short circuit, and grounding; through cross-validation and noise simulation evaluation, the model verifies its robustness with high-order calculus metrics, and can check the feature attention distribution in a visual way. When the real-time recording reaches the threshold, it automatically triggers incremental training and applies a contrast memory mechanism to adaptively absorb new fault information, forming a multi-label diagnosis closed loop, effectively improving the accuracy and continuous optimization ability of multi-scenario fault identification in smart power grids.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid fault detection, and specifically to a method for recording the waveform of fault current in an AC power grid. Background Technique

[0002] With the continuous evolution of the power system towards intelligence, distribution, and diversification, the access of large-scale renewable energy generation and power electronic terminals in the distribution link is increasing day by day, and the complexity and dynamics of power grid operation have increased significantly. Multiple harmonics, transient interference, and non-linear arc discharge characteristics are often superimposed in the voltage and current waveforms, and some lines will also show a time-varying state of repeated power flow inversion due to the reverse power flow of distributed power sources.

[0003] In this context, the fault forms are not only limited to the traditional single three-phase short circuit or single-phase grounding forms, but also multiple faults such as harmonic distortion, high-resistance grounding, and instantaneous overvoltage may occur concurrently during the same fault period, resulting in compound fault phenomena. Since these fault types are coupled or superimposed in the waveform domain, it is often impossible to accurately identify each sub-fault component based on a single feature or traditional simple criterion. On the other hand, the diversity of power grid scale and operating conditions also increases the difficulty of fault sample acquisition, data management, and real-time processing, making it an urgent topic to be broken through in the field of intelligent power grid fault diagnosis to deeply excavate multi-source recording information and maintain the adaptive ability of the model to environmental changes.

[0004] In a Chinese invention patent with the authorization announcement number CN113625125B, a method, device, and equipment for detecting high-resistance grounding faults in a distribution network are disclosed. The method includes: periodically obtaining the magnetic field signal corresponding to the distribution network feeder; performing a first noise reduction operation on the magnetic field signal to generate multiple first-order basic components; performing a second noise reduction operation on each first-order basic component to generate multiple second-order basic components; constructing a target curve according to the numerically largest and second-largest second-order basic components selected from the multiple second-order basic components as the abscissa component and the ordinate component; and judging whether the distribution network feeder has a high-resistance grounding fault according to the comparison result between the target slope corresponding to the target curve and a preset slope coefficient, so as to quickly and accurately identify the high-resistance grounding fault of the distribution network feeder.

[0005] However, considering the actual application scenario and the existing technology:

[0006] In the current process of multi-label recognition and online optimization for compound faults, there is still a core problem: how to accurately identify rare faults (such as high-resistance faults) and concurrent fault types from the massive fault waveform data collected, and achieve adaptive updates in a real-time environment. When the power flow direction, load characteristics, or system topology of the power grid change, it is often difficult to track new fault evolution patterns in a timely manner using a static model for fault analysis. At the same time, the distribution deviation between simulation data and real recorded waveforms will lead to insufficient generalization ability of the model for on-site scenarios; even if the data scale is enhanced through physical simulation, it is difficult to avoid the differences between actual waveforms and simulations in terms of noise, edge sampling accuracy, and arc discharge non-linearity, resulting in fault discrimination deviations. In addition, multi-label fusion requires effective decomposition of different types of fault characteristics such as harmonic distortion, short circuit, and overvoltage within the same waveform. If the physical prior and online learning mechanisms are not combined, when new fault patterns or fault distribution changes occur, it is very easy to cause a sharp drop and missed detection of the recognition results.

[0007] Therefore, the present invention provides a method for recording fault current waveforms of an AC power grid. Summary of the Invention

[0008] (I) Technical problems to be solved

[0009] Aiming at the deficiencies of the prior art, the present invention provides a method for recording fault current waveforms of an AC power grid. The data acquisition and preprocessing module denoises and wavelet-enhances multi-source recorded waveforms and simulation data to generate an initial data set with multi-label high-order features. Subsequently, the multi-task learning module corrects the data distribution deviation by combining domain adaptation and adaptive weighted loss, and incorporates physical priors such as arc discharge to identify compound faults such as harmonic distortion, short circuit, and grounding. Through cross-validation and noise simulation evaluation, the model verifies its robustness with high-order calculus metrics and can check the feature attention distribution in a visual way. Finally, when the real-time recorded waveform reaches the threshold during online deployment, it automatically triggers incremental training and applies the contrast memory mechanism to adaptively absorb new fault information, forming a multi-label diagnosis closed-loop, effectively improving the accuracy and continuous optimization ability of multi-scenario fault recognition in smart grids, and solving the technical problems recorded in the background art.

[0010] (II) Technical solutions

[0011] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0012] A method for recording the waveform of AC power grid fault current, including: when it is detected that the waveform distortion occurs in the power grid or the signs of high-resistance fault appear, the data acquisition and preprocessing module automatically extracts multi-source recording and simulation data when the monitoring result exceeds the pre-set fault threshold; through performing noise reduction and anomaly elimination based on physical laws, then using the wavelet method to enhance and expand the sparse fault samples, and enriching the sample identification through hierarchical indexing and label binding, finally generating an initial data set containing multi-label high-order features and unified formatted storage, which can be used for subsequent multi-label fault analysis;

[0013] After the initial data set is formatted and enters the multi-label modeling stage, the multi-task learning module, after obtaining the high-order features and the input of the time-domain sampling branch, corrects the problem of unbalanced distribution through domain adaptation and adaptive weighted loss function, and incorporates physical priors such as arc discharge to limit the abnormal drift of network weights. At the same time, it performs parallel convolution extraction on the harmonic distortion features and outputs the activation value of the hidden vector at the fusion layer, and finally generates a multi-label classification structure model for composite fault identification;

[0014] After the constructed multi-label model completes the preliminary training and enters the evaluation stage, the cross-validation and noise simulation module examines the stability of the model for various fault scenarios through the analysis of segmentation and high-order calculus metric formula, and checks the attention distribution of key features such as harmonic distortion and high-resistance grounding in the interpretability visualization. If there is no weight value violating the power law in the physical prior constraint layer, it outputs the intermediate model parameters with good robustness and containing the ability to analyze the fault segmentation features;

[0015] When the model completes the offline verification and needs to be put into the edge and cloud collaborative environment, the online deployment and continuous optimization module starts the fault identification inference after the real-time recording wave meets the trigger threshold, and monitors the identification quality of new or rare faults through the high-order deviation metric function. Once a significant drift is detected, it automatically uploads the error samples and labels to trigger incremental training, and at the same time retains the historical knowledge through the comparison memory mechanism, and finally forms a long-term adaptive fault diagnosis closed-loop.

