Electrical equipment fault diagnosis method and device based on artificial intelligence

Abstract

The invention provides an electrical equipment fault diagnosis method and device based on artificial intelligence, and relates to the technical field of fault recognition, and the method comprises the steps: obtaining historical operation data collected by a sensor installed in a power system, and enabling the historical operation data to serve as an original data set; expanding the original data set; inputting the expanded original data set into a feature extraction model to train the feature extraction model; training a feature dimension reduction model, and training a classifier model; and inputting real-time operation data collected by a sensor installed in the power system into the trained feature extraction model, the feature dimension reduction model and the classifier model in sequence, and outputting corresponding fault categories. The diversity of a data set is enhanced through a data expansion technology, the generalization ability of a model is improved, feature extraction, feature dimension reduction and classification processing are carried out, deep analysis and processing of data are achieved, and therefore the operation state and the possible fault type of the electrical equipment can be judged more accurately.

Description

Technical Field

[0001] The present application relates to the technical field of fault identification, and in particular to an electrical equipment fault diagnosis method and device based on artificial intelligence. Background Art

[0002] As the complexity of power systems continues to increase, traditional fault diagnosis methods can no longer meet the requirements of modern power systems in terms of accuracy and efficiency. Especially in terms of data acquisition, processing and analysis, facing large and complex data sets, how to effectively predict faults and manage health has become a technical challenge that needs to be solved urgently.

[0003] In the existing electrical equipment fault diagnosis technology, data expansion often uses simple replication or linear interpolation methods, which cannot effectively reflect the complex data structure. As a result, the expanded data cannot effectively support model generalization, the fault diagnosis accuracy is low, and the safe operation of electrical equipment is affected. Summary of the invention

[0004] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0005] To this end, the first purpose of this application is to propose an electrical equipment fault diagnosis method based on artificial intelligence.

[0006] The second object of the present application is to provide a device.

[0007] The third objective of the present application is to provide an electronic device.

[0008] A fourth objective of the present application is to provide a computer-readable storage medium.

[0009] A fifth object of the present application is to provide a computer program product.

[0010] To achieve the above-mentioned purpose, the first embodiment of the present application proposes an electrical equipment fault diagnosis method based on artificial intelligence, comprising:

[0011] Acquire historical operation data collected by sensors installed in the power system as an original data set, and mark the fault category of each sample in the original data set;

[0012] Expanding the original data set;

[0013] Inputting the expanded original data set into a feature extraction model to train the feature extraction model, and outputting a first training feature;

[0014] Inputting the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model, and outputting a second training feature;

[0015] Inputting the second training feature into a classifier to train a classifier model;

[0016] The real-time operation data collected by the sensors installed in the power system are sequentially input into the trained feature extraction model, feature dimension reduction model and classifier model, and the corresponding fault category is output.

[0017] Optionally, the expanding the original data set includes:

[0018] Performing correlation analysis on the training data to determine highly correlated feature combinations;

[0019] Assume that the original data set collected is The number of samples in the original data set is N ds , the Pearson correlation coefficient is calculated for each feature dimension to evaluate the strength of the linear relationship between the features, expressed as:

[0020]

[0021] In the formula, and are the values ​​of the i-th sample on the j-th and k-th dimension features respectively; and Represent the values ​​of the sample on the j-th and k-th dimension features respectively; and are the means of the features of the samples in the jth and kth dimensions respectively; is the correlation coefficient of the sample values ​​on the j-th and k-th dimension features;

[0022] For each minority class sample, the local density is calculated based on the distribution of its k nearest neighbors to determine which minority class samples have relatively sparse sample spaces around them and require more data interpolation. The calculation method of the local density is expressed as:

[0023]

[0024] In the formula, yes The k nearest neighbor sample set; is the local density of the i-th sample; represents the i-th sample; represents the jth sample; σ c is the width parameter of the Gaussian kernel; ∥∥ is the L2 norm.

[0025] Based on the feature correlation and local density of each sample, a sparse feature map is constructed, and new samples are generated based on the local similarity and the sparse feature map;

[0026] The generated new samples are optimized through feature scaling and outlier processing, and new samples are continuously generated until the number of new samples reaches a preset threshold.

[0027] Optionally, the step of inputting the expanded original data set into a feature extraction model to train the feature extraction model, and outputting a first training feature comprises:

[0028] Generate an initial population, each individual represents a configuration of network weights, where the population size is N p , initialize the weights and biases for the i-th individual, expressed as:

[0029]

[0030] Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of

[0031] For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance based on the predetermined loss function;

[0032] According to the fitness of the individuals, individuals with fitness higher than the threshold are selected from the current population and retained as candidate solutions for the next generation;

[0033] Generate new offspring through crossover and mutation operations;

[0034] Repeatedly generate new offspring until the preset stop iteration condition is met;

[0035] The original data set is input into the trained feature extraction model to extract features to output the first training features.

