Fault diagnosis method, device, equipment, storage medium and program product

Abstract

The application relates to a fault diagnosis method, device, equipment, storage medium and program product. The method comprises the following steps: acquiring power equipment operation data; performing feature extraction on the power equipment operation data according to a feature extraction model to obtain feature data, wherein the feature extraction model is realized by using a neural network model based on a skew correction mechanism; inputting the feature data into a preset classification model, and performing fault diagnosis on the power equipment according to an output result of the classification model. The method can improve the accuracy of fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, in particular to a fault diagnosis method, device, equipment, storage medium and program product. BACKGROUND

[0002] With the rapid development of modern power systems, the operating environment of power equipment is becoming increasingly complex, and its safety and stability are crucial to the reliability of the entire power grid. However, there are many types of power equipment failures, including electrical faults, mechanical faults, temperature abnormalities, etc., which have complex manifestations and the early signals of the faults are often hidden and weak.

[0003] In the prior art, fault diagnosis methods mainly rely on expert experience or physical model-based analysis, which are sensitive to data size and complexity, and the diagnosis results are not accurate enough to adapt to the diversified operating conditions of modern power equipment. SUMMARY

[0004] Therefore, it is necessary to provide a fault diagnosis method, device, equipment, storage medium and program product capable of improving the accuracy of fault diagnosis.

[0005] In a first aspect, the present application provides a fault diagnosis method, comprising:

[0006] obtaining power equipment operating data;

[0007] performing feature extraction on the power equipment operating data according to a feature extraction model to obtain feature data, the feature extraction model being implemented by a neural network model based on a skewness correction mechanism;

[0008] inputting the feature data into a preset classification model, and performing fault diagnosis on the power equipment according to the output result of the classification model.

[0009] In one embodiment, the training process of the feature extraction model comprises:

[0010] obtaining an initial feature extraction model, and obtaining a first training sample set, the first training sample set comprising a plurality of first training samples and the fault types corresponding to each first training sample, the first training sample being historical power equipment operating data; obtaining a loss function, and iteratively training the initial feature extraction model by a backpropagation and gradient descent method to obtain the feature extraction model, the loss function being related to the weighted coefficients of each first training sample and cross-entropy loss.

[0011] In one of the embodiments, for each first training sample, the weighting coefficient is determined according to a density of the first training sample under the corresponding fault type and a total density of all first training samples under the fault type; and the density of the first training sample is determined according to a similarity between the feature data of the first training sample and the feature data of other first training samples under the fault type.

[0012] In one of the embodiments, the initial feature extraction model comprises an input layer, a plurality of hidden layers and an output layer, and the loss function is also related to a similarity measure of the output features of each layer of the initial feature extraction model; for the lth layer of the initial feature extraction model, the similarity measure is determined according to a total number of the first training samples and a similarity between the output features of each first training sample at the lth layer.

[0013] In one of the embodiments, the classification model is implemented based on a deep belief network composed of a plurality of layers of restricted Boltzmann machines, and the training process of the classification model comprises:

[0014] obtaining a second training sample set, and pre-training each layer of the restricted Boltzmann machine using a contrastive divergence algorithm to determine an initial deep belief network, the second training sample set comprising a plurality of second training samples and corresponding fault types of each second training sample, and the second training sample being determined according to the feature extraction model and historical operation data of the power equipment; and performing supervised training on the initial deep belief network based on the quantum state of each second training sample to obtain the classification model.

[0015] In one of the embodiments, the supervised training on the initial deep belief network based on the quantum state of each second training sample to obtain the classification model comprises:

[0016] calculating the quantum state of each second training sample, and transforming each quantum state through a quantum gate operation to determine a target quantum state of each second training sample; performing quantum measurement on the target quantum state of each second training sample to output a target fault type of each second training sample; and performing iterative updating on the weight of the initial deep belief network through a back propagation algorithm based on each target fault type and the corresponding fault type of each second training sample in the second training sample set to obtain the classification model.

[0017] In a second aspect, the present application further provides a fault diagnosis device, comprising:

[0018] an obtaining module configured to obtain the operation data of the power equipment;

[0019] a feature extraction module configured to perform feature extraction on the operation data of the power equipment according to a feature extraction model to obtain feature data, and the feature extraction model is implemented by using a neural network model based on a skew correction mechanism;

[0020] The diagnostic module is configured to input the feature data into a preset classification model, and perform fault diagnosis on the power equipment according to an output result of the classification model.

[0021] In one of the embodiments, the fault diagnosis apparatus further comprises a first training module configured to obtain an initial feature extraction model, and obtain a first training sample set comprising a plurality of first training samples and corresponding fault types of the first training samples, the first training samples being historical power equipment operation data; obtain a loss function, and iteratively train the initial feature extraction model by a back propagation and a gradient descent method to obtain the feature extraction model, the loss function being related to a weighted coefficient of each first training sample and a cross-entropy loss.