[0016] Preferably, fault waveform data is obtained from multiple sources, and after the obtained original data is preliminarily cleaned, it is named the initial cleaning data, where: the elimination operation is performed on the waveform segments with obvious distortion; the samples with abnormal timestamps or seriously inconsistent with the known working condition information are marked and temporarily stored in the abnormal library; the common-sense error samples are filtered according to the prior physical constraints;

[0017] Preferably, based on the sparse fault types or composite fault waveforms in the initial cleaning data, targeted data enhancement and high-order feature extraction are performed, and an enhanced feature data set is generated. For the fault types with insufficient sample size, synthetic fault data with real waveform statistical features is constructed, and a part of the enhanced data obtained is integrated with the original data and incorporated into the enhanced feature data set;

[0018] Introduce a high-order waveform energy measure to identify the key time-frequency characteristics of composite faults, and define a high-order waveform energy function , the form of which can be based on the integral accumulation of the high-order derivative in the wavelet domain:

[0019]

[0020] In the formula: represents the discrete wavelet transform coefficients of the fault waveform , is the layer scale, is the translation variable; is the order of the high-order derivative ; is the upper limit of the wavelet translation range;

[0021] Preferably, keywords such as waveform sampling points, labels, GAN enhancement identifiers, and high-order energy measures are stored as a hierarchical table; randomly sample the enhanced feature data according to the required ratio to obtain three subsets of training, validation, and testing; perform a consistency check on the diversified samples to make the high-order waveform energy correspond one-to-one with the original waveform, generating and providing preprocessed output data;

[0022] Preferably, define multiple parallel input branches to process time-domain sampling data and high-order eigenvalues respectively;

[0023] After completing the parallel input, introduce a basic convolutional layer to extract local patterns, set up a feature fusion layer, map the features refined by each branch to the same hidden vector space to form a fused hidden vector; according to the set of target fault labels in the power grid, generate a preliminary classification activation output after retaining the number of channels matching the number of labels;

[0024] Preferably, incorporate an adaptively updatable weight coefficient into the loss function, and define the following multi-label weighted loss function :

[0025]

[0026] Among them: represents the classification activation output; is the corresponding true label; is the weighted coefficient of the fault category, which can be dynamically updated with iterations; is a single-label sub-loss function that measures the difference between the activation value and the true label;

[0027] Introduce a cross-domain mapping function Align the fused hidden vectors and define the cross - domain difference metric as follows:

[0028]

[0029] In the formula: and represent the input features in the simulation domain and the real domain respectively; is a learnable cross - domain mapping function, means that after mapping the feature vector h to the hidden space, at dimension a certain output value or component below; is the domain of definition of the hidden space index; is the fractional - order (or higher - order) derivative of the variable z, where is the fractional - order order; when it degenerates into the conventional first - order derivative; when it is the fractional - order derivative; is the balance coefficient, represents the vector norm or absolute - value operation;

[0030] Preferably, add monotonicity or non - linear transition restrictions to the activation function of the convolutional channels, and achieve it by imposing an additional regular constraint function on the weights of each convolutional kernel:

[0031]

[0032] In the formula: W is the set of weights of all convolutional kernels in this sub - channel; represents the positive activation weight determined to correspond to the high - resistance fault mechanism in the m - th convolutional kernel; is a predefined constraint coefficient, is the overall weight of the physical prior regularization term;

[0033] Under the physical prior constraints, retrain or fine - tune the domain adaptation module and the multi - label adaptive weighted loss function, and output the initial multi - label model after several rounds of iteration, including the network structure and initial weights;

[0034] Preferably, screen the subset with the widest coverage of fault label categories and diverse time - frequency features from the training dataset and match - load it with the initial multi - label model; for categories with scarce sample sizes such as high - resistance faults, perform oversampling or increase the training frequency in the pre - processed output data of the original dataset, and cooperate with the adaptive weighted loss function to increase the attention to rare categories;

[0035] Send the prepared training set and the initial multi - label model into the deep learning framework, according to the comprehensive loss function Perform several rounds of iteration to obtain the preliminarily optimized network weights and generate the first intermediate training model;

[0036] Preferably, divide the training set into folds, and take one fold as the validation set one by one, and the remaining folds as the training set, and perform training-validation cycles on the first intermediate training model, and record the multi-label classification results after each training;

[0037] Introduce additional noise simulation in the training or validation stage, perform amplitude perturbation or random high-frequency interference injection on the waveform, and introduce the following robustness metric function :

[0038]

[0039] In the formula: represents the response value shown by the model for a certain key performance index when the noise amplitude is under the -fold validation;

[0040] Different correspond to different intensities or types of noise injection, is the value range of the noise amplitude or interference intensity;

[0041] represents derivative of with respect to the noise amplitude order, ; represents the value of the continuous wavelet transform coefficients of in the noise dimension at one or several scales , ,, is the balance coefficient; combining the cross-validation results and the robustness metric function , perform further fine-tuning iteration on the first intermediate training model until the validation metrics for each fold converge basically, and finally form a more reliable model weight and structure configuration, and the output is the second intermediate training model;

[0042] Preferably, by drawing an attention distribution map, show the high-order response correlation of the key intervals concerned by the model in the time domain or frequency domain of the waveform: check the change slope or change curve of the model output label when the input fault waveform is fine-tuned on the key features, so as to judge whether the model strictly identifies around the physical prior. After the interpretability verification and physical completeness check, an interpretable training model is obtained;

[0043] Preferably, install a fault identification and reasoning module at the edge device of the nearby substation to achieve low-latency fault detection. Retain the global data aggregation and analysis function on the cloud platform for storing historical fault records and subsequent evaluation;

[0044] Import the network structure and weights of the interpretable training model into the edge inference engine, and set the real-time sampling frequency, batch size, and number of core threads according to the grid operation parameters; Continuously monitor the fault identification latency, accuracy, and resource occupancy within a set period, collect the multi-label output results of online identification, and compare them with known fault events, and record the real-time error list;

[0045] Preferably, compare the real-time identification results reported by the edge side with the fault event records, and statistically calculate the accuracy and missed detection rate for each sub-fault type and centrally store them in the incremental update model;

[0046] Define a high-order deviation metric function , and the formula is as follows:

[0047]

[0048] In the formula: represents the multi-label fault identification deviation at time ; represents the -th order derivative operation with respect to time is the fractal dimension metric function, is the trade-off coefficient, is the time window size;

[0049] When the high-order deviation metric function exceeds the deviation threshold, trigger an update; If the high-order deviation metric function exceeds the deviation threshold or a new fault category is detected, pack the corresponding data and labels and send them back to the cloud;

[0050] For retraining or incremental training of new data, generate a new incremental update model; After training is completed, send the incremental update model back to the edge side and automatically replace the currently used fault identification model to form a deployed update model;