[0036] Optionally, the loss function is:

[0037]

[0038] Where, L pi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; represents the composite loss function, and fsig() represents the neural network model function; Represents the characteristics of the jth sample; represents the label of the jth sample;

[0039] The loss function contains a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as:

[0040]

[0041] Where MSE() is the mean square error function, Reg(W pi ) is the regularization term, λ ps is the regularization parameter.

[0042] Optionally, the calculation formula of the fitness is:

[0043]

[0044] Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual.

[0045] Generate new offspring through crossover and mutation operations, including:

[0046] W′ pi =α pcs W p1 +(1-α pcs )W p2

[0047] b′ pi =α pcs b p1 +(1-α pcs )b p2

[0048] In the formula, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first individual selected, b p1 is the bias of the neural network corresponding to the first individual selected, W p2 is the weight of the neural network corresponding to the second individual selected, b p2 is the bias of the neural network corresponding to the second individual selected, W′ pi is the weight of the neural network corresponding to the individual after the crossover operation, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation;

[0049] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0050]

[0051] In the formula, τ 2 represents the variance of variation, W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b″ pi is the bias of the neural network corresponding to the individual after the mutation operation.

[0052] Optionally, inputting the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model includes:

[0053] Inputting the first training feature into the autoencoder neural network, mapping it to the initial low-dimensional feature space through a multi-layer nonlinear mapping structure, and outputting the initial low-dimensional feature;

[0054] Generate feature weights according to feature importance in the current feature space, and assign the feature weights to the initial low-dimensional features;

[0055] Adjust the weights of each initial low-dimensional feature to generate optimized features;

[0056] Re-mapping the optimized features back to the high-dimensional space and calculating the loss function;

[0057] Optimizing the weights of the features for multiple rounds until an iteration stop condition is met;

[0058] The first training feature is input into the trained feature dimensionality reduction model for dimensionality reduction to output the second training feature.

[0059] Optionally, the step of inputting the second training feature into a classifier to train a classifier model includes:

[0060] Initialize the quantum state, and represent the second training feature after dimension reduction of each feature as a quantum bit;

[0061] Through the quantum state adaptive conversion mechanism, each quantum bit is operated;

[0062] Map the expanded quantum features onto a Riemannian manifold and use the local and global geometric structures of the manifold to process the data;

[0063] Train the support vector machine model on the data processed by quantum states and Riemann manifolds, and use the improved feature space to make the final classification decision to ensure that the model can effectively distinguish different types of electrical faults;

[0064] Minimizing the overall classification error is the goal of adjusting the first quantum gate parameter and the second quantum gate parameter.

[0065] To achieve the above-mentioned purpose, the second embodiment of the present application proposes an electrical equipment fault diagnosis device based on artificial intelligence, comprising:

[0066] A data acquisition module, used to obtain historical operation data collected by sensors installed in the power system as an original data set, and mark the fault category of each sample in the original data set;

[0067] An expansion module, used to expand the original data set;

[0068] A first training module, used for inputting the expanded original data set into a feature extraction model to train the feature extraction model, and outputting a first training feature;

[0069] A second training module, used for inputting the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model, and outputting a second training feature;

[0070] A third training module, used for inputting the second training feature into a classifier to train a classifier model;

[0071] The fault diagnosis module is used to input the real-time operation data collected by the sensors installed in the power system into the trained feature extraction model, feature dimension reduction model and classifier model in sequence, and output the corresponding fault category.

[0072] To achieve the above-mentioned purpose, the third aspect of the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0073] The memory stores computer-executable instructions;

[0074] The processor executes the computer-executable instructions stored in the memory to implement the method as described in any one of the first aspects.

[0075] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present application proposes a computer-readable storage medium, in which computer-readable storage medium is stored computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement the method as described in any one of the first aspects.

[0076] To achieve the above-mentioned purpose, the fifth aspect of the present application proposes a computer program product, which implements any method in the first aspect when executed by a processor.

[0077] The electrical equipment fault diagnosis method, device, electronic device and storage medium based on artificial intelligence provided by the present application have the following beneficial effects:

[0078] The local similarity-based SMOTE algorithm is used for data expansion. New samples are dynamically generated by evaluating the correlation and sparsity of local features to solve the problem of insufficient training data.