[0022] In one of the embodiments, for each first training sample, the weighted coefficient is determined according to a density of the first training sample under the corresponding fault type and a total density of all first training samples under the fault type; and the density of the first training sample is determined according to a similarity between the feature data of the first training sample and feature data of other first training samples under the fault type.

[0023] In one of the embodiments, the initial feature extraction model comprises an input layer, a plurality of hidden layers and an output layer, and the loss function is further related to a similarity measure of output features of each layer of the initial feature extraction model; for the lth layer of the initial feature extraction model, the similarity measure is determined according to a total number of the first training samples and a similarity between output features of each first training sample at the lth layer.

[0024] In one of the embodiments, the classification model is implemented based on a deep belief network, the deep belief network is composed of a plurality of layers of restricted Boltzmann machines, and the fault diagnosis apparatus further comprises a second training module configured to obtain a second training sample set comprising a plurality of second training samples and corresponding fault types of the second training samples, and pre-train each layer of the restricted Boltzmann machines by using a contrastive divergence algorithm to determine an initial deep belief network, the second training samples being determined according to the feature extraction model and the historical power equipment operation data; and perform supervised training on the initial deep belief network based on quantum states of the second training samples to obtain the classification model.

[0025] In one of the embodiments, the second training module is specifically configured to calculate quantum states of the second training samples, and transform each quantum state by a quantum gate operation to determine a target quantum state of each second training sample; perform quantum measurement on the target quantum state of each second training sample to output a target fault type of each second training sample; and iteratively update weights of the initial deep belief network by a back propagation algorithm based on the target fault type and the corresponding fault type of each second training sample in the second training sample set to obtain the classification model.

[0026] In a third aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method of any one of the first aspect when executing the computer program.

[0027] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method of any one of the first aspect.

[0028] In a fifth aspect, the present application also provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the method of any one of the first aspect.

[0029] The above fault diagnosis method, device, equipment, storage medium and program product, obtain power equipment operation data; perform feature extraction on the power equipment operation data according to a feature extraction model to obtain feature data, the feature extraction model is implemented by using a neural network model based on a skew correction mechanism; input the feature data into a preset classification model, and perform fault diagnosis on the power equipment according to an output result of the classification model. Using the skew correction mechanism can process the training data, correct or adjust the skew of the training data, so that the feature extraction model can better learn the features of various samples. In this way, the feature extraction model can still accurately capture potential fault patterns in the case of less abnormal data, and the combination of the feature data and the classification model can accurately diagnose rare faults and improve the accuracy of fault diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0031] Figure 1 A flowchart of a fault diagnosis method in an embodiment;

[0032] Figure 2 A flowchart of a training step of a feature extraction model in an embodiment;

[0033] Figure 3 A flowchart of a training step of a classification model in an embodiment;

[0034] Figure 4 A flowchart of a training step of a classification model in another embodiment;

[0035] Figure 5 A system architecture block diagram corresponding to the fault diagnosis method in an embodiment;

[0036] Figure 6 A structural block diagram of a fault diagnosis device in an embodiment;

[0037] Figure 7 A structural block diagram of a fault diagnosis device in another embodiment;

[0038] Figure 8 An internal structural diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0039] In order to make the purposes, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0040] With the rapid development of modern power systems, the operating environment of power equipment is becoming increasingly complex, and its safety and stability are crucial to the reliability of the entire power grid. However, there are various types of power equipment failures, including electrical faults, mechanical faults, temperature abnormalities, etc., which have complex manifestations and early signals of the faults are often hidden and weak.

[0041] In the prior art, fault diagnosis methods mainly rely on expert experience or physical model-based analysis, which are sensitive to data size and complexity, and the diagnosis results are not accurate enough to adapt to the diversified operating conditions of modern power equipment.

[0042] Therefore, the present application provides a fault diagnosis method that can improve the accuracy of fault diagnosis. The fault diagnosis method provided in the embodiments of the present application can have a fault diagnosis device as its execution subject, which can be realized by software, hardware or a combination of software and hardware. It can be embedded in or independent of the processor in the computer device in hardware form, or it can be stored in the memory in the computer device in software form. In the following method embodiments, the execution subject is taken as an example of a computer device, which can be a server, and the specific type of the computer device is not limited in the embodiments of the present application.

[0043] In an exemplary embodiment, as shown in Figure 1 a fault diagnosis method is provided, which includes the following steps 101 to 103. Among them:

[0044] Step 101, obtaining power equipment operating data.

[0045] Optionally, the power equipment operation data can include voltage, current, temperature, vibration, sound, pressure or load data of the power equipment during operation.

