Inverter open-circuit fault diagnosis method and system

Abstract

The invention belongs to the field of open-circuit fault diagnosis, and particularly relates to an inverter open-circuit fault diagnosis method and system. Inputting the collected current waveform of the motor stator into the established open-circuit fault diagnosis model to obtain a fault diagnosis result; the establishment mode of the diagnosis model comprises the following steps: initializing a population corresponding to a to-be-optimized hyper-parameter of the diagnosis model, then carrying out iteration, dividing particles in the population into particle pairs in pairs during each iteration, taking the particle with a relatively large fitness value in each pair as a winner, taking the rest as a septicem, enabling each winner to directly enter the population subjected to the next iteration, and carrying out the optimization of the to-be-optimized hyper-parameter of the diagnosis model; after being updated according to the Cauchy variation strategy, a septiceper enters a population of next iteration; iterating until a termination condition is met, substituting the value of the to-be-optimized hyper-parameter determined according to the value of the particle into the diagnosis model, and training the diagnosis model through the training sample to complete the establishment of the diagnosis model. By using the Cauchy variation strategy to update the septicees in the particles, the accuracy and the diagnosis efficiency of fault diagnosis can be effectively improved.

Description

Technical Field

[0001] The invention belongs to the field of open circuit fault diagnosis, and in particular relates to an inverter open circuit fault diagnosis method and system. Background Art

[0002] High-voltage disconnectors are important switching devices used to isolate voltage in power systems. They can protect the safety of high-voltage electrical appliances and devices during maintenance work. The inverter is an important component of the motor drive system of the high-voltage disconnector. Its function is to convert direct current into alternating current to supply the AC servo motor. However, due to environmental interference, mechanical vibration, circuit aging and other factors, the inverter may have an open circuit fault. If the open circuit fault of the inverter is not discovered in time, it may directly lead to the failure of the high-voltage disconnector, thus causing safety hazards. Therefore, the research on the open circuit fault diagnosis method of the inverter of the motor drive system of the high-voltage disconnector is of great significance.

[0003] With the continuous deepening of the intelligentization of power grids, the fault diagnosis method of high-voltage disconnectors based on deep learning has become a hot topic among current scholars. Common deep learning models include convolutional neural networks, deep belief networks, long short-term memory networks, etc. Applying deep learning models to open-circuit fault diagnosis is a relatively advanced and reasonable technical direction in this field. However, when deep learning models are used for inverter open-circuit fault diagnosis, the model may often fall into a local optimal solution due to the limitations of heuristic search, insufficient sample data, or improper parameter settings, resulting in insufficient search capabilities of the model, which in turn makes the performance of the deep learning model unstable, and the reliability of its output results is insufficient, affecting the accuracy of the deep learning model in fault diagnosis and the diagnostic efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide an inverter open circuit fault diagnosis method and system, which is used to solve the problem that when using a deep learning model to diagnose an inverter open circuit fault in the prior art, the model falls into a local optimal solution due to the limitations of heuristic search, insufficient sample data or improper parameter settings, thereby affecting the accuracy of the deep learning model in fault diagnosis and the diagnostic efficiency.

[0005] In order to achieve the above object, the present invention provides an inverter open circuit fault diagnosis method, which inputs the collected current waveform of the motor stator into an established open circuit fault diagnosis model to obtain a fault diagnosis result; the establishment method of the diagnosis model includes:

[0006] Initialize the population corresponding to the hyperparameter to be optimized of the diagnostic model and then iterate. In each iteration, divide the particles in the population into pairs of particles. The particle with the larger fitness value in each pair is regarded as the winner, and the others are losers. The winners directly enter the population of the next iteration, and the losers enter the population of the next iteration after being updated according to the Cauchy mutation strategy. Repeat the iteration until the termination condition is met. After determining the value of the hyperparameter to be optimized according to the current value of the particle and substituting it into the diagnostic model, the diagnostic model is trained through the training sample to complete the establishment of the diagnostic model.

[0007] The way of updating the losers according to the Cauchy mutation strategy includes: obtaining the speed of each loser in the t+1th iteration according to the product of the speed of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the setting neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function;

[0008] The position of each loser in the t+1th iteration is obtained according to the position of each loser in the tth iteration and the speed of each loser in the t+1th iteration.