[0079] The feature extraction model is trained using a neural network optimization algorithm based on dynamic population evolution, and the evolution rules of network weights and biases are adjusted to solve the problem of insufficient generalization ability of traditional training methods when faced with variable data characteristics.

[0080] The feature dimension reduction is performed using an autoencoding neural network based on feature refinement, and the features are finely managed through a dynamically adjusted weight matrix to address the problems of feature redundancy and information loss in the original technology.

[0081] By utilizing a quantum state-based support vector machine algorithm and using quantum state super-position state simulation to enhance data feature expression, we address the limitations of traditional support vector machines in processing complex data patterns.

[0082] The newly generated data samples maintain high-quality attributes similar to the original data, effectively increasing data diversity and improving the training effect of the model.

[0083] Through the application of self-correction mechanism, the flexibility and adaptability of the algorithm are improved, thereby improving the model's ability to handle new and complex data.

[0084] The optimized dimensionality reduction model reduces the data dimension while ensuring the retention of important information and improving the efficiency of data processing.

[0085] The improved support vector machine classification algorithm proposed in the present invention enhances the model's ability to recognize data patterns and improves classification accuracy and efficiency.

[0086] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0088] Figure 1 A flowchart of an electrical equipment fault diagnosis method based on artificial intelligence provided in an embodiment of the present application;

[0089] Figure 2 A training flowchart for a neural network algorithm based on dynamic population evolution optimization;

[0090] Figure 3 It is a training flow chart of the autoencoder neural network algorithm based on feature refinement;

[0091] Figure 4 A schematic diagram of the structure of an electrical equipment fault diagnosis device based on artificial intelligence provided in an embodiment of the present application. DETAILED DESCRIPTION

[0092] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0093] As the complexity of power systems continues to increase, traditional fault diagnosis methods can no longer meet the requirements of modern power systems in terms of accuracy and efficiency. Especially in terms of data acquisition, processing and analysis, facing large and complex data sets, how to effectively predict faults and manage health has become a technical challenge that needs to be solved urgently.

[0094] In the existing electrical equipment fault diagnosis technology, data expansion often uses simple replication or linear interpolation methods, which cannot effectively reflect the complex data structure, resulting in the expanded data being unable to effectively support model generalization.

[0095] Traditional feature extraction methods often rely on static network structures and parameter configurations. When faced with changes in data characteristics, model adjustments are not flexible enough, which affects the processing effect.

[0096] In terms of feature dimensionality reduction, conventional methods may lead to the loss of important information, making the reduced data insufficient to support accurate subsequent analysis and decision-making.

[0097] Traditional classifiers have limited effectiveness when processing highly complex or nonlinearly separable data, especially in application scenarios such as electrical fault diagnosis, where they are unable to effectively identify complex fault modes.

[0098] To address this problem, the present invention provides an electrical equipment fault diagnosis method based on artificial intelligence. Figure 1 The following is a flow chart of an electrical equipment fault diagnosis method based on artificial intelligence provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0099] Step 101, obtaining historical operation data collected by sensors installed in the power system as an original data set, and marking the fault category of each sample in the original data set;

[0100] Step 102, expanding the original data set;

[0101] Step 103, inputting the expanded original data set into a feature extraction model to train the feature extraction model, and outputting a first training feature;

[0102] Step 104, inputting the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model, and outputting a second training feature;

[0103] Step 105, inputting the second training feature into a classifier to train a classifier model;

[0104] Step 106, input the real-time operation data collected by the sensors installed in the power system into the trained feature extraction model, feature dimension reduction model and classifier model in sequence, and output the corresponding fault category.

[0105] In this embodiment, data collection mainly comes from sensors installed in the power system. These sensors can monitor and record key parameters such as current, voltage, temperature, etc. of electrical equipment during operation. The data is stored in a structured format in the database of the cloud server, and the specific storage format is a CSV file.

[0106] In one embodiment, the attributes of the data include:

[0107] Ia is the current value, Va is the voltage value, Ta is the temperature value, Ha is the humidity value, Fa is the frequency value, Pa is the power value, Ca is the capacitance value, La is the inductance value, Sa is the status code, and Na is the noise level.

[0108] It should be noted that this embodiment is only used to illustrate one data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds.

[0109] Furthermore, the collected data is labeled. The labeling method of the present invention is manual labeling. In one embodiment, the labeling categories include: normal operation, functional failure, and serious failure.