[0046] Optionally, the power equipment operation data can be obtained from the database according to the identification information of the power equipment, and can also be real-time operation data of the power equipment obtained through an operation monitoring system of the power equipment. The operation monitoring system of the power equipment can be a distributed sensor network, and different types of sensors can be arranged at different parts of the power equipment, for example, a current sensor is used to monitor current fluctuation, a temperature sensor is used to monitor overheating of the equipment, and a vibration sensor is used to detect mechanical damage of the equipment.

[0047] Optionally, during operation of the power equipment, the distributed sensor network can be used to sample at a preset time interval, continuously collect power equipment operation data, and send the collected power equipment operation data to the computer device in real time through the Internet of Things technology.

[0048] For example, the data attributes of the power equipment operation data include Ra as voltage (unit: V), Da as current (unit: A), Ta as temperature (unit: °C), Va as vibration (unit: m / s²), Sa as sound (unit: dB), Pa as pressure (unit: Pa), La as load (unit: kW), Ca as equipment operation state (0 represents normal, 1 represents abnormal), Na as operator operation parameter (such as start-up time, shutdown time, etc.), and Fa as equipment failure history (0 represents no failure, 1 represents failure). It should be noted that this embodiment is only used to illustrate a data format and type of the present application, and in actual application, the number of data attributes is usually more than 10, and the number of data attributes can reach dozens or even hundreds.

[0049] In step 102, the power equipment operation data is subjected to feature extraction according to a feature extraction model to obtain feature data, and the feature extraction model is implemented by using a neural network model based on a skewness correction mechanism.

[0050] Optionally, before the power equipment operation data is subjected to feature extraction, the obtained power equipment operation data can be preprocessed, including data cleaning, denoising, normalization and format conversion, to ensure data quality.

[0051] Optionally, in the feature extraction model training process, the training samples in the training sample set often exist in the condition of skew distribution, that is, the data quantity of some classes is much more than that of other classes, or the distribution of some feature values of the data presents serious imbalance. Such skewness will lead to insufficient learning of the neural network to the tail information of the minority class or feature distribution in the training process, thereby affecting the comprehensive and accurate extraction of the model to the data features. The skewness correction mechanism can be used to correct or adjust the skewness of the training samples, so that the feature extraction model can more effectively process various data and features and improve the quality of feature extraction.

[0052] In a possible implementation manner, the skewness correction mechanism is used to correct or adjust the skewness of the training samples, which can be performed at a data preprocessing level, and resampling operations including oversampling and undersampling are performed on the training samples. The oversampling can be increasing the number of minority class training samples, for example, new training samples can be generated by interpolating the minority class training samples or the like, so as to balance the training sample set. The undersampling can be reducing the number of majority class training samples, so as to achieve the purpose of relatively balanced data distribution.

[0053] In another possible implementation manner, the skewness correction mechanism is used to correct or adjust the skewness of the training samples, which can be performed at a neural network structure level. A neural network with a multi-branch structure is used, and each branch is responsible for processing different parts or feature subsets of the data. In this way, the skewness problem of the data can be more flexibly coped with. Different branches can learn different scales and types of features, and then the outputs of the branches are fused to obtain more comprehensive and robust feature representations.

[0054] In another possible implementation manner, the skewness correction mechanism is used to correct or adjust the skewness of the training samples, which can also be performed at a loss function level. Based on the focal loss function, a modulation factor is introduced on the basis of the cross-entropy loss function, so as to automatically reduce the attention degree to the samples that are easy to classify, and focus more on the samples that are difficult to classify. In this way, the neural network can better capture subtle features and differences in the data in the training process.

[0055] In step 103, the feature data is input into a preset classification model, and the power equipment is diagnosed for faults according to an output result of the classification model.

[0056] Optionally, the output of the classification model can include multiple categories, for example, category 1 represents a normal operating state, i.e., the power equipment is in a normal operating state without any abnormality; category 2 represents an electrical fault, such as voltage, current abnormality or short circuit, etc.; category 3 represents a mechanical fault, such as excessive vibration, noise or mechanical component damage, etc.; category 4 represents a temperature abnormality, such as excessive equipment temperature, poor heat dissipation, etc.; category 5 represents a load overload, such as the equipment load exceeding its rated carrying capacity, etc.; and category 6 represents a comprehensive fault, i.e., multiple attributes (such as voltage, current, temperature, etc.) of the equipment occur abnormality at the same time.

[0057] Optionally, the classification model can be a support vector machine, a decision tree, a multilayer perception, a random forest or a gradient boosting tree, etc. The embodiments of the present application do not limit this.

[0058] The above fault diagnosis method obtains power equipment operating data; performs feature extraction on the power equipment operating data according to a feature extraction model to obtain feature data, the feature extraction model being implemented by a neural network model based on a skew correction mechanism; inputs the feature data into a preset classification model, and performs fault diagnosis on the power equipment according to the output result of the classification model. Using the skew correction mechanism can process the training data, correct or adjust the skew of the training data, so that the feature extraction model can better learn the features of various samples. In this way, the feature extraction model can still accurately capture potential fault patterns in the case of less abnormal data, and in combination with the feature data and the classification model, it can accurately diagnose rare faults and improve the accuracy of fault diagnosis.

