Three-phase voltage type PWM rectifier switch tube open-circuit fault diagnosis method

Abstract

The invention discloses a three-phase voltage type PWM rectifier switch tube open-circuit fault diagnosis method, which comprises the steps of 1, implementing three-layer wavelet decomposition on an input side three-phase current historical database after a three-phase voltage type PWM rectifier power switch tube open-circuit fault by adopting a wavelet decomposition method, extracting a characteristic value corresponding to the switch tube fault and finishing normalization; and 2, performing parameter optimization on the support vector machine through an improved mouse group search algorithm, and determining an optimal penalty factor and a kernel parameter so as to carry out data training and fault diagnosis. The invention aims to analyze the waveform distortion condition of the three-phase current at the input side after the open-circuit fault of the three-phase rectifier switch tube, so as to optimize the fault characteristic quantity and realize the accurate diagnosis of the open-circuit fault of the power switch tube. According to the invention, the fault identification capability of the rectifier system under the abnormal working condition can be effectively improved, so that safe and stable operation of power electronic equipment is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rectifier fault diagnosis, in particular to a combined algorithm based on wavelet decomposition and swarm algorithm to optimize support vector machine, which is used to realize accurate identification of open-circuit fault of power switch tube of three-phase voltage type PWM rectifier. BACKGROUND

[0002] With the progress of power electronic device manufacturing process, if the three-phase PWM rectifier has an open-circuit fault of switch tube, its operation stability will be significantly reduced: the load side voltage is difficult to maintain the set value, and phenomena such as pulsation and drop occur, affecting the work of downstream equipment, and in severe cases, causing system shutdown; the input side current waveform is distorted, which destroys the power quality of the power grid. More prominent is that such faults often do not immediately trigger the protection device, are difficult to detect quickly, and long-term operation with faults will exacerbate the stress of other devices, causing secondary faults and expanding economic losses.

[0003] Currently, diagnostic methods for open-circuit faults in three-phase voltage-source PWM rectifier switching transistors are mainly divided into three categories: 1) Signal-based diagnostic techniques, which diagnose by comparing normal and fault signals, and are divided into voltage and current signals. However, current signal diagnosis is prone to misjudgment due to similar waveform distortion when there are many harmonics, power grid fluctuations, or sudden load changes; voltage signal diagnosis usually requires additional sensors or hardware circuits to obtain sensitive signals; 2) Model-based diagnostic techniques, which judge faults by comparing the deviation between measured values ​​and model calculated values. The diagnosis speed is relatively fast, but accurate modeling is difficult, and model deviations and interference from power grid distortion and load fluctuations can easily lead to misjudgment or missed detection; 3) Artificial intelligence-based diagnostic techniques, which use a large amount of historical data to analyze faults through artificial intelligence and other technologies. No circuit model needs to be built, but a large amount of data and high computing power are required. In addition, fault diagnosis methods based on support vector machine models are widely used. However, fault diagnosis methods based on support vector machine models still have some limitations. First, the performance of support vector machines largely depends on the quality of feature extraction and selection. If the input features fail to fully represent the differences between different fault modes, the classification accuracy will be limited. Secondly, Support Vector Machines (SVMs) are essentially binary classification models. When dealing with multi-class and complex faults in three-phase rectifiers, they require the construction of multi-classifiers using strategies such as one-to-many or one-to-one classification. This can lead to complex classification structures, increased computational burden, and imbalanced samples between classes, which can affect diagnostic stability. Furthermore, the kernel function type, penalty coefficient, and kernel parameters of SVMs significantly impact classification performance. However, these parameters often rely on empirical settings or grid search, making optimization time-consuming and prone to getting trapped in local optima, resulting in insufficient adaptability in real-time online diagnostic scenarios. Finally, traditional SVMs are sensitive to the size of the training samples. When fault samples are scarce or unevenly distributed, the model's generalization ability decreases, making it difficult to cope with unknown fault types or changes in operating conditions. Therefore, how to optimize feature engineering by combining rectifier fault characteristics, improve multi-classification strategies, and achieve adaptive adjustment of model parameters remains a key issue that needs to be addressed in current research. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by proposing a method for diagnosing open-circuit faults in three-phase voltage-type PWM rectifier switching transistors based on an improved mouse swarm search optimized support vector machine. The aim is to achieve rapid and accurate diagnosis of open-circuit faults in rectifier power switching transistors when the input current waveform is distorted. This solves problems such as slow response, low accuracy, weak anti-interference, and poor adaptability to operating conditions, and improves the reliability and stability of three-phase rectifier operation.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: The present invention provides a method for diagnosing open-circuit faults in the switching transistors of a three-phase voltage-source PWM rectifier, characterized by the following steps: Step 1: Obtain the M-type fault states of the switching transistors in the three-phase rectifier and the D-type operating states under each fault state, thereby obtaining the three-phase current database of the three-phase rectifier. , This represents the S-phase current data of the three-phase rectifier, and , This represents the s-phase current data of the three-phase rectifier under the i-th type of operating state, and , express middle The phase current value at time s, I represents the overall operating state of the rectifier, and I = M × D, T represents the total duration of a discrete cycle; Let the fault code of the i-th type of operating state be denoted as The fault state vector is obtained. ; Step 2, for Wavelet analysis is performed to obtain the s-phase fault feature vector. Step 3: Based on the s-phase fault feature vector, optimize the parameters of the support vector machine model using an improved mouse swarm algorithm to obtain the optimal penalty coefficient. and kernel function parameters ; Step 4: Optimal penalty coefficient and kernel function parameters The support vector machine model is used as a fault diagnosis model to perform fault diagnosis on the s-phase current data of the online-acquired three-phase rectifier.

