Power Equipment Fault Detection Method and System Based on Deep Learning Network

Abstract

The present invention provides a power equipment fault detection method and system based on a deep learning network. The method includes: obtaining state parameter data of a power equipment during an observation period, forming a first multi-dimensional time series according to the state parameter data, inputting the first multi-dimensional time series into a deep learning model to obtain a fault probability prediction value corresponding to each time node; obtaining a fault coefficient according to the fault probability prediction values of all time nodes, and judging whether the power equipment has a fault according to the fault coefficient; if the power equipment has a fault, inputting the fault probability prediction value corresponding to each time node into a fault classification model to obtain a fault category set corresponding to the observation period, and obtaining the fault category of the current time node according to the fault category set. The present invention can avoid the situation that inaccurate diagnosis results are caused by data anomalies due to fault judgment through data of a single dimension, and improves the accuracy of fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and particularly to a method and system for fault detection of power equipment based on a deep learning network. Background Art

[0002] In a power system, the stable operation of power equipment is crucial for ensuring the safety of the system. Fault detection of power equipment is carried out to timely know whether a power equipment has a fault. When a power equipment has a fault, it needs to be repaired to avoid the continued operation of the faulty power equipment and eliminate potential safety hazards. However, in actual applications, the complexity of the equipment and the diversity of the operating environment make it difficult to detect equipment faults.

[0003] To improve the accuracy of fault detection, the invention application with the publication number CN115979456A in the prior art discloses a method, device and equipment for fault detection of power equipment, which collects the numerical values of multiple target dimensions of the power equipment and detects the corresponding fault level of the power equipment, that is, uses machine learning technology to detect the corresponding fault level of the power equipment, and can improve the accuracy of the fault detection result of the power equipment.

[0004] In actual applications, overheating of power equipment may be caused by faults, but the fault conditions of power equipment are not only related to temperature. Feature data in other dimensions can also reflect relevant faults of power equipment. For example, for motor fault diagnosis, a vibration sensor can detect machine vibration, and the machine vibration data reflects mechanical imbalance faults. Current and voltage sensors can detect the current and voltage of power equipment, and the current and voltage data can reflect electrical faults. Generally, a threshold is set for multi-dimensional data. In the case where the data exceeds the threshold, it can only be judged that the data is abnormal, and it cannot directly judge whether there is a fault in the power equipment. The method of fault judgment through the threshold is relatively single, and the accuracy of fault diagnosis is not high. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for fault detection of power equipment based on a deep learning network, aiming to solve the problem that the traditional fault detection technology can only directly obtain that the data is abnormal by setting a threshold, and cannot directly judge whether there is a fault in the power equipment. The method of fault judgment is relatively single, and the accuracy of fault diagnosis is not high.

[0006] In the first aspect, the present invention provides a method for fault detection of power equipment based on a deep learning network, and the method includes:

[0007] Construct a deep learning model composed of a memory module and a convolutional neural network, and construct a fault classification model based on the convolutional neural network;

[0008] Select an observation period to obtain the state parameter data of the power equipment during the observation period, and form a first multi-dimensional time series according to the state parameter data. The first multi-dimensional time series includes the state parameter data of the current time node and at least one past time node. Input the first multi-dimensional time series into the deep learning model to obtain the fault probability prediction values corresponding to each time node respectively;

[0009] Obtain a fault coefficient according to the fault probability prediction values of all time nodes, and judge whether the power equipment has a fault according to the fault coefficient;

[0010] If the power equipment has a fault, input the fault probability prediction values corresponding to each time node into the fault classification model to obtain a fault category set corresponding to the observation period, and obtain the fault category of the current time node according to the fault category set.

[0011] Further, the steps of constructing the deep learning model composed of a memory module and a convolutional neural network, and constructing the fault classification model based on the convolutional neural network include:

[0012] For any time point t, collect the historical data of each state parameter of the power equipment from time point t - n to t to form a second multi-dimensional time series;

[0013] Input the second multi-dimensional time series into the memory module, perform feature extraction in the time dimension and output it into the convolutional neural network, and perform local feature extraction and high-level feature extraction in sequence to obtain the fault probability prediction value;

[0014] Adjust the weights of each layer in the memory module and the convolutional neural network according to the fault probability prediction value.