[0051] Preferably, appropriately retain the key data and its feature distribution in the early stage during training, or use regularization constraints to suppress large offsets of network weights, and introduce a contrastive memory constraint function , as follows:

[0052]

[0053] In the formula: is the set of weights of the previous stable model version, and W is the weight after the current retraining;

[0054] For The nth derivative measures the distribution pattern of the difference between the weights of the new and old models in the weight dimension ;

[0055] When it is confirmed that the model recognition result is incorrect or missed, this data is archived and manually labeled to form operation and maintenance labeled data, and the operation and maintenance labeled data is periodically incorporated into the retraining set; after multiple rounds of updates, merged memories, and operation and maintenance feedback, an integrated and deployed updated model is obtained;

[0056] (III) Beneficial effects

[0057] The present invention provides a method for recording the fault current waveform of an AC power grid, having the following beneficial effects:

[0058] This solution provides an overall solution idea with high precision, high robustness, and sustainable evolution for composite fault recognition from multi-source data acquisition and preprocessing to online deployment and continuous optimization, including:

[0059] By first obtaining the initial cleaned data and enhanced feature data and outputting the preprocessed output data, it can effectively balance the sample distributions of common faults and rare faults (such as high-resistance grounding), and with the assistance of wavelet transform or GAN data augmentation, strengthen the extraction of complex waveform features such as transient spikes and harmonic distortions, retaining both the diversity of real fault scenarios and highlighting key fault signals with high-order energy metrics or fractal features, so that the model training stage is no longer limited to a single feature dimension;

[0060] The initial multi-label model realizes parallel discrimination of multiple fault types such as harmonic distortion, short circuit, and grounding at the levels of fused hidden vectors and classification activation outputs through parallel feature embedding and domain adaptation modules, and introduces an adaptive weighting factor and physical prior regularization into the loss function to improve the model's adaptability to high-resistance faults and the differences between the real-simulation domains, significantly reducing the misjudgment probability of the model in the face of multi-label composite fault scenarios and making the network structure more interpretable;

[0061] The training of intermediate model one and training of intermediate model two are sequentially tested through cross-validation and noise simulation, measure the robustness with high-order calculus metrics, and verify the consistency of power mechanism and network weights under the visualization mechanism. The finally output interpretable training model not only avoids overfitting relying on a single scenario but also ensures stable recognition performance under various working conditions;

[0062] The constructed real-time detection and online update process enables the real-time error list to monitor the fault identification deviation at any time. Once a significant deviation is detected, it automatically triggers cloud incremental training, generates an incremental update model, and quickly deploys it into the final deployed model. Combined with the operation and maintenance annotation and contrast memory mechanism, the model can continuously adapt to new fault samples and environmental changes while avoiding forgetting the learned knowledge. Generally speaking;

[0063] This solution effectively solves the problems of data imbalance, simulation-real domain difference, model interpretability, and online update in multi-label composite fault identification, and can be widely applied to the detection and diagnosis of multi-scenario faults in smart grids. Brief Description of the Drawings

[0064] Figure 1 It is a schematic flowchart of the method for recording the fault current waveform of the AC power grid of the present invention. Detailed Embodiments

[0065] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0066] Please refer to Figure 1 , the present invention provides a method for recording the fault current waveform of an AC power grid, including,

[0067] In a smart grid, the fault types are becoming increasingly diverse and prone to overlap with each other, thus forming composite faults and causing the waveform characteristics to become complex. In order to achieve more accurate multi-label fault type identification in the subsequent steps, it is necessary to perform sufficient and detailed preprocessing and preliminary analysis on multi-source fault waveform data in the early stage; among them:

[0068] Step 1: When it is detected that the waveform of the power grid is distorted or there are signs of high-resistance faults, after the detection threshold reaches the set threshold, the data acquisition and preprocessing module automatically extracts multi-source recording wave and simulation data;

[0069] By performing denoising and anomaly rejection based on physical laws, then using the wavelet method to enhance and expand rare fault samples, and enriching the sample identification through hierarchical indexing and label binding, finally generating an initial data set containing multi-label high-order features and unified formatted storage, which can be used for subsequent multi-label fault analysis;

[0070] The said Step 1 includes the following contents:

[0071] Step 101: Acquisition and preliminary cleaning of multi-source waveform data

[0072] Obtain fault waveform data from multiple sources, including collecting fault waveforms triggered under various operating conditions through a real power grid fault recorder (or online monitoring system), and generating synthetic fault data using a high-fidelity simulation platform under different distribution network parameters, fault point locations, and interference conditions;

[0073] After preliminary cleaning of the obtained raw data, it is uniformly named initial cleaned data, where:

[0074] Perform deletion operations on significantly distorted waveform segments (such as constant value curves caused by sensor failures); mark and temporarily store samples with abnormal timestamps or seriously inconsistent with known operating condition information in the abnormal library; filter out common sense error samples based on prior physical constraints (such as intervals where current and voltage cannot be negative at the same time);

[0075] It can initially ensure data quality and exclude invalid information caused by sensor defects or communication failures; at the same time, unify real and simulated waveforms into the initial cleaned data; by fusing real power grid recordings and simulation data, the diversity of fault samples is improved. With preliminary cleaning and physical rule filtering, obvious abnormal or invalid waveforms can be effectively removed, reducing interference in subsequent analysis, and laying a high-quality and wide-coverage data foundation for subsequent data enhancement and feature extraction;

[0076] Step 102: Data enhancement and high-order feature extraction

[0077] Based on rare fault types (such as high-resistance faults) or composite fault waveforms in the initial cleaned data, perform targeted data enhancement and high-order feature extraction, and generate an enhanced feature data set, focusing on the expansion of rare samples and the mining of key features. The core idea is to achieve a deeper characterization of composite fault signals through high-order transformation measures to support subsequent multi-label recognition models, where:

[0078] For fault types with insufficient sample sizes such as high-resistance faults, use generative adversarial networks (GANs) or wavelet perturbation methods to construct synthetic fault data with real waveform statistical characteristics, balance the distribution of different fault categories in the initial cleaned data, and integrate some of the enhanced data obtained with the original data into the enhanced feature data set;

[0079] In the generated waveform samples, introduce high-order waveform energy measures to identify the key time-frequency characteristics of composite faults. To depict more sensitive transient details, define a high-order waveform energy function , and its form can be based on the integral accumulation of high-order derivatives in the wavelet domain:

[0080]

[0081] In the formula: represents the fault waveform The discrete wavelet transform coefficients obtained are for the layer scale and is the translation variable; is the order of the high-order derivative and is used to highlight the drastic changes in the waveform in the high-frequency or spike region; is the upper limit of the wavelet translation range, which is determined according to the waveform sampling duration and resolution;

[0082] This high-order waveform energy is used to quantify the transient feature intensity of the composite fault at a specific wavelet scale. When embedding it as a key feature into the subsequent fault identification model, it can improve the sensitivity to waveform variations caused by the superposition of multiple types of faults such as harmonic distortion and short circuits; when used, it can significantly increase the diversity and quantity of rare fault waveforms, so that the subsequent model training can avoid over-biasing towards common fault types. Through high-order waveform energy and other advanced waveform measures, it can more effectively distinguish the transient features of various faults and avoid subtle differences that cannot be captured by simple time-domain or frequency-domain methods. Using wavelet transform or GAN for data augmentation of rare samples such as high-resistance faults, balancing the fault category distribution, and extracting key features such as the superposition of transients and harmonics by defining high-order energy measures, making the composite fault features more prominent in the high-frequency or spike region.