[0059] In an exemplary embodiment, as shown in Figure 2 Optionally, the training process of the feature extraction model includes the following steps 201 to 202. Wherein:

[0060] Step 201, obtaining an initial feature extraction model and obtaining a first training sample set.

[0061] The first training sample set includes multiple first training samples and the fault types corresponding to each first training sample, and the first training sample is historical power equipment operating data.

[0062] Optionally, the initial feature extraction model can include an input layer, multiple hidden layers and an output layer, each layer including a corresponding number of nodes and an activation function, and the activation function of each layer can be expressed as:

[0063]

[0064] Wherein, is the activation output of the lth layer of the neural network; is the input power equipment operating data; is a weight of the i-th layer of the neural network; is a bias of the i-th layer of the neural network; is an activation function of the neural network, optionally, the activation function of the neural network can adopt a LeakyReLU activation function, and can also adopt a ReLU activation function, which is not limited by the embodiments of the present application.

[0065] Optionally, the fault type corresponding to the first training sample can be determined in an artificial labeling manner.

[0066] In step 202, an initial feature extraction model is obtained by iteratively training the initial feature extraction model through a back propagation and gradient descent method, and a feature extraction model is obtained. The loss function is related to the weighted coefficient of each first training sample and the cross-entropy loss.

[0067] Optionally, the value of the loss function can be used as the target of the initial feature extraction model training, and the gradient of the loss function with respect to the initial feature extraction model parameters is calculated through the back propagation and gradient descent method, so as to guide the updating direction of the initial feature extraction model parameters, and make the initial feature extraction model optimize in the direction of reducing the value of the loss function. The parameters of the initial feature extraction model are continuously adjusted to obtain the feature extraction model.

[0068] Optionally, the gradient of the loss function with respect to the initial feature extraction model parameters is calculated through the back propagation and gradient descent method, so as to guide the updating direction of the initial feature extraction model parameters, which can specifically include: according to the derivative of the loss function with respect to the output layer, the error of the output layer neuron is calculated; then, starting from the output layer, the error is sequentially propagated backward to each layer using the chain rule, and the error of each layer neuron and the gradient of the loss function with respect to each layer parameter (weight and bias) are calculated; then, according to the calculated gradient, the learning rate is multiplied to obtain the updating amount of the parameter, wherein the learning rate determines the step size of each parameter update, and is an important hyperparameter; then, the current parameter value is subtracted by the updating amount to obtain a new parameter value, and one parameter update is completed, which can be represented by the following formula:

[0069]

[0070]

[0071] In the formula, w is a weight of the neural network, b is a bias of the neural network, x is an input of the neural network, y is an output of the neural network, and f is an activation function of the neural network. is a parameter updating operation; is a learning rate of the neural network; is a partial derivative symbol; is a loss function; for example, is set to 0.01.

[0072] Optionally, the above parameter update process is repeated to iteratively train the initial feature extraction model. The iterative stopping conditions may include: when the loss function value converges to a smaller value, or reaches a preset maximum number of iterations, or the performance on the validation set no longer improves, etc., the iterative training is stopped to obtain a trained feature extraction model.

[0073] Optionally, the loss function is used to measure the difference between the fault type predicted by the model and the fault type corresponding to the first training sample. In an embodiment of the present application, after feature extraction of the i-th first training sample, the feature data obtained can be calculated by a preset Softmax function to obtain the predicted fault type.

[0074] In one possible implementation, the loss function can be expressed as follows:

[0075]

[0076] in, is the cross entropy loss of the neural network, For the The fault type corresponding to the first training sample, For the The predicted fault type of the first training sample, is the total number of the first training samples, is the weighting coefficient of the i-th first training sample.

[0077] The following describes the process of determining the weighting coefficients of each first training sample:

[0078] Optionally, for each first training sample, the weighting coefficient is determined based on the density of the first training sample under the corresponding fault type and the total density of all first training samples under the fault type; the density of the first training sample is determined based on the similarity between the characteristic data of the first training sample and the characteristic data of other first training samples under the fault type.

[0079] Optionally, the weighting coefficient of the first training sample is calculated as follows:

[0080]

[0081] Where, is the density of the i-th first training sample under the fault type to which it belongs; is the total density of all first training samples of this fault type; is the density of the kth first training sample under the fault type to which it belongs; is the number of samples of the current fault type; is the weighting coefficient adjustment constant, for example, may be set to 0.3.