[0015] Further, the steps of inputting the second multi-dimensional time series into the memory module, performing feature extraction in the time dimension and outputting it into the convolutional neural network, and performing local feature extraction and high-level feature extraction in sequence to obtain the fault probability prediction value include:

[0016] In the memory module, the feature extraction process in the time dimension is as follows:

[0017] Calculate the forget gate through the following formula:

[0018] ;

[0019] Calculate the input gate and the candidate cell state at the current time point t through the following formula:

[0020] ;

[0021] ;

[0022] Update the cell state at the current time point \(t\) using the forget gate, input gate, and candidate cell state through the following formula:

[0023] ;

[0024] Calculate the output gate and update the hidden state at the current time \(t\) through the following formula:

[0025] ;

[0026] ;

[0027] In the formula, is the forget gate, is the input gate, is the candidate cell state at time point \(t\), is the cell state at time point \(t\), is the output gate, is the hidden state at time point \(t\), is the sigmoid function, is the hidden state at time point \(t - 1\), is the real-time data of the state parameter at time point \(t\), , , , , , , , are all learnable parameters in the memory module, tanh is the activation function, is the cell state at time point \(t - 1\).

[0028] Furthermore, the step of inputting the second multi-dimensional time series into the memory module, extracting features in the time dimension and outputting them into the convolutional neural network, and sequentially performing local feature extraction and high-level feature extraction to obtain the fault probability prediction value further includes:

[0029] Substitute the hidden state at time point \(t\) into the convolutional neural network, and the convolutional neural network is set to two layers;

[0030] In the first-layer convolutional neural network, use the following formula to perform convolution on the hidden state at time point \(t\) and then perform non-linear transformation to obtain the output value:

[0031] ;

[0032] ;

[0033] In the formula, , are the learnable parameters of the first-layer convolutional neural network, is the first convolution value, A is the output value, and ReLu is the activation function;

[0034] In the second-layer convolutional neural network, the following formula is used to perform a convolution operation on the output value and convert the convolution value into a fault probability prediction value based on the softmax function:

[0035] ;

[0036] ;

[0037] wherein, 、 are the learnable parameters of the second-layer convolutional neural network, is the second convolution value, and p is the fault probability prediction value.

[0038] Further, the step of obtaining a fault coefficient according to the fault probability prediction values at all time nodes and determining whether the power equipment has a fault according to the fault coefficient includes:

[0039] Obtain the fault coefficient according to the following formula:

[0040] ;

[0041] wherein, is the fault probability prediction value at the t-th time point, T represents the observation period, and Y represents the fault coefficient under the observation period.

[0042] Judge whether the fault coefficient under the observation period is greater than a preset fault coefficient threshold;

[0043] If the fault coefficient under the observation period is greater than the preset fault coefficient threshold, it is determined that the power equipment has a fault.

[0044] Further, the step of, if the power equipment has a fault, inputting the fault probability prediction values corresponding to each time node into the fault classification model to obtain a fault category set corresponding to the observation period and obtaining the fault category of the current time node according to the fault category set includes:

[0045] Map the fault probability prediction value of the convolutional neural network at each time point to the fault category parameter u through the fully connected layer:

[0046] ;

[0047] wherein, and are both parameters of the fully connected layer, and p is the fault probability prediction value;

[0048] The fault category parameter u is transformed into probabilities of different fault categories through the softmax function:

[0049] ;

[0050] In the formula, represents the fault probability that the time point corresponding to the fault category parameter u is the j-th fault category, c is the total number of fault categories, and exp represents the natural exponential function;

[0051] The maximum fault probability is selected from all the fault probabilities at the same time point, and the fault category corresponding to the maximum fault probability is used as the final fault category at this time point;

[0052] The final fault probabilities at all time points during the observation period are output.

[0053] Furthermore, the method further includes:

[0054] During the training process of the memory module and the convolutional neural network, the objective function is defined as:

[0055]

[0056] In the formula, is the total number of samples, is the actual state of the power equipment, is the predicted value of the fault probability obtained through the deep learning model, indicates that the actual state of the power equipment at time point t is normal, indicates that the actual state of the power equipment at time point t is abnormal.