[0009] Beneficial effects: The present invention provides a new inverter open circuit fault diagnosis method, which actually constructs an open circuit fault diagnosis model in advance, and inputs the collected current waveform of the motor stator into the model to obtain the fault diagnosis result. The model is constructed by using a competitive particle swarm algorithm to optimize the hyperparameters of the model, that is, after the winner and the loser are separated by particle competition, the loser is updated using the Cauchy mutation strategy. The strategy can enable the loser to learn from the winner and the neighboring particles at the two levels of speed and position; the purpose of learning from the winner is to improve the search object with the winner as a reference, and the purpose of learning from the neighboring particles is to increase the population diversity. By using the Cauchy mutation strategy to update the loser in the particles, the ability of the algorithm to jump out of the local optimal solution can be enhanced, thereby effectively improving the global search ability of the open circuit fault diagnosis model, making the performance of the model more stable, so that the fault diagnosis results output by it are more reliable, thereby effectively improving the accuracy and efficiency of fault diagnosis.

[0010] Furthermore, the fitness value of the particle is the fault diagnosis accuracy of the open circuit fault diagnosis model network for the training samples when the particle is used as a hyperparameter of the open circuit fault diagnosis model network.

[0011] Further, the method of obtaining the speed of each loser in the t+1th iteration according to the product of the speed of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function includes:

[0012] The velocity of each loser in the t+1th iteration is obtained by weighted summing the product of the velocity of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function; the weight of the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function is set as the adaptive inertia weight, and the weights of the other two items are 1;

[0013] The adaptive inertia weight is:

[0014]

[0015] in, is the adaptive inertia weight; To set the minimum inertia weight; is to set the maximum value of inertia weight; t is the number of iterations; t max is the maximum number of iterations.

[0016] Furthermore, the open circuit fault diagnosis model adopts a BiLSTM-Attention network.

[0017] Furthermore, the hyperparameters of the open circuit fault diagnosis model include the number of neurons, the number of iterations and the learning rate.

[0018] Furthermore, the method of determining the value of the hyperparameter to be optimized according to the current value of the particle includes: taking the value of the particle with the largest current fitness value as the value of the hyperparameter to be optimized.

[0019] Furthermore, the value range of setting the minimum inertia weight is [0.2, 0.5]; the value range of setting the maximum inertia weight is [0.7, 0.95].

[0020] The present invention also provides an inverter open circuit fault diagnosis system, comprising a processor for executing a computer program to implement the steps of the inverter open circuit fault diagnosis method.

[0021] The inverter open circuit fault diagnosis system can achieve the same beneficial effects as the above-mentioned inverter open circuit fault diagnosis method. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flowchart of an inverter open circuit fault diagnosis method in an inverter open circuit fault diagnosis method embodiment;

[0023] Figure 2 An example diagram of a process of updating a loser by using a Cauchy mutation strategy including adaptive weights in an adaptive competitive particle swarm algorithm in an inverter open circuit fault diagnosis method embodiment;

[0024] Figure 3 It is a structural schematic diagram of an LSTM network in an inverter open circuit fault diagnosis method embodiment;

[0025] Figure 4 It is a structural schematic diagram of a BiLSTM network in an inverter open circuit fault diagnosis method embodiment;

[0026] Figure 5 Schematic diagram of the principle of implementing the attention mechanism in the inverter open circuit fault diagnosis method embodiment. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0028] Inverter open circuit fault diagnosis method embodiment

[0029] This embodiment provides an inverter open circuit fault diagnosis method, which implements open circuit fault diagnosis through an open circuit fault diagnosis model; in the process of building the model, the hyperparameters are optimized by applying the Cauchy mutation strategy to particle competition to update the loser. The selected hyperparameters are selected from the optimized hyperparameters and used as the hyperparameters used in training the model. The method of using the Cauchy mutation strategy to update the loser to optimize the hyperparameters can avoid the competitive particle swarm algorithm from falling into a local optimal solution, enhance the search ability of the model, and improve the accuracy and efficiency of fault diagnosis.