[0110] In this embodiment, five of the data are shown in Table 1:

[0111]

[0112]

[0113] Table 1

[0114] The step 102 expands the original data set, including:

[0115] Performing correlation analysis on the training data to determine highly correlated feature combinations;

[0116] It is understandable that in the task of the present invention, the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. The present invention uses a SMOTE algorithm driven by local similarity to generate samples, thereby achieving data expansion. On the basis of the traditional SMOTE algorithm, the present invention adopts a SMOTE algorithm driven by local similarity and adopts a sparse feature mapping mechanism. Based on the correlation and sparsity of local features, the feature value of each newly generated sample is dynamically adjusted, so that the newly generated sample is not just a simple linear combination of the original sample, but is optimized and generated according to the inherent structure of the feature, so as to better meet the needs of electrical equipment data expansion.

[0117] Assume that the original data set collected is The number of samples in the original data set is N ds , the Pearson correlation coefficient is calculated for each feature dimension to evaluate the strength of the linear relationship between the features, expressed as:

[0118]

[0119] In the formula, and are the values ​​of the i-th sample on the j-th and k-th dimension features respectively; and Represent the values ​​of the sample on the j-th and k-th dimension features respectively; and are the means of the features of the samples in the jth and kth dimensions respectively; is the correlation coefficient of the sample values ​​on the j-th and k-th dimension features;

[0120] For each minority class sample, the local density is calculated based on the distribution of its k nearest neighbors to determine which minority class samples have relatively sparse sample spaces around them and require more data interpolation. The calculation method of the local density is expressed as:

[0121]

[0122] In the formula, yes The k nearest neighbor sample set; is the local density of the i-th sample; represents the i-th sample; represents the jth sample; σ c is the width parameter of the Gaussian kernel; ∥∥ is the L2 norm.

[0123] In one embodiment, the Gaussian kernel width parameter σ c The calculation of depends on the standard deviation between samples, expressed as:

[0124]

[0125] Based on the feature correlation and local density of each sample, a sparse feature map is constructed, and new samples are generated based on the local similarity and the sparse feature map;

[0126] The sparse feature map not only reflects the interaction between features, but also considers the distribution characteristics of each feature in the local area. The calculation method of the sparse feature map matrix is ​​expressed as:

[0127]

[0128] Where M c is the sparse feature map matrix; is the element in the jth row and kth column of the sparse feature map matrix; α c is a coefficient that weighs relevance and sparsity, ∈ c is a small constant added to avoid the denominator being zero; is the local density of the ith sample.

[0129] In one embodiment, the coefficient for weighing relevance and sparsity is obtained by solving an optimization problem, and the calculation method is expressed as:

[0130]

[0131] Where d is the dimension of the feature. Preferably, ∈ c Set to 0.001.

[0132] When generating new samples, not only interpolation is performed numerically, but also feature values ​​are adjusted through mapping to ensure the rationality and diversity of new samples in the feature space. The method of generating new samples based on local similarity and sparse feature mapping is expressed as:

[0133]

[0134] In the formula, is the newly generated sample; and is the selected minority class sample pair, specifically the i-th sample and the j-th sample; λ c is an interpolation coefficient randomly sampled from the interval [0, 1], used to control the distance between the new sample and the original sample; is the corresponding element in the sparse feature map matrix, which is used to adjust the interpolation direction and amplitude.

[0135] The generated new samples are optimized through feature scaling and outlier processing, and new samples are continuously generated until the number of new samples reaches a preset threshold.

[0136] For the generated new samples Optimize by scaling features and handling outliers to ensure that their quality is consistent with the original data set, while optimizing their contribution to the prediction model. The calculation method is expressed as:

[0137]

[0138] In the formula, is the new sample after optimization; γ c and β c are the scaling factor and offset respectively.

[0139] In one embodiment, the scaling factor and the offset are used to adjust the scale and position of the new sample to ensure that the feature value of the new sample is within a reasonable range. The calculation method is expressed as:

[0140]

[0141] The above process is repeated until the number of generated samples reaches a preset threshold number. In one embodiment, assuming that the original collected samples are 800 and the data expansion obtains 200 samples, the expanded data set contains 1000 samples.

[0142] The expanded data is input into the feature extraction model to train the feature extraction model. The present invention uses a 6-layer fully connected neural network for feature extraction. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, gradient vanishing, gradient explosion or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a neural network algorithm based on dynamic population evolution optimization as a feature extraction model. In traditional population evolution algorithms, the evolution of all individuals is based on fixed rules. The present invention uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current training data, and adjusts the probability of crossover and mutation according to the changing trend of the loss function, so that the algorithm can adapt to different data distributions more flexibly and improve the generalization ability and training efficiency of the model.