[0057] In a second aspect, the present invention provides a power equipment fault detection system based on a deep learning network, and the system includes:

[0058] A model construction module, configured to construct a deep learning model composed of a memory module and a convolutional neural network, and construct a fault classification model based on the convolutional neural network;

[0059] A probability prediction module, configured to select an observation period to obtain state parameter data of the power equipment during the observation period, form a first multi-dimensional time series according to the state parameter data, where the first multi-dimensional time series includes state parameter data of the current time node and at least one past time node, and input the first multi-dimensional time series into the deep learning model to obtain predicted values of fault probabilities corresponding to each time node respectively;

[0060] A fault detection module, configured to obtain a fault coefficient according to the predicted values of the fault probabilities of all time nodes, and determine whether the power equipment has a fault according to the fault coefficient;

[0061] The fault category output module is used to input the fault probability prediction value corresponding to each time node into the fault classification model if there is a fault in the power equipment, obtain the fault category set corresponding to the observation period, and obtain the fault category of the current time node according to the fault category set.

[0062] In a third aspect, the present invention provides a storage medium, which stores one or more programs, which, when executed by a processor, implement the above-mentioned power equipment fault detection method based on deep learning network.

[0063] In a fourth aspect, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein:

[0064] The memory is used to store computer programs;

[0065] When the processor is used to execute the computer program stored in the memory, the above-mentioned power equipment fault detection method based on deep learning network is implemented.

[0066] In summary, according to the above-mentioned power equipment fault detection method based on deep learning network, the deep learning model is trained through the historical data of various state parameters of the power equipment, and the fault probability prediction value of the fault and the fault coefficient of the power equipment are accurately calculated according to the deep learning model. If the fault coefficient exceeds the threshold, further fault classification detection is performed, and the fault coefficient of the power equipment can be calculated through multi-dimensional state parameters. The fault coefficient is used as a judgment standard for the current operating status of the power equipment. If the operating status is not good, further fault classification detection is performed, which improves the efficiency of the fault diagnosis process. In addition, the fault classification model is combined with the fault prediction probability obtained based on the multi-dimensional state parameter analysis to jointly perform comprehensive classification diagnosis on the fault, avoiding the situation where fault judgment is made through single-dimensional data and the diagnosis result is inaccurate due to data abnormality, thereby improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A flowchart of a method for detecting faults in electric power equipment based on a deep learning network proposed in one embodiment of the present invention;

[0068] Figure 2 This is a schematic diagram of the structure of a power equipment fault detection system based on a deep learning network proposed in one embodiment of the present invention.

[0069] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein shall have the ordinary meaning as understood by those of ordinary skill in the art to which the present invention pertains. As used herein, words such as "including" and the like are intended to mean that the elements or items appearing before the word cover the elements or items listed after the word and their equivalents, without excluding other elements or items.

[0071] As Figure 1 shown, an embodiment of the present invention provides a power equipment fault detection method based on a deep learning network. The method includes steps S101 to S104, where:

[0072] Step S101: Construct a deep learning model composed of a memory module and a convolutional neural network, and construct a fault classification model based on the convolutional neural network;

[0073] It should be noted that in some embodiments, during the training process of the memory module and the convolutional neural network, the objective function is defined as:

[0074] ;

[0075] In the formula, N is the total number of samples, is the actual state of the power equipment, is the predicted fault probability value obtained through the deep learning model. The optimization objective of this deep learning model is to minimize the gap between the predicted fault probability value and the actual state of the power equipment. After optimizing the deep learning model using the above objective function, the predicted fault probability value is calculated using actual data, and the mean value of the predicted fault probability value is obtained to get the fault coefficient of the power equipment.

[0076] In addition, in some other embodiments of the present invention, during the training process of the memory module and the convolutional neural network, the objective function is defined as:

[0077]

[0078] In the formula, is the total number of samples, is the actual state of the power equipment, is the predicted fault probability value obtained through the deep learning model, indicates that the actual state of the power equipment at time point t is normal, Indicates that the actual state of the power equipment at time point t is abnormal. During the training process, if the actual state of the power equipment is normal, the goal is to minimize the residual, that is, to minimize the gap between the predicted fault probability value and the actual state of the power equipment. If the actual state of the power equipment is abnormal, the goal is to maximize the residual, that is, to maximize the gap between the predicted fault probability value and the actual state of the power equipment. After optimizing the deep learning model using the above objective function, it is possible to effectively monitor the predicted fault probability value of the actual state abnormality of the power equipment, increase the data of the predicted fault probability value under the abnormal state of the power equipment. In this case, the predicted fault probability value can be calculated more accurately, and the mean value of the predicted fault probability value is obtained to get the fault coefficient of the power equipment.