[0030] The method specifically inputs the collected current waveform of the motor stator into the established open circuit fault diagnosis model to obtain the fault diagnosis result; the establishment method of the diagnosis model includes:

[0031] Initialize the population corresponding to the hyperparameters to be optimized of the diagnostic model and then iterate. In each iteration, divide the particles in the population into pairs. The particle with the larger fitness value in each pair is the winner, and the others are losers. The winners directly enter the population of the next iteration, and the losers enter the population of the next iteration after being updated according to the Cauchy mutation strategy. Repeat the iteration until the termination condition is met. After determining the value of the hyperparameter to be optimized according to the current value of the particle and substituting it into the diagnostic model, the diagnostic model is trained through the training samples to complete the establishment of the diagnostic model.

[0032] The way of updating the losers according to the Cauchy mutation strategy includes: obtaining the speed of each loser in the t+1th iteration according to the product of the speed of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the setting neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function;

[0033] The position of each loser in the t+1th iteration is obtained according to the position of each loser in the tth iteration and the speed of each loser in the t+1th iteration.

[0034] Specifically, for the inverter of the high-voltage disconnect switch motor drive system, this embodiment selects the current waveform of the motor stator as a sample; Figure 1 As shown, the sample is input into the pre-established and trained open circuit fault diagnosis model, and the output of the model is the fault diagnosis result. The model is established by iterating the particle population corresponding to the hyperparameters to be optimized of the initial diagnosis model (i.e., the initial model framework, whose hyperparameters are also the hyperparameters to be optimized in the initial setting), so that the hyperparameters after the final iteration are involved in the training.

[0035] This embodiment uses a competitive particle swarm algorithm to iterate the particle population; Figure 2 As shown in Figure 1, taking the particle population P(t) corresponding to a certain hyperparameter to be optimized as an example, assuming that the total number of particles in the population P(t) is m (that is, the population P(t) contains m particles, m is an even number, and t is the iteration index), the population P(0) is randomly initialized, then the position of the i-th particle in the t-th generation is X i (t) = (x i,1 (t),x i,2 (t),.. i x. ,n t)(, speed is V i (t)=(v i,1 (t),v i,2 (t),...v i,n (t)); n is the dimension of the search space, that is, the number of hyperparameters to be optimized.

[0036] At each iteration, the particles in the population P(t) are divided into m / 2 pairs (i.e., the particles in the population are divided into pairs of particles); the two particles in each pair are divided into pairs based on the fitness value f(X i (t)) competes, and the particle with the larger fitness value in each particle pair is regarded as the winner, and the position of the winner particle is X w (t), speed is V w (t), k=1,2,...,m / 2; the rest are losers, and the position of the loser particle is X l (t), speed is V l (t). The winner of all particle pairs in the population directly enters the population P(t+1) of the next iteration; the loser enters the population P(t+1) of the next iteration after being updated according to the Cauchy mutation strategy.

[0037] The specific way to update the losers with the Cauchy mutation strategy is as follows: the speed of each loser in the t+1th iteration is obtained by multiplying the speed of each loser in the tth iteration by the standard Cauchy distribution function, the difference between the positions of each winner and each loser in the tth iteration by the Cauchy distribution function, and the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration by the Cauchy distribution function. The formula for calculating the speed of each loser in the t+1th iteration is as follows:

[0038]

[0039] Among them, cauchy1(0,1), cauchy2(0,1) and cauchy3(0,1) are all standard Cauchy distribution functions; V l,k (t+1) is the speed of each loser in the t+1th iteration; V l,k (t) is the speed of each loser in the tth iteration; X w,k (t) is the position of each winner in the tth iteration; X l,k (t) is the position of each loser in the tth iteration; is the adaptive inertia weight; is the average value of the position of the neighboring particles. In the above formula, cauchy1(0,1)V l,k (t) is the inertia term, which is used to ensure the stability of the search process; cauchy2(0,1)(X w,k (t)-X l,k (t)) indicates that the loser learns the winner's position and uses the winner as a reference to improve the search direction; It represents the loser’s learning of the neighboring particles and introduces the average value of the neighboring particle positions to increase the population diversity and thus enhance the search capability.

[0040] After finding the speed of each loser in the t+1th iteration, it is necessary to calculate the position of each loser in the t+1th iteration. The specific calculation method is: according to the position of each loser in the tth iteration and the speed of each loser in the t+1th iteration, the position of each loser in the t+1th iteration is obtained. The formula for calculating the position of each loser in the t+1th iteration is as follows:

[0041] X l,k (t+1)=X l,k (t)+V l,k (t+1)

[0042] Among them, X l,k (t+1) is the position of each loser in the t+1th iteration; X l,k (t) and V l,k The meaning of (t+1) is the same as that in the formula for calculating the speed of each loser in the t+1th iteration.