[0083] Step 103, Standardize data format and diversity

[0084] After completing data augmentation and high-order feature extraction, it is necessary to perform unified formatting and diversity operations on the augmented feature data to form the final preprocessed output data, which includes high-order feature values such as high-order waveform energy , label information (the fault type or combination corresponding to each waveform), and the denoised time-frequency domain sampling data. The specific execution process is as follows:

[0085] Store keyword fields such as waveform sampling points, labels, GAN augmentation flags, and high-order energy measures in a consistent data structure, such as a hierarchical table or binary file format, to ensure that multi-label and time-frequency features can be successfully indexed in the same structure; randomly sample the augmented feature data according to the required ratio (such as 6:2:2), and ensure that the distribution of each fault category is relatively balanced to obtain three subsets for training, validation, and testing; perform a consistency check on the diverse samples to ensure that each sample has complete waveform data and corresponding labels, and ensure that the high-order waveform energy corresponds one-to-one with the original waveform;

[0086] Through format unification and diversification, it provides preprocessed output data for subsequent multi-label model training, and realizes the functional division of data in the form of training, validation, and test sets. It synchronizes and verifies the high-order energy measure and the original waveform, reduces confusion or incorrect calls during data usage, improves the reliability of model training and evaluation, and writes the GAN enhanced identifier into the metadata, which helps to conduct differential evaluation in the subsequent model training and validation stages and observe the impact of real waveforms and synthetic waveforms on the recognition performance.

[0087] Step 101 focuses on preliminary cleaning and multi-source data fusion, providing a good foundation for data augmentation and feature extraction in Step 102: Step 102 conducts targeted augmentation and high-order measure extraction on the basis of the cleaned data, and outputs a richer and more discriminative waveform set. Step 103 then normalizes and diversifies its results into preprocessed output data.

[0088] Step 2: After the initial dataset is formatted and enters the multi-label modeling stage, the multi-task learning module, after obtaining high-order features and the input of the time-domain sampling branch, corrects the problem of uneven distribution through an adaptive weighted loss function, incorporates physical priors such as arc discharge to limit abnormal drift of network weights, and at the same time conducts parallel convolution extraction on harmonic distortion features and outputs the activation value of the hidden vector at the fusion layer, finally generating a multi-label classification structure model for composite fault recognition;

[0089] The above Step 2 includes the following contents:

[0090] Step 201, Multi-label Feature Embedding and Primary Network Construction

[0091] Based on the multi-source waveform data and high-order features provided by the preprocessed output data, construct the initial input layer and feature embedding layer of the network, and uniformly map the high-order waveform energy measure and the original waveform time-domain sampling points from the preprocessed output data to the input vector of the neural network to ensure that the model pays attention to both transient features and global features;

[0092] Input layer configuration: Define multiple parallel input branches to process time-domain sampling data and high-order feature values respectively. For example, convolution kernels more sensitive to high-frequency transients can act on the high-order feature branch, while convolution kernels more sensitive to the overall waveform trend can act on the original waveform branch;

[0093] Basic convolution and fusion layer: After completing the parallel input, introduce a basic convolution layer (which can be a one-dimensional convolution Conv1D or a two-dimensional convolution Conv2D, depending on the waveform data arrangement) to extract local patterns; then set up a feature fusion layer to map the features extracted from each branch to the same hidden vector space to form a fused hidden vector H201;

[0094] Initial multi-label mapping: Based on the set of target fault labels in the power grid (such as harmonics, short circuits, high-resistance grounding, etc.), after retaining the number of channels matching the number of labels, a preliminary classification activation output is generated;

[0095] At the early stage of the network, the original waveform and high-order features are fused, making the model more sensitive to weak or rapidly changing signals generated by complex faults. Through parallel input and feature fusion, the model's ability to capture different types of fault features is effectively improved. With the help of the multi-channel parallel structure and high-order feature fusion, the limitation that traditional single-channel convolution may dilute complex fault features is broken through.

[0096] Step 202, Multi-label weighted loss function and domain adaptation module

[0097] On top of the classification activation output, an adaptive weighted loss function is defined for the characteristics of multi-label scenarios and uneven data distributions of different fault categories; to narrow the possible distribution differences between real waveforms and simulation waveforms, a domain adaptation module is introduced to make the model more suitable for the actual power system operation environment. The execution process is as follows:

[0098] In the multi-label recognition problem, each sample often contains multiple target fault labels simultaneously. To take into account the error impact of rare fault categories (such as high-resistance faults), a weight coefficient that can be adaptively updated needs to be incorporated into the loss function , define the following multi-label weighted loss function :

[0099]

[0100] Where: represents the classification activation output; is the corresponding true label (1 if this fault type occurs, otherwise 0); is the weighted coefficient of the fault category, which can be dynamically updated with iterations; is a single-label sub-loss function for measuring the difference between the activation value and the true label (such as based on logistic regression or cosine similarity, etc., more specifically measuring the discrimination of different sub-fault types);

[0101] This design can adaptively increase the weight of rare categories during training and reduce the risk of common categories dominating the loss;

[0102] Regarding the problem of distribution differences between simulation data and real data, a cross-domain mapping function is introduced to align the fused hidden vector H201, making the model have a more consistent feature expression when processing simulation data or real data, and defining the cross-domain difference metric as follows:

[0103]

[0104] In the formula: and respectively represent the input features in the simulation domain and the real domain; is a learnable cross-domain mapping function, indicating that after mapping the feature vector h to the latent space, at dimension a certain output value or component below; is the domain of definition of the latent space index;

[0105] is the fractional-order (or higher-order) derivative of the variable z, where is the fractional-order; when it degenerates into the conventional first-order derivative; when it is the fractional-order derivative; when

[0106] , is the balance coefficient, used to weigh the importance of the numerical difference of the mapping output itself (the first term) and the change difference of the output in the latent space (the second term); if is closer to 1, more attention is paid to the numerical gap: if k is closer to 0, more attention is paid to the alignment of the deformation process of the mapping function; represents the vector norm or absolute value operation (the norm, norm, etc. can be adopted according to specific requirements), ensuring the non-negativity of the difference description;

[0107] Cross-domain difference metric measures the distance between the features of the two domains in the mapped space. The sampling range Z depends on the dimension of the latent vector. Finally, is incorporated into the overall loss, aiming to minimize this value, thereby reducing the difference between the simulation and the real data distribution and improving the performance of the model in the actual scenario. The combined loss obtained by combining multi-label weighting and domain adaptation is:

[0108]

[0109] where is the balance coefficient of the domain adaptation term, used to adjust the relative importance of multi-label classification and cross-domain alignment.