[0079] Step S102: Select an observation period to obtain the state parameter data of the power equipment during the observation period, and form a first multi-dimensional time series according to the state parameter data. The first multi-dimensional time series includes the state parameter data of the current time node and at least one past time node. Input the first multi-dimensional time series into the deep learning model to obtain the predicted fault probability value corresponding to each time node.

[0080] It should be noted that the state parameter data includes, but is not limited to, the temperature data, vibration amplitude, vibration frequency, current and voltage of the power equipment, etc.

[0081] It should be noted that the fault prediction situations in different observation periods may be different. For example, in the actual application process, the equipment was replaced once three months ago, and the historical data in the past three months has reference value, while the previous data has no reference value for the fault prediction result, and the accuracy of the predicted probability value obtained is far less than that in the past three months. Therefore, in actual applications, according to the actual situation, only the historical data in the past three months is collected for the training of the deep learning model.

[0082] In addition, for the structures of different power equipment, different continuous periods also need to be selected to calculate the fault coefficients of the corresponding power equipment. For example, for the vibration of the bearing seat, a continuous period of 2h can be selected as the observation period. During the observation period, the predicted probability value is calculated at every other time period, and the mean value of the predicted probability value is obtained to get the fault coefficient of the power equipment. Select a longer time as the observation period, and the change of this state parameter will not have a great impact on the whole, but only have an impact under long-term abnormal operation, and it can be reflected in time when the fault coefficient is abnormal during the on-site observation period.

[0083] For example, if the corresponding current and voltage are abnormal, a continuous time period of 10 s can be selected as the observation period. During the observation period, the probability prediction value is calculated every other time period (for example, 1 s), and the mean value of the probability prediction values is obtained to get the fault coefficient of the power equipment. Selecting a shorter time as the observation period, the change of this state parameter will have a great impact on the whole, and the short-term abnormality will affect the operation of the power equipment. With a short observation period, the fault coefficient of the power equipment within a short time period can be calculated in a timely and effective manner and reflected.

[0084] Step S103: Obtain the fault coefficient according to the fault probability prediction values of all time nodes, and judge whether the power equipment has a fault according to the fault coefficient;

[0085] Specifically, a continuous time period T including the current time point is selected as the observation period. During the observation period T, the probability prediction value is calculated every other time period, and the mean value of the probability prediction values is obtained to get the fault coefficient of the power equipment:

[0086] ;

[0087] In the formula, is the fault probability prediction value at the t-th time point, T represents the observation period, and Y represents the fault coefficient under the observation period.

[0088] Then judge whether the fault coefficient under the observation period is greater than the preset fault coefficient threshold;

[0089] If the fault coefficient under the observation period is greater than the preset fault coefficient threshold, it is determined that the power equipment has a fault;

[0090] If the fault coefficient under the observation period is less than or equal to the preset fault coefficient threshold, it is determined that the power equipment has no fault, and there is no need to enter the fault classification at this time.

[0091] Step S104: If the power equipment has a fault, input the fault probability prediction values corresponding to each time node into the fault classification model to obtain a fault category set corresponding to the observation period, and obtain the fault category of the current time node according to the fault category set.

[0092] It should be noted that in the fault classification model, based on the fault probability prediction value of the convolutional neural network at each time point, it is mapped to the fault category parameter u through the fully connected layer:

[0093] ;

[0094] In the formula, and are both parameters of the fully connected layer, and p is the fault probability prediction value;

[0095] The fault category parameter u is transformed into the probabilities of different fault categories through the softmax function:

[0096] ;

[0097] In the formula, represents the fault probability that the time point corresponding to the fault category parameter u is the j - th fault category. c is the total number of fault categories, and exp represents the natural exponential function;

[0098] The maximum fault probability is screened out from all the fault probabilities at the same time point, and the fault category corresponding to the maximum fault probability is taken as the final fault category at this time point; finally, the final fault probabilities of all time points in the observation period are output. That is to say, through the method proposed in this step, the relationship between the fault category parameter data and the fault category can be obtained at a certain time point. At the first time point, a fault category A can be obtained, at the second time point, a fault category B can be obtained... and so on, so as to obtain the one - to - one correspondence between the state parameter data and the fault category under continuous time, thus achieving the purpose of fault classification.