[0082] Optionally, the density of the first training sample under the fault type to which the first training sample belongs can be determined by a kernel density estimation method. Considering the distribution difference between the first training samples, in order to dynamically measure the local distribution of each first training sample, the density of the first training sample under the fault type to which the first training sample belongs can also be realized according to the similarity between the feature data of the first training sample, and the calculation manner is represented as:

[0083]

[0084] wherein, is the feature data obtained by inputting the i-th first training sample into the initial feature extraction model; is the feature data obtained by inputting the i-th first training sample into the initial feature extraction model; is the feature data obtained by inputting the i-th first training sample into the initial feature extraction model; is the feature data obtained by inputting the i-th first training sample into the initial feature extraction model; is the L2 norm; is the scale parameter of the Gaussian kernel, and optionally, may be set to 0.3.

[0085] Optionally, based on the category distribution of each first training sample, higher weighting coefficients are given to category samples that are scarce or difficult to learn, solving the problem of insufficient learning of neural networks for a small number of categories caused by uneven distribution of power equipment operation data, and the attention of power equipment operation data feature extraction to complex samples can be improved.

[0086] In another possible implementation manner, the initial feature extraction model includes an input layer, a plurality of hidden layers and an output layer, and the loss function is also related to the similarity measurement of the output features of each layer of the initial feature extraction model; for the l-th layer of the initial feature extraction model, the similarity measurement is determined according to the total number of the first training samples and the similarity between the outputs of each first training sample at the l-th layer.

[0087] Optionally, the loss function can be represented as:

[0088]

[0089] wherein, may represent the total loss including the weighted loss and the similarity loss; is a hyperparameter for adjusting the overall weight of the similarity loss, and exemplarily, may be set to 0.3; is the weight coefficient of the l-th layer similarity loss, and the weight coefficients of the similarity losses of the layers are preset; is the total number of layers of the initial feature extraction model; is the similarity measurement of the output features of the l-th layer of the initial feature extraction model.

[0090] Optionally, the similarity measure of the output feature of the initial feature extraction model at the lth layer The similarity measure can be represented by the following formula:

[0091] S p (l) = 1 n p ( n p -1) ∑ i≠j exp(- ∥ Z p (l) [i]- Z p (l) [j] ∥ 2 2 σ p (l)2 )

[0092] wherein, Z p (l) [i] is the output feature of the neural network at the lth layer for the mth first training sample; is the output feature of the neural network at the lth layer for the mth first training sample; Z p (l) [j] is the output feature of the neural network at the lth layer for the mth sample; is a scale parameter of the similarity measure of the output feature of the neural network at the lth layer.

[0093] Optionally, the sensitivity of the similarity measure can be adaptively adjusted according to the change of the distribution of the output feature of each layer, and the adaptive scale parameter calculation can be realized according to the variance of the hierarchical output, and the calculation manner is represented as:

[0094] σ p (l) = α p ⋅ 1 n p ∑ i=1 n p  ∥ Z p (l) [i]- μ p (l) ∥ 2

[0095] wherein, is the mean vector of the output feature of the neural network at the lth layer; is a coefficient for adjusting the sensitivity of the similarity, which can be optionally set to 0.3.

[0096] Optionally, the loss function can combine the skewness correction and the similarity measure, so that the feature extraction model can learn the local spatial relationship of the power equipment operation data while taking into account the uneven distribution of the power equipment operation data, thereby enhancing the generalization ability and stability of the feature extraction model.

[0097] In an exemplary embodiment, as shown in Figure 3 the classification model is implemented based on a deep belief network composed of multiple layers of restricted Boltzmann machines, and the training process of the classification model includes the following steps 301 to step 302. Wherein:

[0098] ​​Step 301, a second training sample set is obtained, and each layer of the restricted Boltzmann machine is pre-trained using a contrast divergence algorithm to determine an initial deep belief network.

[0099] The second training sample set includes a plurality of second training samples and corresponding fault types of each second training sample, and the second training sample is determined according to a feature extraction model and historical power equipment operation data.

[0100] Optionally, the second training sample can be input into the trained feature extraction model, and the output feature data is taken as the second training sample.

[0101] Optionally, the deep belief network (DBN) is a representative deep generative model, which is a deep neural network stacked by a plurality of restricted Boltzmann machines (RBM), each layer of the DBN is an RBM, adjacent RBM layers interact through connection weights, the output of the lower layer RBM is taken as the input of the upper layer RBM, and each RBM includes visible units and hidden units.

[0102] Optionally, each restricted Boltzmann machine in the deep belief network can be pre-trained layer by layer through unsupervised learning, and the weights and biases of each layer of the restricted Boltzmann machine are optimized by contrast divergence, and the weight update increment The calculation method of the weight update increment

[0103]

[0104] In the formula, is the weight from the visible unit to the hidden unit is a learning rate; is an expected value based on the feature data after feature extraction; is an expected value based on the feature data after feature extraction after reconstruction.