[0143] The training flow chart of the neural network algorithm based on dynamic population evolution optimization is as follows Figure 2 As shown, the step 103 inputs the expanded original data set into the feature extraction model to train the feature extraction model, and outputs the first training feature including:

[0144] Generate an initial population, each individual represents a configuration of network weights, where the population size is N p , initialize the weights and biases for the i-th individual, expressed as:

[0145]

[0146] Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of

[0147] For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to the predetermined loss function; for the i-th individual, use its weight and bias to calculate the loss on the training data set, and the loss function is:

[0148]

[0149] Where, L pi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; represents the composite loss function, and fsig() represents the neural network model function; Represents the characteristics of the jth sample; represents the label of the jth sample;

[0150] The loss function contains a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as:

[0151]

[0152] Where MSE() is the mean square error function, Reg(W pi ) is the regularization term, λ ps is the regularization parameter.

[0153] In one embodiment, the composite loss function includes a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as:

[0154]

[0155] Where MSE() is the mean square error function, Reg(W pi ) is the regularization term, λ ps is the regularization parameter.

[0156] Furthermore, the calculation method of the regularization term is expressed as:

[0157]

[0158] Where W pi,k is the kth weight of the neural network corresponding to the i-th individual.

[0159] According to the fitness of the individuals, individuals with fitness higher than the threshold are selected from the current population and retained as candidate solutions for the next generation;

[0160] According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of an individual being selected is expressed as:

[0161]

[0162] Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual. Preferably, γ pse Set to 2.

[0163] Generate new offspring through crossover and mutation operations;

[0164] The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0165] W′ pi =α pcs W p1 +(1-α pcs )W p2

[0166] b′ pi =α pcs b p1 +(1-α pcs )b p2

[0167] In the formula, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first individual selected, b p1 is the bias of the neural network corresponding to the first individual selected, W p2 is the weight of the neural network corresponding to the second individual selected, b p2 is the bias of the neural network corresponding to the second individual selected, W′ pi is the weight of the neural network corresponding to the individual after the crossover operation, b′ piis the bias of the neural network corresponding to the individual after the crossover operation. Preferably, α pcs Set to 0.3.

[0168] Moreover, the mutation operation performs a small random perturbation on the newly generated individual weights, which can be expressed as:

[0169]

[0170] In the formula, τ 2 represents the variance of variation, W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b p ″ i is the bias of the neural network corresponding to the individual after the mutation operation. Preferably, τ 2 Set to 0.04.

[0171] Repeatedly generate new offspring until the preset stop iteration condition is met;

[0172] In this embodiment, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0173] The original data set is input into the trained feature extraction model to extract features to output the first training features.

[0174] The data after feature extraction is input into the feature dimensionality reduction model to train the feature dimensionality reduction model. The present invention adopts an autoencoding neural network based on feature refinement as a dimensionality reduction model. The autoencoding neural network based on feature refinement consists of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the features after dimensionality reduction back to the original space to ensure the reversibility of the dimensionality reduction process. The feature adjustment module dynamically adjusts the feature space after dimensionality reduction through recursive feature adaptive optimization, so that important features can be strengthened and secondary features are gradually weakened, so that the feature representation after dimensionality reduction has better simplicity while retaining data information.

[0175] Specifically, the training flow chart of the autoencoder neural network algorithm based on feature refinement is as follows: Figure 3 As shown, step 104 inputs the first training feature into the feature dimensionality reduction model to train the feature dimensionality reduction model, including:

[0176] Inputting the first training feature into the autoencoder neural network, mapping it to the initial low-dimensional feature space through a multi-layer nonlinear mapping structure, and outputting the initial low-dimensional feature;

[0177] In this embodiment, the low-dimensional feature space is expressed as:

[0178] Z r =Sig enc (W r X r +b r )

[0179] In the formula, Z r represents the initial low-dimensional features, W r is the weight matrix of the encoder, b r is the encoder bias vector, Sig enc () is the multi-layer Sigmoid activation function of the encoder.

[0180] Generate feature weights according to feature importance in the current feature space, and assign the feature weights to the initial low-dimensional features;

[0181] In this embodiment, after the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weight is assigned to all features so that the subsequent steps can be gradually adjusted according to the feature contribution. The calculation method of the initial weight matrix is ​​expressed as follows:

[0182] A r =diag(α r )

[0183] In the formula, A r is the initial weight matrix; α r is the feature weight vector, α r Each element α in r,i Initialized to the same value, indicating that all features have the same importance in the initial stage; diag() is a function for extracting the diagonal elements of the matrix.

[0184] Furthermore, the adjusted features can be expressed as:

[0185] Z′ r =A r Z r

[0186] In the formula, Z′ r is the feature representation after feature weight adjustment.