[0110] The multi-label adaptive weighted loss function can effectively improve the model's attention to rare fault types, avoid overfitting to common faults caused by unbalanced sample distribution, and align the simulation and real waveforms at the latent vector level, which can alleviate the recognition bias caused by distribution differences during actual deployment of the model. The organic combination of multi-label weighting and domain adaptation improves the recognition accuracy and generalization ability of the model in complex power scenarios. The adaptively updated weight The mechanism can dynamically adjust the attention to each fault category during the training process, enabling the model to spontaneously adapt to different fault distribution changes.

[0111] Step 203: Deep mapping incorporating power physics prior

[0112] Incorporate the physical prior of specific fault mechanisms in the power system into the network structure or activation function, so that the model does not violate the basic power system mechanisms when identifying compound faults. Driven by the obtained comprehensive loss, further perform structural constraints and parameter correction on the network, and finally output the model structure and initial training weights.

[0113] According to the characteristics of power mechanisms such as arc discharge and line coupling, add physical prior constraints to the sub-channels specifically set for high-resistance faults or harmonic distortion faults in the network. Among them, add a certain monotonicity or non-linear transition limit to the activation function of the convolutional channel to avoid the network learning mappings that are contrary to the arc discharge law. Such priors can be achieved by imposing an additional regular constraint function on the weights of each convolutional kernel:

[0114]

[0115] Where: W is the set of weights of all convolutional kernels in this sub-channel; Represents the positive activation weight in the m-th convolutional kernel determined to correspond to the high-resistance fault mechanism;

[0116] Is a predefined constraint coefficient used to limit the sign flip or abnormal sudden increase of the weights in the convolutional kernel that violate the physical characteristics of arc discharge; Is the overall weight of the physical prior regularization term to prevent excessive interference with the normal fitting ability of the network;

[0117] Under the physical prior constraints, retrain or fine-tune the domain adaptation module and the multi-label adaptive weighted loss function, so that the network not only takes into account rare faults and cross-domain alignment, but also strictly adheres to the basic fault laws of the power system; finally, after several rounds of iteration, output the initial multi-label model as the model result of this step, including the network structure (such as the size of each layer's convolutional kernel and the number of feature channels) and the initial weights.

[0118] Through physical prior constraints, the mapping that conflicts with the actual operation rules of the power grid learned by the model is significantly reduced, the interpretability and robustness of complex faults are enhanced, which complements the domain adaptation mechanism, and ensures that the model can still maintain recognition consistency in a wider range of power operating conditions; Sub-step 201 completes multi-source feature fusion in the early stage of the network, providing the basic activation values for sub-step 202 to perform multi-label weighting and domain adaptation; Sub-step 202 uses the adaptive weighted loss function and cross-domain alignment mechanism to effectively handle rare faults and simulation-real distribution biases. On this basis, sub-step 203 further integrates the power physics prior, enabling the model to have a higher degree of fit to the real fault mechanism, and at the same time outputting the preliminary training weights, laying a solid foundation for subsequent training evaluation.

[0119] For example, the high-resistance fault current shows pulsed or fuse-like non-linear fluctuations in the low-value range. It can be declared that the fluctuation range is approximately 0 - 75A, and the activation function is clipped in combination with a specific threshold.

[0120] For harmonic faults, the ratio limit of the harmonic component to the fundamental wave amplitude (such as not exceeding a certain safety factor) can be introduced in the network, so that the features learned by the convolution kernel or the mapping layer do not violate the basic distribution law of harmonics in the power grid; Taking arc discharge as an example, the typical arc current frequency band observed in simulation and actual tests (such as inter-harmonic peak values, etc.) can be disclosed, and how to impose restrictions on the network weights corresponding to this frequency band in the regularization term is explained.

[0121] Step 3: After the constructed multi-label model completes the preliminary training and enters the evaluation stage, the cross-validation and noise simulation module examines the stability of the model for various fault scenarios through the analysis of the splitting and high-order calculus metric formula, and verifies the attention distribution of key features such as harmonic distortion and high-resistance grounding in the interpretability visualization.

[0122] If there are no weights that violate the power laws in the physical prior constraint layer, the intermediate model parameters with good robustness and the ability to analyze fault subdivision features are output.

[0123] The said Step 3 includes the following contents:

[0124] Step 301: Multi-source training data integration and distribution balance strategy

[0125] Based on the preprocessed output data and the initial multi-label model, first perform the distribution balance processing on the training set, and then start the initial training iteration. Specifically, it includes:

[0126] According to the training data set (including real and simulated fault waveforms, multi-label annotations, and high-order features), select the subset with the widest coverage of fault label categories and diverse time-frequency features from the training data set and load it in matching with the initial multi-label model.

[0127] For sample - scarce categories such as high - resistance faults, by performing oversampling or enhancing the training frequency in the output data of the original dataset pre - processing, the proportion of each fault label in the training batches is made more balanced, so as to cooperate with the adaptive weighted loss function and increase the attention to rare categories; the prepared training set and the initial multi - label model are sent into the deep - learning framework together, and according to the comprehensive loss function (including multi - label weighting and domain adaptation) perform several rounds of iteration to obtain the preliminarily optimized network weights and generate the first intermediate training model;

[0128] During use, the combination of oversampling and adaptive weighting focuses on solving the problem of insufficient training volume for rare - category faults, effectively uses the pre - processing output data and the initial multi - label model for initial training iteration to obtain the first intermediate training model. The fusion of multi - source data ensures that the model covers more working conditions and initially reduces the risk of overfitting;

[0129] Step 302: Cross - validation and noise reliability assessment

[0130] Based on the first intermediate training model, test the generalization ability and robustness of the model through cross - validation and noise simulation, and gradually iterate the model parameters during the training process to generate a more reliable intermediate result, the second intermediate training model. To achieve a higher - level metric, a high - order calculus - type evaluation formula is used to measure the consistency and effectiveness of the model's performance in different data analyses; where:

[0131] Divide the training set into folds (where the common choices are 5 or 10). One fold is used as the validation set one by one, and the remaining folds are used as the training set to perform training - validation cycles on the first intermediate training model. After each training, record the multi - label classification results (such as accuracy, precision, etc.) and uniformly summarize them into the metric items;