[0105] Optionally, and The calculation method of the weight update increment

[0106]

[0107]

[0108] In the formula, is a Sigmoid function, is a value of a reconstructed visible unit, ​is the number of second training samples, and may be the value of the reconstructed visual unit obtained by Gibbs sampling, and the Gibbs sampling can be sampling the second training samples based on a conditional probability distribution.

[0109] Optionally, the initial deep belief network can be determined according to the pre-trained restricted Boltzmann machines, and for example, the initial deep belief network can be formed by stacking the restricted Boltzmann machines.

[0110] At step 302, the initial deep belief network is supervised trained based on the quantum states of the second training samples, to obtain a classification model.

[0111] Optionally, the quantum state of the second training sample can be mapping the second training sample to a quantum state of a quantum system through a specific encoding manner, and the quantum state can be a superposition state of a quantum bit, an entangled state or other many-body quantum states.

[0112] Optionally, quantum state feedback can be used to adjust and optimize the parameters of the initial deep belief network, for example, the quantum measurement result of the quantum state of the second training sample is used as feedback information to update the parameters of the initial deep belief network, so that the model can converge faster and the risk of overfitting can be reduced.

[0113] Optionally, as shown in Figure 4 , the initial deep belief network is supervised trained based on the quantum states of the second training samples, to obtain a classification model, including steps 401 to 403. Wherein:

[0114] At step 401, the quantum state of each second training sample is calculated, and the quantum state of each second training sample is transformed through quantum gate operation to determine the target quantum state of each second training sample.

[0115] Optionally, the quantum state of each second training sample can be an initial quantum bit.

[0116] In one possible implementation, the quantum state of each second training sample can be determined according to an amplitude encoding manner, and after each second training sample is normalized to a unit vector, the amplitude of the quantum state can be directly determined.

[0117] In another possible implementation, the quantum state of each second training sample can be determined according to the following formula:

[0118]

[0119] Wherein, represents the quantum state of the i-th second training sample; and​ respectively are the polar angle and the azimuthal angle on the Bloch sphere of the qubit, the polar angle and the azimuthal angle are determined by the parameterization through the i-th second training sample; and are the basic states of the qubit, define as the basic state of the first qubit, as the basic state of the second qubit.

[0120] Optionally, the polar angle and the azimuthal angle are calculated based on the second training sample, which can be represented as:

[0121]

[0122]

[0123] wherein, indicates normalizing the value of the second training sample to the range [0, 1]:

[0124]

[0125] wherein, and are the minimum value and the maximum value in the second training sample set respectively.

[0126] Optionally, the transformation of each quantum state through the quantum gate operation can be performed in the following way:

[0127]

[0128] wherein, is the target quantum state of the quantum state of the i-th second training sample after transformation through the quantum gate operation; is the quantum gate operation, which can transform the quantum state according to the optimization parameters and wherein, is the first optimization angle parameter, is the second optimization angle parameter.

[0129] Optionally, the quantum gate operation can be a rotation gate, which can be represented as:

[0130]

[0131] Step 402, performing quantum measurement on the target quantum state of each second training sample, and outputting the target fault type of each second training sample.

[0132] Optionally, quantum measurement can be to collapse the target quantum state to a specific ground state according to a preset quantum measurement method, and return the corresponding measurement result, which is a definite classical value.

[0133] Optionally, the quantum measurement result is achieved based on the probability distribution of the target quantity state, and the category corresponding to the measurement result is the target fault type of each second training sample.

[0134] Optionally, the quantum measurement method may include a projection measurement method, a weak measurement method, a quantum tomography method, etc. The quantum measurement method is not limited in the embodiments of the present application.

[0135] Step 403 : Based on each target fault type and the fault type corresponding to each second training sample in the second training sample set, the weights of the initial deep belief network are iteratively updated by a back propagation algorithm to obtain a classification model.

[0136] Optionally, when the weights of the initial deep belief network are iteratively updated through the back-propagation algorithm, the calculation method of the weight update increment can be expressed as:

[0137]

[0138] in, is the learning rate; is an error function. Exemplarily, the error function adopts a cross entropy error function, which can be determined according to the target fault type and the fault type corresponding to the second training sample in the second training sample set. Determined by quantum state feedback.

[0139] Optionally, based on quantum state feedback, the update increment of the weights in the initial deep belief network The calculation method can be expressed as:

[0140]

[0141] in, It is The quantum measurement results of the second training sample; is the average value of all second training sample quantum measurement results; is the feedback strength parameter; is the learning rate, exemplarily, It can be set to 0.3, Can be set to 0.01.

[0142] Optionally, the above steps 401 to 403 may be iteratively performed to update the weights of the initial deep belief network.

[0143] Optionally, in one iteration, an angle parameter in a quantum gate operation can also be included in the optimization according to an error function E and The updating manner of the first and second optimized angle parameters is represented as:

[0144]

[0145]

[0146] wherein, is a learning rate for updating the optimized angle parameter, and an exemplary value is 0.01.