[0187] Adjust the weights of each initial low-dimensional feature to generate optimized features;

[0188] In this embodiment, the feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the features in the previous round, gradually enhances those features with important influences, and gradually weakens redundant or noise features. Assume that in the tth round of iteration, the weight update rule is as follows:

[0189]

[0190] In the formula, represents the feature weight of the t+1th iteration, represents the feature weight of the tth iteration, η r is the learning rate of the autoencoder neural network, L r () is the loss function of the autoencoder neural network, Y r is the label data, represents the gradient of the loss function with respect to the feature weight. Preferably, the loss function of the autoencoder neural network is a reconstruction error loss function. Preferably, η r Set to 0.01.

[0191] Re-mapping the optimized features back to the high-dimensional space and calculating the loss function;

[0192] In this embodiment, the reconstruction process of the decoder is expressed as:

[0193] X′ r =Sig dec (W′ r Z′ r +b′ r )

[0194] In the formula, X′ r is the reconstructed high-dimensional data, W′ r is the weight matrix of the decoder, b′ r is the bias vector of the decoder, Sig dec () is the multi-layer Sigmoid activation function of the decoder.

[0195] Optimizing the weights of the features for multiple rounds until an iteration stop condition is met;

[0196] In this embodiment, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0197] The first training feature is input into the trained feature dimensionality reduction model for dimensionality reduction to output the second training feature.

[0198] The step 105 inputs the second training feature into a classifier to train a classifier model, including:

[0199] Initialize the quantum state, and represent the second training feature after dimension reduction of each feature as a quantum bit;

[0200] In this embodiment, the initialization method is expressed as:

[0201]

[0202] In the formula, ψ u Represents the quantum state of the data sample after feature dimensionality reduction; α i,u is the complex amplitude, indicating the state |φ i,u >Probability amplitude; |φ i,u > is the ground state, that is, the basic state in quantum computing; Nu represents the number of quantum bits, which is the same as the characteristic dimension of the reduced data.

[0203] Through the quantum state adaptive conversion mechanism, each quantum bit is operated;

[0204] In this embodiment, each qubit is operated through a quantum state adaptive conversion mechanism to simulate the state superposition and entanglement phenomena in quantum computing, thereby enriching the expressiveness of data features and improving the sensitivity of the algorithm to complex data, which is expressed as:

[0205] ψ′ u =U(θ u ,γ u )ψ u

[0206] In the formula, ψ′ u is the new quantum state after applying quantum gate U; U(θ u ,γ u ) is a parameterized quantum gate, and θ u and γ u It is a parameter dynamically adjusted according to the data, which is used to control the conversion process of the quantum state. u is the first quantum gate parameter, γ u is the second quantum gate parameter.

[0207] In one embodiment, the quantum gate U(θ u ,γ u ) dynamically adjusts its parameters according to the data, and the calculation method is expressed as:

[0208]

[0209] Where, X u and Z u are the Pauli-X and Pauli-Z operators in quantum computing respectively; let θu ·X u +γ u ·Z u is the matrix M, represents the unitary transformation described by the matrix M; i * Is an imaginary unit.

[0210] Map the expanded quantum features onto a Riemannian manifold and use the local and global geometric structures of the manifold to process the data;

[0211] In this embodiment, the local and global geometric structures of the manifold are used to further process the data to reduce the dimension of the feature space and optimize the computational efficiency and classification performance of the algorithm. Specifically, the process of mapping quantum features to the Riemann manifold is expressed as:

[0212] χ u =f Riem (ψ′ u )

[0213] In the formula, χ u is the eigenvector after mapping, f Riem () is the mapping function from quantum state space to Riemann manifold.

[0214] In one embodiment, the mapping function from the quantum state space to the Riemann manifold uses the Riemannian geometry method to optimize the geometric structure of the data to meet the input requirements of the support vector machine. The calculation method is expressed as:

[0215]

[0216] Where Exp() is the exponential mapping on the manifold, π(u) represents the base point of the data point u on the manifold, and ω k The weights are automatically adjusted according to the data to ensure the adaptability and flexibility of the mapping.

[0217] Train the support vector machine model on the data processed by quantum states and Riemann manifolds, and use the improved feature space to make the final classification decision to ensure that the model can effectively distinguish different types of electrical faults;

[0218] In this embodiment, the data on the Riemann manifold is used to train the support vector machine model, and the optimization problem is solved as follows:

[0219]

[0220] The constraints are: ξ i,u ≥0

[0221] In the formula, w u is the normal vector of the classification hyperplane; b uis the bias term of the support vector machine; ξ i,u is a slack variable used to handle nonlinear separability; y i,u is the label of the i-th sample; C is the regularization parameter, which controls the trade-off between error and model complexity; Mu is the number of samples input in the current batch.