[0132] Introduce additional noise simulation in the training or validation stage (corresponding to the denoising in step one), perform amplitude perturbation or random high - frequency interference injection on the waveform, observe the degree of decrease in the model recognition rate. To quantitatively represent the reliability performance, introduce the following robustness metric function :

[0133]

[0134] In the formula: represents the response value of the model to a certain key performance indicator (such as multi - label accuracy) when the noise amplitude is in the th fold validation (or the th group of data); different For different intensities or types of noise injection, is the value range of the noise amplitude or interference intensity;

[0135] denotes the order (or fractional order) derivative with respect to the noise amplitude, ;

[0136] If takes a non-integer value, the fractional order derivative defined by Riemann-Liouville or Caputo can be used to capture more complex progressive characteristics in high-resistance fault or arc discharge scenarios;

[0137] denotes the value of the continuous wavelet transform (CWT) coefficients in the noise dimension at one or several scales to measure the local oscillation degree of the response with respect to the change in noise; is the wavelet scale or frequency band, which is often used to identify violent fluctuations of different magnitudes. When is small, it focuses on depicting rapid detail fluctuations, when it is large, it focuses on the overall change trend;

[0138] , is the balance coefficient, which is used to make a trade-off between the fractional order derivative metric term and the wavelet metric term; when is close to 1, the evaluation of the model robustness depends more on the measurement of the slope or step change of in the noise dimension; when α is close to 0, it emphasizes more on capturing the mutations or detail jitters in the wavelet domain;

[0139] Based on the comprehensive cross-validation results and the robustness metric function , the intermediate training model one is further fine-tuned and iterated until the validation metrics for each fold basically converge, and finally a more reliable model weight and structure configuration are formed, and the output is the intermediate training model two;

[0140] During use, cross-validation fully checks the performance of the model on multi-fold data under different fault scenarios and different data splitting methods, which helps to discover potential overfitting problems or biases towards certain types of faults; by using the robustness metric function to measure the change curve of the accuracy or other metrics when injecting noise, the tolerance of the model to interference can be evaluated more precisely. Introducing the fractional order derivative into the evaluation of the performance metrics with respect to the change in noise breaks through the limitations of the traditional measurement of the difference in accuracy or the error variance, making the reliability test more in-depth and sensitive.

[0141] Step 303, Interpretability Analysis and Model Completeness Verification

[0142] Using the robust network weights and structure of the trained intermediate model 1, perform interpretability analysis on the model; at the same time, combined with the mechanism constraints of the power system, verify the completeness of the model to ensure that it will not produce abnormal decisions that significantly violate physical laws in real application scenarios. Finally, output the interpretable training model as the final result of step three; among them:

[0143] By drawing the attention distribution map, display the high-order response correlation of the key intervals concerned by the model at the waveform time domain or frequency domain level: check the change slope or change curve of the model output label when the input fault waveform is fine-tuned on key features (such as arc discharge features) to determine whether the model strictly identifies around physical priors;

[0144] Combined with power physics prior constraints (such as the non-linear characteristics of arc discharge), sample and detect the weight distribution of the internal convolution kernels or mapping functions of the model to check whether there are extreme weights that violate the prior; for fault types with high mutual correlation (such as harmonic distortion and overvoltage), check the prediction coherence of the model when both types of faults occur simultaneously to prevent the model from only identifying one and ignoring the actual occurrence of the other fault; for example: if convolution kernels or feature channels are used to detect arc features, the range is defined as the high-frequency arc region of interest, and the amplitude of the convolution kernel weights in this frequency band is regarded as the key monitoring object. For example, when most of the activated channels fall in the lower frequency band but there are some high-amplitude high-frequency convolution kernel weights, and their distribution cannot be explained by conventional faults (such as three-phase short circuit, low-frequency harmonics), it is determined that the prior of the arc non-linear characteristics may be violated; when weights with a strong corresponding relationship with this harmonic band are found in the convolution kernel or feature mapping layer but do not match the annotation, it can be determined that there is an illegal distribution. For example, if the expected harmonics are strong around the 5th order, but the network unreasonably amplifies at the 10th order and above, it may indicate that the model confuses other fault features or there are weight deformations.

[0145] In the inference link, output multi-label results for each sample, such as Fault Type 1: Harmonic Distortion; Fault Type 2: Overvoltage; Fault Type 3: High-Impedance Fault, etc. If some labels have strong correlations (such as harmonic distortion and overvoltage) that usually exist simultaneously in the true data annotation, but only a single label is given in this inference, there may be a missed fault. Therefore, a layer of coherence verification can be added after the inference, and the rule is defined as: when the correlation between label A and B exceeds R% (obtained according to historical statistics or power physics mechanism), if A is detected but the confidence of B is extremely low, then recalculate or refer to the associated fault probability to improve the recognition accuracy of the scenario where both of them exist simultaneously.

[0146] After the interpretability verification and physical completeness check, the final form and parameters of the obtained model are named the interpretable training model, which has both the ability to identify complex fault scenarios and a certain degree of interpretability;

[0147] When in use, through interpretability analysis and physical constraint verification, the model can provide a reasonable explanation for the decision-making process when facing complex or new power faults, reduce the uncertainty in actual applications, ensure the logical association and consistency between different faults in multi-label fault identification, and avoid the phenomenon of one-sided identification or ignoring associated faults. Combining high-order response correlation analysis with power physics prior knowledge provides a more systematic means for the interpretability evaluation of the model under complex waveform features.

[0148] The distribution balance, noise reliability, and interpretability during the training process are hierarchically optimized to improve the actual applicability and understandability of the model. The output interpretable training model not only ensures the high-precision identification ability of multi-label composite faults but also takes into account the robustness and interpretability consistent with the power system mechanism. Combining the attention mechanism or Grad-CAM, the key attention points of the model in the waveform time domain or feature domain are visualized, and the distribution of network parameters is detected based on power physics prior knowledge to prevent abnormal decisions that seriously violate the mechanism.

[0149] Step 4: When the model completes offline verification and needs to be deployed in an edge-cloud collaborative environment, the online deployment and continuous optimization module starts fault identification reasoning after the real-time recording wave meets the trigger threshold, and monitors the identification quality of new or rare faults through a high-order deviation metric function. Once a significant drift is detected, it automatically uploads the error samples and labels to trigger incremental training, and at the same time retains historical knowledge through a comparison memory mechanism, finally forming a long-term adaptive fault diagnosis closed-loop;

[0150] The content of the above Step 4 includes the following:

[0151] Step 401: Online deployment of the model and initial operation monitoring

[0152] Install a fault identification reasoning module on the edge device of the nearest substation to achieve low-latency fault detection. Retain the global data summary and analysis function on the cloud platform for storing historical fault records and subsequent evaluation; the fault identification reasoning module can be used to make a preliminary determination of waveform distortion, abnormal current, or harmonic content, decide when to call the reasoning process, and at what level of detection threshold to perform emergency reasoning or regular reasoning. For example, a high-resistance fault current threshold of 10A - 30A (combined with factors such as the system voltage level) is configured in the substation automation system. When the current in the data stream is continuously detected to exceed this threshold and shows an atypical waveform distortion, the reasoning module is triggered to start fault analysis.