[0099] In addition, in some embodiments, in the process of constructing the deep learning model, first, for any time point t, the historical data of each state parameter of the power equipment from time point t - n to t is collected to form a second multi - dimensional time series; then the second multi - dimensional time series is input into the memory module, and after feature extraction in the time dimension, it is output to the convolutional neural network, and local feature extraction and high - level feature extraction are carried out in turn to obtain the fault probability prediction value; finally, according to the fault probability prediction value, the weights of each layer in the memory module and the convolutional neural network are adjusted. Specifically, in the memory module, the process of feature extraction in the time dimension is as follows:

[0100] The forget gate is calculated through the following formula:

[0101] ;

[0102] The input gate and the candidate cell state at the current time point t are calculated through the following formula:

[0103] ;

[0104] ;

[0105] The cell state at the current time point t is updated by using the forget gate, the input gate, and the candidate cell state through the following formula:

[0106] ;

[0107] The output gate is calculated and the hidden state at the current time t is updated through the following formula:

[0108] ;

[0109] ;

[0110] Wherein, is the forgetting gate, is the input gate, is the candidate cell state at time point t, is the cell state at time point t, is the output gate, is the hidden state at time point t, is the sigmoid function, is the hidden state at time point t-1, is the real-time data of the state parameter at time point t, , , , , , , , are all learnable parameters in the memory module, tanh is the activation function, is the cell state at time point t-1.

[0111] In addition, in the construction of the fault classification model, the hidden state at time point t also needs to be substituted into the convolutional neural network, and the convolutional neural network is set to two layers;

[0112] In the first-layer convolutional neural network, the following formula is used to perform convolution on the hidden state at time point t and then perform non-linear transformation to obtain the output value:

[0113] ;

[0114] ;

[0115] Wherein, , are the learnable parameters of the first-layer convolutional neural network, is the first convolution value, A is the output value, ReLu is the activation function;

[0116] In the second-layer convolutional neural network, the following formula is used to perform convolution operation on the output value and convert the convolution value into a fault probability prediction value based on the softmax function:

[0117] ;

[0118] ;

[0119] Wherein, and are the learnable parameters of the second convolutional neural network, is the second convolution value, and p is the predicted value of the fault probability.

[0120] By constructing two new algorithms, namely the deep learning model and the fault classification model, both of which use the historical data of power equipment as the original training set. After collecting the data in a specific observation period, the two models cooperate closely to efficiently and accurately analyze whether there is a fault and the type of the existing fault.

[0121] In summary, according to the above-mentioned power equipment fault detection method based on the deep learning network, the deep learning model is trained through the historical data of each state parameter of the power equipment, and the predicted value of the fault probability and the fault coefficient of the power equipment are accurately calculated according to the deep learning model. If the fault coefficient exceeds the threshold, further fault classification detection is carried out. The fault coefficient of the power equipment can be calculated through multi-dimensional state parameters. The fault coefficient is used as the judgment standard for the current operating state of the power equipment. If the operating state is not good, further fault classification detection is carried out, which improves the efficiency of the fault diagnosis process. In addition, the fault classification model and the fault prediction probability obtained by analyzing multi-dimensional state parameters are combined to comprehensively classify and diagnose the fault, avoiding the situation that the diagnosis result is inaccurate due to data anomalies caused by judging the fault through data in a single dimension, and improving the accuracy of the fault diagnosis.

[0122] As Figure 2 shown, an embodiment of the present invention further provides a power equipment fault detection system based on the deep learning network. The system includes:

[0123] A model construction module 10, configured to construct a deep learning model composed of a memory module and a convolutional neural network, and construct a fault classification model based on the convolutional neural network;

[0124] A probability prediction module 20, configured to select an observation period to obtain the state parameter data of the power equipment during the observation period, and form a first multi-dimensional time series according to the state parameter data. The first multi-dimensional time series includes the state parameter data of the current time node and at least one past time node, and input the first multi-dimensional time series into the deep learning model to obtain the predicted value of the fault probability corresponding to each time node;

[0125] A fault detection module 30, configured to obtain a fault coefficient according to the predicted values of the fault probabilities of all time nodes, and judge whether the power equipment has a fault according to the fault coefficient;

[0126] The fault category output module 40 is configured to, if a fault exists in the power device, input the fault probability prediction values corresponding to each time node into the fault classification model to obtain a fault category set corresponding to the observation period, and obtain the fault category of the current time node according to the fault category set.