[0222] Minimizing the overall classification error is the goal of adjusting the first quantum gate parameter and the second quantum gate parameter.

[0223] In this embodiment, the optimization objective function can be expressed as:

[0224]

[0225] Where, L SVM It is the loss function of the support vector machine, which calculates the loss based on the difference between the actual classification results and the theoretical results.

[0226] In one embodiment, the loss function of the support vector machine is calculated as follows:

[0227] L SVM (y i,u ,x i,u (θ u ,γ u ))=max(0,1-y i,u ·( <w u ,x i,u (θ u ,γ u )>+b u )) δ

[0228] Wherein, δ is an adjustable parameter for controlling the sensitivity and smoothness of the loss function. Preferably, δ is set to 2.

[0229] After the above model training is completed, the trained model is used to diagnose electrical equipment faults. In one embodiment, the collected raw data is input into the trained feature extraction and feature dimension reduction model for feature processing, and further, the processed features are input into the classifier model for classification, thereby obtaining classification results. In this embodiment, the classification categories include: normal operation, functional failure, and serious failure.

[0230] In order to implement the above embodiments, the present application also proposes an electrical equipment fault diagnosis device based on artificial intelligence. Figure 4 The present invention provides a schematic diagram of an artificial intelligence-based electrical equipment fault diagnosis device. Figure 4 As shown, the device comprises:

[0231] The data acquisition module 410 is used to obtain historical operation data collected by sensors installed in the power system as an original data set, and mark the fault category of each sample in the original data set;

[0232] An expansion module 420, configured to expand the original data set;

[0233] A first training module 430, configured to input the expanded original data set into a feature extraction model to train the feature extraction model, and output a first training feature;

[0234] A second training module 440 is used to input the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model, and output a second training feature;

[0235] A third training module 450, configured to input the second training feature into a classifier to train a classifier model;

[0236] The fault diagnosis module 460 is used to input the real-time operation data collected by the sensors installed in the power system into the trained feature extraction model, feature dimension reduction model and classifier model in sequence, and output the corresponding fault category.

[0237] In order to implement the above embodiments, the present application also proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method provided by the above embodiments.

[0238] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0239] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0240] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0241] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0242] The present application is expected to provide an implementation scheme for users to selectively block the use or access of personal information data. That is, the present disclosure is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, the risk can be minimized by limiting data collection and deleting the data. In addition, when applicable, such personal information is de-identified to protect the privacy of the user.

[0243] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0244] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0245] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0246] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0247] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0248] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0249] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0250] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An electrical equipment fault diagnosis method based on artificial intelligence, characterized in that: The following steps are involved: Acquire historical operation data collected by sensors installed in the power system as an original data set, and mark the fault category of each sample in the original data set; Expanding the original data set; Inputting the expanded original data set into a feature extraction model to train the feature extraction model, and outputting a first training feature; Inputting the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model, and outputting a second training feature; Inputting the second training feature into a classifier to train a classifier model; The real-time operation data collected by the sensors installed in the power system are sequentially input into the trained feature extraction model, feature dimension reduction model and classifier model, and the corresponding fault category is output.

2. The method according to claim 1, characterized in that The expanding the original data set includes: Performing correlation analysis on the training data to determine highly correlated feature combinations; Assume that the original data set collected is The number of samples in the original data set is N ds , the Pearson correlation coefficient is calculated for each feature dimension, expressed as: In the formula, and are the values ​​of the i-th sample on the j-th and k-th dimension features respectively; and Represent the values ​​of the sample on the j-th and k-th dimension features respectively; and are the means of the features of the samples in the jth and kth dimensions respectively; is the correlation coefficient of the sample values ​​on the j-th and k-th dimension features; For each minority class sample, the local density is calculated based on the distribution of its k nearest neighbors to determine which minority class samples have relatively sparse sample spaces around them and require more data interpolation. The calculation method of the local density is expressed as: In the formula, yes The k nearest neighbor sample set; is the local density of the i-th sample; represents the i-th sample; represents the jth sample; σ c is the width parameter of the Gaussian kernel; ∥∥ is the L2 norm. Based on the feature correlation and local density of each sample, a sparse feature map is constructed, and new samples are generated based on the local similarity and the sparse feature map; The generated new samples are optimized through feature scaling and outlier processing, and new samples are continuously generated until the number of new samples reaches a preset threshold.