[0153] Import the network structure and weights of the interpretable training model into the edge inference engine, set the real-time sampling frequency, batch size, and the number of core threads according to the power grid operation parameters to ensure fast response; continuously monitor the fault identification delay, accuracy, and resource occupancy within the set time period, collect the multi-label output results of online identification, and compare them with the known fault events (if any), and record the real-time error list;

[0154] Through the collaborative deployment of the edge and the cloud, it can balance real-time performance and global data management. The inference performance of the model in the initial stage is verified in the real environment, laying a foundation for subsequent continuous optimization. The multi-label feature extraction and power physical constraints in the interpretable training model are actually verified here, and the output error set, the real-time error list, is used for model deviation detection and update triggering;

[0155] For online inference, the priority is usually higher than batch retraining; the computing tasks can be queued and scheduled in the cloud or at the edge, and the inference tasks are set to high priority to ensure real-time performance. When a new fault mode appears, only the key data segments need to be uploaded for small-batch supplementary training, and there is no need to frequently transmit a large amount of full-scale fault data. For example, how much computing power core should be reserved for online inference in a power substation or an edge server: the training tasks should run in the background and automatically cede resources to the inference threads.

[0156] Step 402, Intelligent Deviation Detection and Dynamic Retraining Trigger

[0157] Utilize the collected online identification results and error information to monitor the possible accuracy degradation or fault mode drift of the model. When significant deviations or new fault modes are found, automatically trigger the retraining or correction process to maintain the model performance. The specific steps are as follows:

[0158] Compare the real-time identification results reported by the edge side with the fault event records, count the accuracy and missed detection rate for each sub-fault type and centrally save them in the incremental update model; define a high-order deviation metric function , to capture the differences and trends between the model and the actual faults within different time windows. The example formula is as follows:

[0159]

[0160] In the formula: Characterizes the multi-label fault identification deviation at time (for example, a mapping value of a certain high-level error metric or a multi-label error set), which can be defined as the difference degree between the model output and the true label according to the actual scenario;

[0161] Represents the -order derivative operation with respect to time ​ It can be of integer order or fractional order (such as Caputo, Riemann-Liouville, etc.), and is used to characterize the mutation, slow change or memory characteristics over time. When it is the case, it can more sensitively capture the metastable or mutation region of the deviation curve. If it is, then it is more sensitive to severe jitters of the second order and above; is a fractal dimension metric function, which is used to evaluate the complexity or irregularity degree of the deviation curve at a very small scale. Common definition methods include box dimension-based, fractal box counting, or local fractal dimension based on wavelet decomposition;

[0162] For example, the wavelet domain fractal metric can be adopted:

[0163]

[0164] where is the continuous wavelet coefficient of , is the wavelet scale, and are adjustable coefficients, and this integral reflects the fractal structure of the deviation curve at different scales;

[0165] is a trade-off coefficient, which is distributed between the derivative term and the fractal dimension metric term. When is close to 1, more attention is paid to the calculus properties of the deviation over time (such as the severity of multiple-order derivatives). When is close to 0, more attention is paid to the roughness or detail complexity of the deviation itself in the fractal sense;

[0166] is the time window size, which is used to accumulate the overall trend of the deviation within the period from to ; in the power system, it can be set according to the average duration of the fault event, the sampling period, etc.;

[0167] When the high-order deviation metric function exceeds the deviation threshold, it indicates that the model has a large amplitude or sudden recognition error recently, triggering an update;

[0168] If the high-order deviation metric function exceeds the deviation threshold or a new fault category is detected (such as a special harmonic overvoltage mode that has not appeared before), the corresponding data and labels will be packed and sent back to the cloud;

[0169] Load the network structure and weights of the interpretable training model, start retraining or incremental training for new data to generate a new incremental update model; after training is completed, transmit the incremental update model back to the edge side and automatically replace the currently used fault identification model to form a deployed update model;

[0170] Capture and quantify the accuracy change of the model in a dynamic environment in real time, avoid automatically triggering retraining when being in a misaligned state for a long time, and introduce new fault mode data in a timely manner, enabling the model to have the ability of self-iteration and improvement, high-order deviation metric function Adopt fractional derivative integration, which can more sensitively capture the prediction error fluctuations caused by the drift of fault modes or the surge of noise interference, avoid being in a misaligned state for a long time, and enable the model to maintain a rapid adaptation to the dynamic changes of the power grid.

[0171] Step 403, Knowledge Consolidation and Collaborative Update

[0172] To prevent the model from forgetting old fault modes or high-resistance fault characteristics during incremental training, key data and its feature distribution in the early stage can be appropriately retained during training, or regular constraints can be used to suppress large offsets of network weights, introducing a contrastive memory constraint function , as follows:

[0173]

[0174] In the formula: is the weight set of the previous stable model version, and W is the weight after the current retraining;

[0175] is the order derivative (optional fractional order) measures the distribution form of the weight difference between the new and old models in the weight dimension;

[0176] When the operation and maintenance engineer or the fault investigation record confirms that the model recognition result is incorrect or missed, archive this data and manually label it to form operation and maintenance labeled data, and periodically incorporate the operation and maintenance labeled data into the retraining set to strengthen the learning depth of error-prone scenarios or difficult faults and improve the overall accuracy of the model; after multiple rounds of updates, merged memories, and operation and maintenance feedback, the cloud integrates the merged model versions to obtain the deployed update model, which is deployed at the edge side and enters normal operation; the deployed update model retains the ability to identify previous fault modes and is also adaptable to new fault modes, becoming the model finally put into use in the fourth step stage;

[0177] Through the contrastive memory constraint function , effectively alleviates the problem of forgetting historical knowledge caused by frequent updates, maintains the recognition accuracy of known fault modes, and can gradually purify the data and strengthen the recognition performance for rare and complex scenarios by using the secondary annotation of operation and maintenance engineers; significantly reduces the missed reports of error-prone faults, and the output deployed update model integrates new and old knowledge in continuous iteration, and long-term supports the recognition of multi-fault scenarios in smart grids.