[0127] On the other hand, the present invention also proposes a storage medium, on which one or more programs are stored, and when the programs are executed by a processor, the above-mentioned power device fault detection method based on a deep learning network is implemented.

[0128] On the other hand, the present invention also proposes an electronic device, including a memory and a processor, where the memory is used to store a computer program, and the processor is used to execute the computer program stored on the memory to implement the above-mentioned power device fault detection method based on a deep learning network.

[0129] Those skilled in the art can understand that the logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0130] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part with one or more wirings (electronic device), a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or processing it in other suitable ways when necessary, and then storing it in a computer memory.

[0131] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above-described 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 in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), and the like.

[0132] Although the embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and changes can be made to these embodiments. However, it should be understood that such modifications and changes are all within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein can have other embodiments and can be implemented or realized in various ways.

Claims

1. A power equipment fault detection method based on a deep learning network, characterized in that, The method includes: Construct a deep learning model composed of a memory module and a convolutional neural network, and construct a fault classification model based on the convolutional neural network; Substitute the hidden state at time point t into the convolutional neural network, and the convolutional neural network is set to two layers; In the first layer of the convolutional neural network, the following formula is used to perform convolution on the hidden state at time point t and then perform a non-linear transformation to obtain an output value: ； ； In the formula, are the learnable parameters of the first convolutional neural network, is the first convolution value, A is the output value, is the activation function; In the second layer of the convolutional neural network, the following formula is used to perform a convolution operation on the output value and convert the convolution value into a fault probability prediction value based on the softmax function: ； ； wherein, are learnable parameters of the second convolutional neural network, is the second convolution value, and p is the predicted fault probability value; During the training process of the memory module and the convolutional neural network, the objective function is defined as: In the formula, is the total number of samples, is the actual state of the power equipment, is the predicted fault probability value obtained from the deep learning model, indicates that the actual state of the power equipment at time point t is normal, indicates that the actual state of the power equipment at time point t is abnormal; Select an observation period to obtain the state parameter data of the power equipment during the observation period, and form a first multi-dimensional time series according to the state parameter data. The first multi-dimensional time series includes the state parameter data of the current time node and at least one past time node. Input the first multi-dimensional time series into the deep learning model to obtain the fault probability prediction values corresponding to each time node respectively; Obtain a fault coefficient according to the fault probability prediction values of all time nodes, and judge whether the power equipment has a fault according to the fault coefficient; Obtain the fault coefficient according to the following formula: ； Wherein, is the predicted value of the failure probability at the t-th time point, T represents the observation period, and Y represents the failure coefficient under the observation period; Judge whether the fault coefficient under the observation period is greater than a preset fault coefficient threshold; If the fault coefficient under the observation period is greater than the preset fault coefficient threshold, it is determined that the power equipment has a fault; If the power equipment has a fault, input the fault probability prediction values corresponding to each time node into the fault classification model to obtain a fault category set corresponding to the observation period, and obtain the fault category of the current time node according to the fault category set; Map the fault probability prediction value at each time point based on the convolutional neural network to the fault category parameter u through a fully connected layer: ； In the formula, and are both parameters of the fully connected layer, and p is the predicted value of the fault probability; Select the maximum fault probability from all the fault probabilities at the same time point, and use the fault category corresponding to the maximum fault probability as the final fault category at this time point; Output the final fault probabilities of all time points under the observation period.

2. The power equipment fault detection method based on a deep learning network according to claim 1, characterized in that The steps of constructing a deep learning model composed of a memory module and a convolutional neural network, and constructing a fault classification model based on the convolutional neural network include: For any time point t, collect the historical data of each state parameter of the power equipment from time point t-n to t to form a second multi-dimensional time series; Input the second multi-dimensional time series into the memory module, perform feature extraction in the time dimension and output it into the convolutional neural network, and sequentially perform local feature extraction and high-level feature extraction to obtain the fault probability prediction value; Adjust the weights of each layer in the memory module and the convolutional neural network according to the fault probability prediction value.

