CN-122020311-A - Driving circuit fault diagnosis method and system based on improved CAE network

CN122020311ACN 122020311 ACN122020311 ACN 122020311ACN-122020311-A

Abstract

The invention discloses a driving circuit fault diagnosis method and system based on an improved CAE network, which comprises the steps of obtaining state monitoring data of a driving circuit in a nuclear power station instrument control system, preprocessing the state monitoring data to obtain preprocessed state monitoring data, carrying out modal decomposition on the preprocessed state monitoring data based on an integrated empirical mode decomposition algorithm to obtain modal decomposed data, carrying out feature extraction on the modal decomposed data based on the improved CAE network to obtain fault feature data and representing the fault feature data as a high-order sparse form, training a pre-built depth classifier according to the fault feature data to obtain a trained depth classifier, wherein the depth classifier is a multi-class classifier which is constructed by using one flattening layer and three FC layers and is used for fault diagnosis, and carrying out fault diagnosis on the state monitoring data to be detected based on the trained depth classifier. The invention realizes accurate fault diagnosis of the driving circuit.

Inventors

Yuan yannan
MA YU
ZHANG JIAWEI
LI ANG
YANG CHENG
HAN WENXING
WANG FU
LI JUEYING
MIN YUAN
XIAHOU TANGFAN
LIU YU
YAO ZHANG

Assignees

中国核动力研究设计院

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (13)

1. The driving circuit fault diagnosis method based on the improved CAE network is characterized by comprising the following steps: Acquiring state monitoring data of a driving circuit in a nuclear power station instrument control system, and preprocessing the state monitoring data to obtain preprocessed state monitoring data, wherein the state monitoring data is three-phase current data of an output end of a switch reluctance motor in the driving circuit; Performing modal decomposition on the preprocessed state monitoring data based on an integrated empirical mode decomposition algorithm to obtain data after modal decomposition; Performing feature extraction on the data after modal decomposition based on an improved CAE network to obtain fault feature data and representing the fault feature data as a high-order sparse form; Training a pre-constructed depth classifier according to the fault characteristic data to obtain a trained depth classifier, wherein the depth classifier is a multi-class classifier which is constructed by using one flattening layer and three FC layers and is used for fault diagnosis; and performing fault diagnosis on the state monitoring data to be detected based on the trained depth classifier to obtain a fault diagnosis result.
2. The driving circuit fault diagnosis method based on the improved CAE network of claim 1, characterized in that the preprocessing comprises: Searching for the maximum value and the minimum value in each phase of current data according to the state monitoring data, and carrying out normalization processing on the state monitoring data based on a min-max normalization method to obtain normalized data; And converting the normalized data into a form and a file format recognizable by a subsequent algorithm.
3. The driving circuit fault diagnosis method based on the improved CAE network according to claim 1, wherein the improved CAE network is constructed and obtained by combining the CAE network and network sparsification characteristics, comprising: Constructing a CAE network, and extracting fault characteristics in data after modal decomposition based on the CAE network, wherein the CAE network is constructed by applying a one-dimensional convolutional neural network, batch normalization, reLu activation functions and ELU activation function layers and adopting an AE structure as the overall architecture of the network; And the network sparsification characteristic is that L2 regularization and relative entropy regularization are selected as sparsification constraint limits of the network, and the two regularization items are added into a loss function of the CAE network in the form of a loss function penalty item.
4. The improved CAE network based driving circuit fault diagnosis method of claim 1, characterized in that the outputs of the first two layers of FCs in the depth classifier are activated by means of a Tanh activation function; the last layer of FC is used as an output layer, the number of the included neurons is the same as the number of the fault diagnosis categories, and output data is activated by adopting a Softmax activation function and is used for classifying and judging the positions of faults in the driving circuit.
5. The method for diagnosing a driving circuit fault based on an improved CAE network as claimed in claim 4, wherein a Dropout layer is built between every two FC layers for preventing the depth classifier from over-fitting during the training process.
6. The method for diagnosing a drive circuit failure based on an improved CAE network of claim 5, wherein a Dropout rate of the Dropout layer is 0.5.
7. The method for driving circuit fault diagnosis based on improved CAE network as claimed in claim 3, wherein the loss function The expression of (2) is: ; ; ; ; ; ; ; Wherein, the Representing the first input CAE The data samples are taken from the data samples, Represent CAE output of the first The data samples are taken from the data samples, The total sample quantity is the total three-phase current data quantity; representing the operation of the two norms, For the number of network layers contained in the CAE, Represent the first The number of neurons in the layer network, Represent the first Sample pair number Neurons in layers Is used for the weight of the (c), Is the first Layer number The bias term of the individual neurons is set, The activation function is represented as a function of the activation, Represent the first The average activation of the individual neurons is determined, As a function of the KL-divergence, Representing a logarithmic function; The value of the sparsity parameter is 0.05; And The weight coefficients of the L 2 regularization term and the relative entropy regularization term are represented respectively, And The training learning rate is the same as that of training; And (3) with Respectively representing real and predicted labels, wherein the label formats are in a single-hot coding form, And (3) with Respectively is And (3) with Contains one-hot coding digit number which satisfies And for all And (3) with Satisfy the following requirements , For improved weighting values between the CAE network and the classifier loss function, Taken as 0.5.
8. A drive circuit fault diagnosis system based on an improved CAE network, the system comprising: the acquisition unit is used for acquiring state monitoring data of a driving circuit in the instrument control system of the nuclear power station, wherein the state monitoring data is three-phase current data of an output end of a switch reluctance motor in the driving circuit; the preprocessing unit is used for preprocessing the state monitoring data to obtain preprocessed state monitoring data; The modal decomposition unit is used for carrying out modal decomposition on the preprocessed state monitoring data based on an integrated empirical mode decomposition algorithm to obtain data after modal decomposition; the improved CAE network unit is used for carrying out feature extraction on the data after modal decomposition based on the improved CAE network to obtain fault feature data and representing the fault feature data as a high-order sparse form; The model training unit is used for training the pre-constructed depth classifier according to the fault characteristic data to obtain a trained depth classifier, wherein the depth classifier is a multi-class classifier which is constructed by using one flattening layer and three FC layers and is used for fault diagnosis; And the fault diagnosis unit is used for carrying out fault diagnosis on the state monitoring data to be detected based on the trained depth classifier to obtain a fault diagnosis result.
9. The improved CAE network based drive circuit fault diagnosis system of claim 8, wherein the improved CAE network is constructed in combination with CAE network and network sparseness characteristics, comprising: Constructing a CAE network, and extracting fault characteristics in data after modal decomposition based on the CAE network, wherein the CAE network is constructed by applying a one-dimensional convolutional neural network, batch normalization, reLu activation functions and ELU activation function layers and adopting an AE structure as the overall architecture of the network; And the network sparsification characteristic is that L2 regularization and relative entropy regularization are selected as sparsification constraint limits of the network, and the two regularization items are added into a loss function of the CAE network in the form of a loss function penalty item.
10. The improved CAE network based drive circuit fault diagnosis system of claim 8, wherein the outputs of the first two layers of FCs in the depth classifier are activated using a Tanh activation function; the last layer of FC is used as an output layer, the number of the included neurons is the same as the number of the fault diagnosis categories, and output data is activated by adopting a Softmax activation function and is used for classifying and judging the positions of faults in the driving circuit.
11. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the drive circuit failure diagnosis method based on an improved CAE network according to any of claims 1 to 7 when executing the computer program.
12. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the drive circuit failure diagnosis method based on the modified CAE network of any one of claims 1 to 7.
13. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the improved CAE network based drive circuit fault diagnosis method of any of claims 1 to 7.