[0043] After finding the speed and position of each loser in the t+1th iteration, that is, after obtaining the optimized hyperparameters, it is determined whether the termination condition is met; if the termination condition is not met, the particle population is continued to be iterated; if the termination condition is met, the iteration is terminated and the particle with the largest fitness value is selected from the current particle population, and its value is substituted into the open circuit fault diagnosis model (since the quality of the particles after the termination of the iteration is relatively high, in other embodiments, if the optimization requirement for the hyperparameters is not very high, a particle can be adaptively selected from the particles after the termination of the iteration according to specific requirements, and its value is substituted into the open circuit fault diagnosis model). Then, a commonly used model training method is used for training. After the training is completed, the open circuit fault diagnosis model is established, which will not be repeated here.

[0044] In this embodiment, the fitness value of the particle is the fault diagnosis accuracy of the open circuit fault diagnosis model network for the training samples when the particle is used as a hyperparameter of the open circuit fault diagnosis model network (the open circuit fault diagnosis model in this embodiment adopts the BiLSTM-Attention network).

[0045] Specifically, let the fitness value f(X i (t)) is the fault diagnosis accuracy of the BiLSTM-Attention network for training samples when particles are used as the hyperparameters of the open circuit fault diagnosis model network. The specific calculation formula is as follows:

[0046]

[0047] in, is the fault diagnosis label of the BiLSTM-Attention network for the training samples; is the true label of the training sample; sum() represents the calculation and The total number of correct matches; N train is the total number of training samples.

[0048] The BiLSTM-Attention network used in this embodiment has a high optimization complexity of its hyperparameters, and the optimization process is prone to fall into local optimality; therefore, the Cauchy mutation strategy is introduced when updating the loser to enhance the ability of the competitive particle swarm algorithm to jump out of the local optimality.

[0049] The above BiLSTM-Attention network includes Figure 3 The LSTM network shown in the figure is mainly composed of a forget gate, an input gate, and an output gate.

[0050] The forget gate determines whether information is retained or discarded. The calculation process is as follows:

[0051] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0052] Among them, σ represents the sigmoid function, f t is the output of the forget gate at time t, h t-1 is the output of the hidden layer at time t-1, x t is the input at time t, W f and b f They are the weight matrix and bias term of the forget gate respectively;

[0053] The input gate controls whether the information is accepted by the memory unit. The calculation process is as follows:

[0054] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0055] Among them, i t is the output of the input gate at time t, W i and b i They are the weight matrix and bias term of the input gate respectively;

[0056] The output gate determines the output information, and the calculation process is as follows:

[0057] o t =σ(W o ·[h t-1 ,x t ]+bo )

[0058] Among them, t is the output of the output gate at time t, W o and b o They are the weight matrix and bias term of the output gate respectively;

[0059] C t is the state of the temporary memory unit at the moment, and the calculation process is as follows:

[0060] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0061] In the formula, o t is the output of the output gate at time t, W o , b o They are the weight matrix and bias term of the output gate respectively;

[0062] C t is the state of the temporary memory unit at time t, and the calculation process is as follows:

[0063] C t =tanh(W C ·[h t-1 ,x t ]+b C )

[0064] Where tanh represents the tanh function, W o and b o They are the weight matrix and bias term of the temporary memory unit respectively;

[0065] C t is the state of the memory unit at time t, and the calculation process is as follows:

[0066]

[0067] In the formula, represents element-wise multiplication, It means element-by-element addition;

[0068] h t is the state of the hidden layer at time t, and the calculation process is as follows:

[0069]

[0070] like Figure 4The BiLSTM network shown is a variant of the LSTM network, which is composed of a pair of forward LSTM networks and backward LSTM networks. It can capture both forward and backward information in sequence data. The implementation of the BiLSTM network includes:

[0071]

[0072] In the formula, LSTM() represents the calculation process of the LSTM network. is the state of the forward hidden layer at time t, is the state of the backward hidden layer at time t, y t is the output of the output layer at time t, W y and b y They are the weight matrix and bias term of the output layer of the BiLSTM network respectively;

[0073] The attention mechanism assigns different weights to information at different positions in the sequence data, allowing the BiLSTM network to more effectively focus on the features of data near the time when the fault occurs, thereby improving the performance of the BiLSTM network for fault diagnosis. The implementation of this attention mechanism includes:

[0074] For the input sequence data y=[y1,y2,...,y n ] is encoded to obtain the query vector q;

[0075] Calculate the scoring function s i (y i ,q), the scoring function is defined as follows:

[0076] s i (y i ,q)=V T tanh(Wy i +Uq)

[0077] Where V, W, and U are parameter matrices to be learned.