3. The power equipment fault detection method based on a deep learning network according to claim 2, wherein, The steps of inputting the second multi-dimensional time series into the memory module, performing feature extraction in the time dimension and outputting it into the convolutional neural network, and sequentially performing local feature extraction and high-level feature extraction to obtain the fault probability prediction value include: In the memory module, the feature extraction process in the time dimension is as follows: Calculate the forget gate through the following formula: ； Calculate the candidate cell state of the input gate and the current time point t through the following formula: ； ； Update the cell state of the current time point t by using the forget gate, input gate, and candidate cell state through the following formula: ； Calculate the output gate and update the hidden state of the current time t through the following formula: ； ； In the formula, is the forget gate, is the input gate, is the candidate cell state at time point t, is the cell state at time point t, is the output gate, is the hidden state at time point t, is the sigmoid function, is the time point of the hidden state, is the real-time data of the state parameter at time point t, are all learnable parameters in the memory module, tanh is the activation function, is the time point of the cell state.

4. The method for detecting faults of power equipment based on a deep learning network according to claim 1, characterized in that, The step of inputting the predicted value of the fault probability corresponding to each time node into the fault classification model if the power device has a fault, obtaining a set of fault categories corresponding to the observation period, and obtaining the fault category of the current time node according to the set of fault categories includes: Convert the fault category parameter u into probabilities of different fault categories through the softmax function: ； In the formula, represents the failure probability that the time point corresponding to the failure category parameter u is the j-th failure category, c is the total number of failure categories, and exp represents the natural exponential function.

5. A power equipment fault detection system based on a deep learning network, characterized in that, The system includes: A model construction module for constructing a deep learning model composed of a memory module and a convolutional neural network, and constructing a fault classification model based on the convolutional neural network; Substitute the hidden state at time point t into the convolutional neural network, and the convolutional neural network is set to two layers; In the first layer of the convolutional neural network, perform convolution on the hidden state at time point t through the following formula and then perform a non-linear transformation to obtain an output value: ； ； In the formula, is the learnable parameter of the first convolutional neural network, is the first convolution value, A is the output value, is the activation function; In the second layer of the convolutional neural network, perform a convolution operation on the output value and convert the convolution value into a predicted value of the fault probability based on the softmax function: ； ； In the formula, are the learnable parameters of the second convolutional neural network, is the second convolution value, and p is the predicted value of the fault probability; During the training process of the memory module and the convolutional neural network, define the objective function as: In the formula, is the total number of samples, is the actual state of the power equipment, is the predicted fault probability value obtained through the deep learning model, indicates that the actual state of the power equipment at time point t is normal, indicates that the actual state of the power equipment at time point t is abnormal; A probability prediction module for selecting an observation period to obtain state parameter data of the power device during the observation period, forming a first multi-dimensional time series according to the state parameter data, where the first multi-dimensional time series includes state parameter data of the current time node and at least one past time node, and inputting the first multi-dimensional time series into the deep learning model to obtain predicted values of the fault probability corresponding to each time node; A fault detection module for obtaining a fault coefficient according to the predicted values of the fault probability of all time nodes, and judging whether the power device has a fault according to the fault coefficient; Obtain the fault coefficient according to the following formula: ； In the formula, is the predicted value of the failure probability at the t-th time point, T represents the observation period, and Y represents the failure coefficient under the observation period; Judge whether the fault coefficient under the observation period is greater than a preset fault coefficient threshold; If the fault coefficient under the observation period is greater than the preset fault coefficient threshold, it is determined that the power device has a fault; A fault category output module for inputting the predicted values of the fault probability corresponding to each time node into the fault classification model if the power device has a fault, obtaining a set of fault categories corresponding to the observation period, and obtaining the fault category of the current time node according to the set of fault categories; Map the predicted value of the fault probability at each time point based on the convolutional neural network to the fault category parameter u through a fully connected layer: ； In the formula, and are both parameters of the fully connected layer, and p is the predicted value of the fault probability; Select the maximum fault probability from all the fault probabilities at the same time point, and use the fault category corresponding to the maximum fault probability as the final fault category at this time point; Output the final fault probabilities of all time points under the observation period.

6. A storage medium, characterized in that, The storage medium stores one or more programs, and when the program is executed by a processor, it implements the method for detecting faults in a power device based on a deep learning network as described in any one of claims 1-4.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, where: The memory is used to store a computer program; When the processor is used to execute the computer program stored on the memory, it implements the power equipment fault detection method based on a deep learning network according to any one of claims 1-4.

Images