[0147] The above-mentioned quantum state feedback mechanism in the deep belief network reduces the risk of overfitting and improves the training efficiency of the model by representing data features with quantum bits and dynamically adjusting the learning rate and parameter updating strategy in the classification stage.

[0148] In an exemplary embodiment, the trained feature extraction model and classification model can be used to analyze newly collected power equipment operation data in real time, quickly determine whether there are early signs of equipment operation failure, and determine the possible failure development trend by comparing with historical data, thereby providing a basis for equipment maintenance decisions.

[0149] Optionally, the running state of the equipment, the fault diagnosis result, and the historical trend analysis can also be displayed through a graphical interface to enable the user to view detailed diagnostic reports, which can include fault location, severity, and recommended maintenance measures.

[0150] Optionally, as shown in Figure 5 the system architecture diagram of the above-mentioned fault diagnosis method includes a data acquisition and sensing unit, a fault diagnosis unit, and a visualization and decision support unit, wherein the fault diagnosis unit includes a data preprocessing and feature extraction unit, a diagnosis unit, and a model training unit.

[0151] It should be understood that although each step in the flowchart involved in each embodiment described above is shown in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0152] Based on the same inventive concept, the embodiments of the present application also provide a fault diagnosis device for implementing the above-mentioned fault diagnosis method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more fault diagnosis device embodiments provided below can refer to the limitations of the fault diagnosis method described above, which will not be described here again.

[0153] In one exemplary embodiment, as shown in Figure 6 a fault diagnosis device 600 is provided, comprising an acquisition module 601, a feature extraction module 602, and a diagnosis module 603, wherein:

[0154] The acquisition module 601 is configured to acquire power equipment operation data.

[0155] The feature extraction module 602 is configured to perform feature extraction on the power equipment operation data according to a feature extraction model to obtain feature data, and the feature extraction model is implemented by using a neural network model based on a skew correction mechanism.

[0156] The diagnosis module 603 is configured to input the feature data into a preset classification model, and perform fault diagnosis on the power equipment according to an output result of the classification model.

[0157] In one embodiment, as shown in Figure 7 the fault diagnosis device 600 further comprises a first training module 604, the first training module 604 is configured to acquire an initial feature extraction model, and acquire a first training sample set, the first training sample set comprises a plurality of first training samples and a fault type corresponding to each first training sample, and the first training sample is historical power equipment operation data; acquire a loss function, and perform iterative training on the initial feature extraction model by using a back propagation and gradient descent method to obtain the feature extraction model, and the loss function is related to a weighted coefficient of each first training sample and a cross-entropy loss.

[0158] In one embodiment, for each first training sample, the weighted coefficient is determined according to a density of the first training sample under the corresponding fault type and a total density of all first training samples under the fault type; and the density of the first training sample is determined according to a similarity between the feature data of the first training sample and the output features of other first training samples under the fault type.

[0159] In one embodiment, the initial feature extraction model comprises an input layer, a plurality of hidden layers, and an output layer, and the loss function is also related to a similarity measure of the feature data of each layer of the initial feature extraction model; for the lth layer of the initial feature extraction model, the similarity measure is determined according to a total number of first training samples and a similarity between the feature data of each first training sample at the lth layer.

[0160] In one embodiment, the classification model is implemented based on a deep belief network, which is composed of multiple layers of restricted Boltzmann machines, as shown in Figure 7 As shown, the fault diagnosis apparatus 600 further comprises a second training module 605, configured to obtain a second training sample set, and pre-train each layer of the restricted Boltzmann machine using a contrastive divergence algorithm to determine an initial deep belief network, the second training sample set comprising a plurality of second training samples and a fault type corresponding to each second training sample, the second training sample being determined according to a feature extraction model and historical operation data of the power equipment; and the initial deep belief network is supervised trained based on a quantum state of each second training sample to obtain the classification model.

[0161] In one embodiment, the second training module 605 is specifically configured to calculate a quantum state of each second training sample, and transform each quantum state through a quantum gate operation to determine a target quantum state of each second training sample; perform quantum measurement on the target quantum state of each second training sample to output a target fault type of each second training sample; and update the weight of the initial deep belief network through a back propagation algorithm based on each target fault type and the fault type corresponding to each second training sample in the second training sample set to obtain the classification model.

[0162] Each module in the above fault diagnosis apparatus can be realized wholly or partially by software, hardware and combinations thereof. Each module described above can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to each module.

[0163] In one exemplary embodiment, a computer device is provided, which can be a server, and an internal structure diagram thereof can be as shown in Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a fault diagnosis method is implemented.

[0164] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0165] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps described in any of the above method embodiments when executing the computer program.

[0166] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps described in any of the above method embodiments are implemented.