[0178] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0179] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0180] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only for some logical function divisions, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling, direct coupling, or communication connection can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0181] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0182] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A method for recording the waveform of fault current in an AC power grid, characterized in that: including When it is detected that the waveform of the power grid is distorted or there are signs of high-resistance faults, and the monitoring results have exceeded the preset fault threshold, multi-source recorded waveforms and simulation data are extracted. Through denoising and anomaly rejection based on physical laws, sparse fault samples are obtained, and after enhancement and expansion, initial data is generated; After the initial data is formatted, it enters the multi-label modeling stage. The distribution imbalance is corrected through domain adaptation and adaptive weighted loss, and key features are extracted in parallel in the convolutional layer to generate a multi-label classification structure model; After preliminary training and entering the evaluation stage, the robustness of the model is tested for multiple-fold subsets and specific interference injection. The attention distribution of harmonic distortion and high-resistance grounding characteristics is verified by visualization means. If there are no weights conflicting with physical priors, the intermediate model parameters are output; When the real-time recorded waveform data meets the trigger threshold, fault identification reasoning is performed, and a high-order deviation metric function is used to monitor new or rare faults. If significant drift is detected, misjudged samples and labels are automatically uploaded for incremental training, and an adaptive diagnostic closed-loop is formed under the contrast memory mechanism.

2. An AC power grid fault current waveform recording method according to claim 1, characterized in that: Fault waveform data is obtained from multiple sources, and the obtained original data is preliminarily cleaned to obtain initial cleaned data; Based on the rare fault types or composite fault waveforms in the initial cleaned data, targeted data enhancement and high-order feature extraction are performed to generate an enhanced feature data set; For fault types with insufficient sample sizes, synthetic fault data with real waveform statistical characteristics is constructed, and part of the enhanced data obtained is integrated with the original data and incorporated into the enhanced feature data set.

3. An AC power grid fault current waveform recording method according to claim 2, characterized in that: After introducing a high-order waveform energy measure to identify the key time-frequency characteristics of composite faults, the waveform sampling points, labels, GAN enhancement identifiers, and high-order energy measures are stored as a hierarchical table; The enhanced feature data is randomly sampled according to the required ratio to obtain three subsets: training, validation, and testing; a consistency check is performed on the diversified samples to make the high-order waveform energy measure correspond one-to-one with the original waveform, generating preprocessed output data.

4. An AC power grid fault current waveform recording method according to claim 3, characterized in that: Define multiple parallel input branches to process time-domain sampling data and high-order feature values respectively; After parallel input is completed, a basic convolutional layer is introduced to extract local patterns, and a feature fusion layer is established to map the features extracted from each branch to the same hidden vector space to form a fused hidden vector; according to the set of target fault labels in the power grid, after retaining the number of channels matching the number of labels, a preliminary classification activation output is generated.

5. An AC power grid fault current waveform recording method according to claim 4, characterized in that: An adaptively updatable weight coefficient is incorporated into the loss function to define a multi-label weighted loss function; A cross-domain mapping function is introduced to align the fused hidden vectors, a cross-domain difference metric is defined, and the cross-domain difference metric is incorporated into the overall loss. After minimizing this value, the difference between the simulation and real data distributions is reduced.

6. A method for recording the waveform of the fault current in an AC power grid according to claim 5, characterized in that: Add monotonicity or nonlinear transition constraints to the activation function of the convolutional channel, which is achieved by imposing an additional regular constraint function on the weights of each convolutional kernel; Under the physical prior constraint, retrain or fine-tune the domain adaptation module and the multi-label adaptive weighted loss function, and output the initial multi-label model after several rounds of iteration, including the network structure and the initial weights.

7. A method for recording the waveform of the fault current in an AC power grid according to claim 6, characterized in that: Screen a subset with a wide coverage of fault label categories and diverse time-frequency characteristics from the training dataset and match it with the initial multi-label model for loading; for sample-scarce categories such as high-resistance faults, perform oversampling or increase the training frequency in the preprocessed output data of the original dataset, and cooperate with the adaptive weighted loss function to increase the attention to rare categories; Send the prepared training set and the initial multi-label model into the deep learning framework, perform several rounds of iteration according to the comprehensive loss function, obtain the initially optimized network weights, and generate the first intermediate training model.

8. A method for recording the waveform of the fault current in an AC power grid according to claim 7, characterized in that: The training set is divided into folds. One fold is used as the validation set one by one, and the remaining folds are used as the training set. A training-validation loop is performed on the intermediate model one. After each training, the multi-label classification results are recorded. Introduce additional noise simulation in the training or validation stage, perform amplitude perturbation or random high-frequency interference injection on the waveform, and introduce a robustness metric function: combine the cross-validation results and the robustness metric function, and further fine-tune and iterate the first intermediate training model until the verification indicators of each fold basically converge, and finally form more reliable model weights and structural configurations, and output the second intermediate training model.

9. A method for recording the waveform of the fault current in an AC power grid according to claim 8, characterized in that: By drawing the attention distribution map, display the high-order response correlation of the key intervals concerned by the model in the time domain or frequency domain of the waveform: check the change slope or change curve of the output label when the input fault waveform is fine-tuned on the key features, so as to judge whether the change slope or change curve of the output label strictly identifies around the physical prior. After the interpretability verification and physical completeness verification, obtain the interpretable training model.

10. A method for recording the waveform of the fault current in an AC power grid according to claim 8, characterized in that: Install a fault identification and inference module at the edge device of the nearest substation for storing historical fault records and subsequent evaluation; import the network structure and weights of the interpretable training model into the edge inference engine, and set the real-time sampling frequency, batch size and number of core threads according to the grid operation parameters; Continuously monitor the fault identification delay, accuracy and resource occupancy within a set time period, collect the multi-label output results of online identification, and compare them with known fault events, and record the real-time error list.

11. A method for recording the waveform of the fault current in an AC power grid according to claim 10, characterized in that: Compare the real-time recognition results reported by the edge side with the fault event records, calculate the accuracy and missed detection rate for each sub-fault type, and centrally store them in the incremental update model; if the defined high-order deviation metric function exceeds the deviation threshold or a new fault category is detected, package the corresponding data and labels and send them back to the cloud; For retraining or incremental training on new data, generate a new incremental update model; After training is completed, send the incremental update model back to the edge side and automatically replace the currently used fault recognition model to form a deployed update model.

12. A method for recording the fault current waveform of an AC power grid according to claim 11, characterized in that: During training, appropriately retain the key data and its feature distribution in the early stage, or use regularization constraints to suppress large deviations of network weights, introduce a contrastive memory constraint function, reduce catastrophic forgetting, and retain the important fault rules learned in the power system; When it is confirmed that the model recognition result is incorrect or a missed judgment occurs, archive this data and manually label it to form operation and maintenance labeled data, and periodically incorporate the operation and maintenance labeled data into the retraining set; after multiple rounds of updates, merged memories, and operation and maintenance feedback, integrate and obtain the deployed update model.

Images