[0078] like Figure 5 As shown, the Softmax function is used for normalization to obtain the weight a of the input sequence data. i , the calculation process is as follows:

[0079]

[0080] According to the weight a i and input data y i Calculate the weighted sum to get the final fault diagnosis output a:

[0081]

[0082] Furthermore, in this embodiment, the speed of each loser in the t+1th iteration is obtained according to the product of the speed of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function, including:

[0083] The speed of each loser in the t+1th iteration is obtained by weighted summing the product of the velocity of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the neighborhood particle position and the position of each loser in the tth iteration and the Cauchy distribution function; the weight of the product of the difference between the average value of the neighborhood particle position and the position of each loser in the tth iteration and the Cauchy distribution function is set as the adaptive inertia weight, and the weights of the other two items are 1.

[0084] Specifically, the speed of each loser in the t+1th iteration can be calculated based on the product of the speed of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the neighborhood particle positions and the positions of each loser in the tth iteration and the Cauchy distribution function.

[0085] The calculation method specifically includes: performing weighted summation on the product of the velocity of each loser in the t-th iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the t-th iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the t-th iteration and the Cauchy distribution function to obtain the velocity of each loser in the t+1-th iteration; the weight of the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the t-th iteration and the Cauchy distribution function is set as the adaptive inertia weight, and the weights of the other two items are 1.

[0086] The adaptive inertia weight is:

[0087]

[0088] in, is the adaptive inertia weight; To set the minimum inertia weight; is to set the maximum value of inertia weight; t is the number of iterations; t maxis the maximum number of iterations (i.e. the maximum number of iterations of the population); the value range of the set minimum inertia weight is [0.2, 0.5] (i.e. the value of the set minimum inertia weight must be greater than or equal to 0.2 and less than or equal to 0.5); the value range of the set maximum inertia weight is [0.7, 0.95] (i.e. the value of the set maximum inertia weight must be greater than or equal to 0.7 and less than or equal to 0.95).

[0089] Therefore, the adaptive inertia weight Combined with the Cauchy mutation strategy and applied to the competitive particle swarm algorithm, the adaptive competitive particle swarm algorithm (ACSO algorithm) used in this embodiment is obtained. Can control The influence of gradually decays with iterations, thereby ensuring that the early stage of the iterative search has a larger inertia weight so that the particles have a larger search step, so that the algorithm can quickly perform global search; while the later stage of the iterative search has a smaller inertia weight so that the particles have a smaller search step, so that the algorithm can perform more refined local search. Due to the high optimization complexity of the BiLSTM-Attention network hyperparameters, the optimization process needs to balance the global search and local search. Considering e -x The function has a decay characteristic, and the cos(x) function is in the interval (0,π). Combining the decay characteristics of the two functions and after multiple experiments, the above-mentioned adaptive inertia weight suitable for BiLSTM-Attention network hyperparameter optimization is given. In this embodiment, the minimum inertia weight is set in the formula Take 0.4 to set the maximum value of inertia weight Take 0.9.

[0090] In this embodiment, the hyperparameters of the open circuit fault diagnosis model include the number of neurons, the number of iterations, and the learning rate. Specifically, the three hyperparameters of the number of neurons, the number of iterations, and the learning rate have a greater impact on the performance of the model in the method of diagnosing the open circuit fault of the inverter of the high-voltage disconnect switch motor drive system through the open circuit fault diagnosis model network (i.e., the BiLSTM-Attention network). Therefore, the hyperparameters mainly set by the model are the above three hyperparameters. In other embodiments, other hyperparameters can also be set according to specific circumstances and specific needs.