[0167] In one embodiment, a computer program product is provided, comprising a computer program, which implements the steps of any of the above method embodiments when executed by a processor.

[0168] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0169] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0170] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A fault diagnosis method, characterized in that: The method comprises: Obtaining power equipment operation data; Extracting features from the power equipment operation data according to a feature extraction model to obtain feature data, wherein the feature extraction model is implemented using a neural network model based on a skew correction mechanism; Inputting the characteristic data into a preset classification model, and performing fault diagnosis on the power equipment according to an output result of the classification model; The classification model is implemented based on a deep belief network, which is composed of a multi-layer restricted Boltzmann machine. The training process of the classification model includes: Obtaining a second training sample set and pre-training each layer of the restricted Boltzmann machine using a contrastive divergence algorithm to determine an initial deep belief network, wherein the second training sample set includes a plurality of second training samples and a fault type corresponding to each second training sample, and the second training sample is determined based on the feature extraction model and historical power equipment operation data; Performing supervised training on the initial deep belief network based on the quantum state of each of the second training samples to obtain the classification model; The supervised training of the initial deep belief network based on the quantum state of each second training sample to obtain the classification model includes: Calculating the quantum state of each second training sample and transforming each quantum state through a quantum gate operation to determine a target quantum state of each second training sample; performing quantum measurement on a target quantum state of each of the second training samples, and outputting a target fault type of each of the second training samples; Based on each of the target fault types and the fault type corresponding to each of the second training samples in the second training sample set, the weights of the initial deep belief network are iteratively updated through a back-propagation algorithm to obtain a classification model.

2. The method according to claim 1, characterized in that The training process of the feature extraction model includes: Obtaining an initial feature extraction model and obtaining a first training sample set, wherein the first training sample set includes a plurality of first training samples and a fault type corresponding to each first training sample, and the first training samples are historical power equipment operation data; A loss function is obtained, and the initial feature extraction model is iteratively trained by back propagation and gradient descent method to obtain a feature extraction model, wherein the loss function is related to the weighted coefficient and cross entropy loss of each of the first training samples.

3. The method according to claim 2, characterized in that For each first training sample, the weighting coefficient is determined based on the density of the first training sample under the corresponding fault type and the total density of all first training samples under the fault type; the density of the first training sample is determined based on the similarity between the feature data of the first training sample and the feature data of other first training samples under the fault type.

4. The method according to claim 2, characterized in that The initial feature extraction model includes an input layer, multiple hidden layers and an output layer, and the loss function is also related to the similarity measure of the output features of each layer of the initial feature extraction model; l layer, the similarity metric is based on the total number of the first training samples and the number of the first training samples in the first l The similarity between the output features of the layers is determined.

5. A fault diagnosis device, characterized in that: The device comprises: An acquisition module is used to obtain the operating data of the power equipment; a feature extraction module, configured to extract features from the power equipment operation data according to a feature extraction model to obtain feature data, wherein the feature extraction model is implemented using a neural network model based on a skew correction mechanism; a diagnosis module, configured to input the characteristic data into a preset classification model and perform fault diagnosis on the power equipment according to an output result of the classification model; The classification model is implemented based on a deep belief network, which is composed of multiple layers of restricted Boltzmann machines. A second training module is used to obtain a second training sample set and pre-train the restricted Boltzmann machines of each layer using a contrastive divergence algorithm to determine an initial deep belief network, wherein the second training sample set includes multiple second training samples and a fault type corresponding to each second training sample, and the second training samples are determined based on the feature extraction model and historical power equipment operation data; the initial deep belief network is supervisedly trained based on the quantum state of each second training sample to obtain the classification model; The second training module is specifically used to calculate the quantum state of each second training sample, and transform each quantum state through quantum gate operations to determine the target quantum state of each second training sample; perform quantum measurement on the target quantum state of each second training sample, and output the target fault type of each second training sample; based on each target fault type and the fault type corresponding to each second training sample in the second training sample set, iteratively update the weights of the initial deep belief network through a backpropagation algorithm to obtain a classification model.

6. The device according to claim 5, characterized in that The fault diagnosis device further comprises: The first training module is used to obtain an initial feature extraction model and a first training sample set, where the first training sample set includes multiple first training samples and a fault type corresponding to each first training sample, and the first training sample is historical power equipment operation data; a loss function is obtained, and the initial feature extraction model is iteratively trained through back propagation and gradient descent methods to obtain a feature extraction model, where the loss function is related to the weighted coefficient and cross entropy loss of each first training sample.

7. The device according to claim 6, characterized in that For each first training sample, the weighting coefficient is determined based on the density of the first training sample under the corresponding fault type and the total density of all first training samples under the fault type; the density of the first training sample is determined based on the similarity between the feature data of the first training sample and the feature data of other first training samples under the fault type.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images