Description

Driving circuit fault diagnosis method and system based on improved CAE network Technical Field The invention relates to the field of deep learning and fault diagnosis, in particular to a driving circuit fault diagnosis method and system based on an improved CAE network. Background The development and utilization of nuclear energy brings new energy, power and development opportunities for humans, but is also accompanied by risks and challenges of nuclear security. The rapid development of nuclear power technology is always limited by potential reliability and safety problems of a reactor when a nuclear power station is used as a carrier for the current development and utilization of nuclear energy of human beings. At present, although human beings can effectively control part of nuclear fission reaction, large-scale equipment such as mechanical equipment, power system and the like for performing nuclear reactor start-stop and power control can gradually degrade the performance until failure in the use process, so that accidents frequently occur, and the life and property safety of people is seriously threatened. Therefore, in order to ensure that the nuclear power plant maintains a good performance state during operation and prevent accidents, a failure determination study on the reactor and the nuclear power equipment thereof is required. The driving circuit is used as a key component of a nuclear power plant instrument control system (Instrument and Control System, ICS), is power equipment for providing electric energy and control signals for a reactor control rod driving mechanism, plays a crucial role in the operation and power regulation of the nuclear power plant, and the reliability level and the operation state of the driving circuit are directly related to the safe and stable operation of the nuclear power plant. However, the soft fault caused by the degradation of the parameters of the electronic components in the driving circuit has weak fault response conditions of the monitoring data compared with the open circuit fault of the insulated gate bipolar transistor (Insulated Gate Bipolar Transistor, IGBT), and if the two types of faults are directly mixed for diagnosis, the model is easy to misjudge the soft fault as a normal state. Therefore, the driving circuit electronic component parameter degradation fault response is weak, so that the mode confusion is caused, and the problems of inaccurate fault diagnosis and the like of the existing driving circuit are solved. Disclosure of Invention The invention aims to solve the technical problems that the mode confusion is caused by weak response of the degradation fault of the parameters of the electronic components of the driving circuit, so that the fault diagnosis of the existing driving circuit is inaccurate, and the like. The invention aims to provide a driving circuit fault diagnosis method and system based on an improved CAE network, which combines an integrated empirical mode decomposition (Ensemble Empirical Mode Decomposition, EEMD) algorithm with the improved CAE network to capture high-dimensional sparse fault characteristics, improve the characteristic mining and capturing capacity of a characteristic extractor, simultaneously avoid the challenge that tiny fault variation in monitoring signals cannot be captured by the network, introduce network parameter sparse characteristics to overcome the problem of low distinguishing property of the fault characteristics, thereby realizing accurate and robust driving circuit fault diagnosis and providing a solid foundation for effective development of subsequent fault positioning and the like. The invention is realized by the following technical scheme: In a first aspect, the present invention provides a driving circuit fault diagnosis method based on an improved CAE network, the method comprising: acquiring state monitoring data of a driving circuit in a nuclear power station instrument control system, and preprocessing the state monitoring data to obtain preprocessed state monitoring data, wherein the state monitoring data is three-phase current data of an output end of a switch reluctance motor in the driving circuit; Performing modal decomposition on the preprocessed state monitoring data based on an integrated empirical mode decomposition algorithm to obtain data after modal decomposition; performing feature extraction on the data after modal decomposition based on an improved CAE network to obtain fault feature data and representing the fault feature data as a high-order sparse form; Training a pre-constructed depth classifier according to fault characteristic data to obtain a trained depth classifier, wherein the depth classifier is a multi-class classifier which is constructed by using one flattening layer and three FC layers and is used for fault diagnosis; And performing fault diagnosis on the state monitoring data to be detected based on the trained depth classifier to obtain a fault diagn