[0091] In this embodiment, the method of determining the value of the hyperparameter to be optimized according to the value of the current particle includes: taking the value of the particle with the largest current fitness value as the value of the hyperparameter to be optimized.

[0092] Specifically, this embodiment uses the size of the fitness value as a condition to measure whether a hyperparameter is a hyperparameter to be optimized; that is, the value of the particle with the largest fitness value in the population P(t) is substituted into the diagnostic model as the value of the hyperparameter to be optimized, and the diagnostic model is trained through training samples to participate in the completion of the establishment of the diagnostic model.

[0093] Inverter open circuit fault diagnosis system embodiment

[0094] This embodiment provides an inverter open circuit fault diagnosis system, which includes a processor having executable program instructions stored therein, and the executable program instructions are used to implement the inverter open circuit fault diagnosis method in the above-mentioned inverter open circuit fault diagnosis method embodiment.

[0095] Since the specific working mode and working principle of the inverter open circuit fault diagnosis system of this embodiment have been described in detail in the above inverter open circuit fault diagnosis method embodiment, they will not be repeated here.

[0096] It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, but do not constitute a limitation to the present invention.

Claims

1. A method for diagnosing an inverter open circuit fault, characterized in that: The collected current waveform of the motor stator is input into the established open circuit fault diagnosis model to obtain the fault diagnosis result; the establishment method of the diagnosis model includes: Initialize the population corresponding to the hyperparameter to be optimized of the diagnostic model and then iterate. In each iteration, divide the particles in the population into pairs of particles. The particle with the larger fitness value in each pair is regarded as the winner, and the others are losers. The winners directly enter the population of the next iteration, and the losers enter the population of the next iteration after being updated according to the Cauchy mutation strategy. Repeat the iteration until the termination condition is met. After determining the value of the hyperparameter to be optimized according to the current value of the particle and substituting it into the diagnostic model, the diagnostic model is trained through the training sample to complete the establishment of the diagnostic model. The way of updating the losers according to the Cauchy mutation strategy includes: obtaining the speed of each loser in the t+1th iteration according to the product of the speed of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the setting neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function; The position of each loser in the t+1th iteration is obtained according to the position of each loser in the tth iteration and the speed of each loser in the t+1th iteration.

2. The inverter open circuit fault diagnosis method according to claim 1, characterized in that: The fitness value of a particle is the fault diagnosis accuracy of the open circuit fault diagnosis model network for training samples when the particle is used as a hyperparameter of the open circuit fault diagnosis model network.

3. The inverter open circuit fault diagnosis method according to claim 1 or 2, characterized in that: Methods for obtaining the speed of each loser in the t+1th iteration according to the product of the speed of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function include: The velocity of each loser in the t+1th iteration is obtained by weighted summing the product of the velocity of each loser in the tth iteration and the standard Cauchy distribution function, the product of the difference between the positions of each winner and each loser in the tth iteration and the Cauchy distribution function, and the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function; the weight of the product of the difference between the average value of the position of the neighborhood particles and the position of each loser in the tth iteration and the Cauchy distribution function is set as the adaptive inertia weight, and the weights of the other two items are 1; The adaptive inertia weight is: in, is the adaptive inertia weight; To set the minimum inertia weight; is to set the maximum value of inertia weight; t is the number of iterations; t max is the maximum number of iterations.

4. The inverter open circuit fault diagnosis method according to claim 1 or 2, characterized in that: The open circuit fault diagnosis model adopts the BiLSTM-Attention network.

5. The inverter open circuit fault diagnosis method according to claim 1 or 2, characterized in that: The hyperparameters of the open circuit fault diagnosis model include the number of neurons, the number of iterations and the learning rate.

6. The inverter open circuit fault diagnosis method according to claim 1 or 2, characterized in that: The method of determining the value of the hyperparameter to be optimized according to the current value of the particle includes: taking the value of the particle with the largest current fitness value as the value of the hyperparameter to be optimized.

7. The inverter open circuit fault diagnosis method according to claim 3, characterized in that: The value range of the minimum inertia weight setting is [0.2, 0.5]; the value range of the maximum inertia weight setting is [0.7, 0.95].

8. An inverter open circuit fault diagnosis system, comprising a processor, characterized in that: The processor is used to execute a computer program to implement the steps of the inverter open circuit fault diagnosis method according to any one of claims 1 to 7.

Images