3. The method according to claim 2, characterized in that The step of inputting the expanded original data set into a feature extraction model to train the feature extraction model, and outputting a first training feature comprises: Generate an initial population, each individual represents a configuration of network weights, where the population size is N p , initialize the weights and biases for the i-th individual, expressed as: Where W pi is the weight of the neural network corresponding to the i-th individual, b pi is the bias of the neural network corresponding to the i-th individual, Represents the weight matrix of the i-th individual in the initial state; represents the bias of the i-th individual in the initial state; σ 2 represents the initialization variance; Indicates that the mean is 0 and the variance is σ 2 Normal distribution of For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the model's output, and evaluate its performance based on a predetermined loss function; According to the fitness of the individuals, individuals with fitness higher than the threshold are selected from the current population and retained as candidate solutions for the next generation; Generate new offspring through crossover and mutation operations; Repeatedly generate new offspring until the preset stop iteration condition is met; The original data set is input into the trained feature extraction model to extract features to output the first training features.

4. The method according to claim 3, characterized in that The loss function is: Where, L pi is the loss of the neural network corresponding to the i-th individual; mps represents the number of samples input in the current batch; l() represents the composite loss function, and fsig() represents the neural network model function; Represents the characteristics of the jth sample; represents the label of the jth sample; The loss function contains a regularization term, which can increase the generalization ability of the model. The calculation method is expressed as: Where MSE() is the mean square error function, Reg(W pi ) is the regularization term, λ ps is the regularization parameter.

5. The method according to claim 4, characterized in that The calculation formula of the fitness is: Where P select (i) represents the probability of the i-th individual being selected; γ pse is the parameter that controls the selection pressure; L pk is the loss of the neural network corresponding to the kth individual.

6. The method according to claim 5, characterized in that The generation of new offspring through crossover and mutation operations includes: W′ pi =a pcp W p1 +(1-a pcs )W p2 b′ pi =a pcs b p1 +(1-a pcs )b p2 In the formula, α pcs is the crossover rate, W p1 is the weight of the neural network corresponding to the first individual selected, b p1 is the bias of the neural network corresponding to the first individual selected, W p2 is the weight of the neural network corresponding to the second individual selected, b p2 is the bias of the neural network corresponding to the second individual selected, W′ pi is the weight of the neural network corresponding to the individual after the crossover operation, b′ pi is the bias of the neural network corresponding to the individual after the crossover operation; The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as: In the formula, τ 2 represents the variance of variation, W″ pi is the weight of the neural network corresponding to the individual after the mutation operation, b″ pi is the bias of the neural network corresponding to the individual after the mutation operation.

7. The method according to claim 6, characterized in that The step of inputting the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model includes: Inputting the first training feature into the autoencoder neural network, mapping it to the initial low-dimensional feature space through a multi-layer nonlinear mapping structure, and outputting the initial low-dimensional feature; Generate feature weights according to feature importance in the current feature space, and assign the feature weights to the initial low-dimensional features; Adjust the weights of each initial low-dimensional feature to generate optimized features; Re-mapping the optimized features back to the high-dimensional space and calculating the loss function; Optimizing the weights of the features for multiple rounds until an iteration stop condition is met; The first training feature is input into the trained feature dimensionality reduction model for dimensionality reduction to output the second training feature.

8. The method according to claim 7, characterized in that The step of inputting the second training feature into a classifier to train a classifier model includes: Initialize the quantum state, and represent the second training feature after dimension reduction of each feature as a quantum bit; Through the quantum state adaptive conversion mechanism, each quantum bit is operated; Map the expanded quantum features onto a Riemannian manifold and use the local and global geometric structures of the manifold to process the data; Train the support vector machine model on the data processed by quantum states and Riemann manifolds, and use the improved feature space to make the final classification decision; Minimizing the overall classification error is the goal of adjusting the first quantum gate parameter and the second quantum gate parameter.

9. An electrical equipment fault diagnosis device based on artificial intelligence, characterized in that: include: A data acquisition module, used to obtain historical operation data collected by sensors installed in the power system as an original data set, and mark the fault category of each sample in the original data set; An expansion module, used to expand the original data set; A first training module, used for inputting the expanded original data set into a feature extraction model to train the feature extraction model, and outputting a first training feature; A second training module, used for inputting the first training feature into a feature dimensionality reduction model to train the feature dimensionality reduction model, and outputting a second training feature; A third training module, used for inputting the second training feature into a classifier to train a classifier model; The fault diagnosis module is used to input the real-time operation data collected by the sensors installed in the power system into the trained feature extraction model, feature dimension reduction model and classifier model in sequence, and output the corresponding fault category.

10. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 8.

Images