[0006] The method for diagnosing open-circuit faults in a three-phase voltage-source PWM rectifier as described in this invention is characterized in that step 2 includes the following steps: Step 2.1, for Perform L-level wavelet analysis to obtain the s-phase low-frequency signal sequence of the L-th layer under the i-th operating state. ,in, The s-phase low-frequency signal of the Lth layer under the i-th type of operating state is represented by the s-phase low-frequency signal of the L-th layer. A low-frequency signal, This represents the signal quantity of the S-phase low-frequency signal, and ; Step 2.2, Extraction Low-frequency signal sequences in the zero range of the positive half-cycle and low-frequency signal sequences in the zero range of the negative half-cycle ,in, express The v-th low-frequency signal is located in the zero-value interval of the positive half-cycle. express The w-th low-frequency signal is located in the zero-value interval of the negative half-cycle. This represents the low-frequency signal quantity within the zero-value interval of the positive half-cycle or the zero-value interval of the negative half-cycle. Step 2.3: Use equation (1) to obtain the energy of the zero-value interval of the positive half-cycle of phase s under the i-th type of operating state. Energy in the zero-value interval of the negative half-cycle of phase s Thus, the energy sequence of the zero-value interval of the positive half-cycle of phase s is obtained. Energy sequence of the negative half-cycle zero value interval of phase s : (1) Step 2.4, and The fault feature vector of phase s in the three-phase rectifier under the i-th type of operating state is normalized to obtain the normalized fault feature vector of phase s in the zero value interval of the positive half-cycle under the i-th type of operating state. and the normalized s-phase negative half-cycle zero value interval under the i-th type of operating state .

[0007] Furthermore, step 3 includes the following steps: Step 3.1: Define the current iteration count as n and initialize n=1; define the maximum iteration count as N. Randomly initialize the location set of the nth generation mouse colony. ,in, Let represent the position vector of the j-th mouse in the n-th generation mouse population, and let represent the position vector of each mouse individual, which is determined by the penalty coefficient in the support vector machine model. and RBF kernel parameters composition, The size of the rat colony; Initialize the energy set of the nth generation mouse colony ,in, This represents the energy of the j-th mouse in the n-th generation mouse colony. This is the initial energy value; Step 3.2: Determine the energy level of the nth generation mouse colony. Is it greater than the set energy value? If yes, proceed to steps 3.3-3.7; otherwise, proceed to steps 3.8-3.9. Step 3.3: Use the position vector of the j-th mouse in the nth generation mouse population. The corresponding penalty coefficient and RBF kernel parameters Construct the j-th support vector machine model in the n-th generation mouse swarm, and... and The j-th support vector machine model from the nth generation mouse swarm is processed to obtain... The corresponding j-th predicted fault state under the nth generation and i-th type of operating state. Thus, the predicted fault state vector of the j-th mouse individual in the nth generation is obtained. ; Step 3.4: Calculate the fitness function value of the j-th mouse individual in the n-th generation mouse population using equation (2). ; (2) In equation (4), It is an indicator function; if the condition inside the parentheses is true, let... =1; otherwise, =0; Step 3.5: Based on the fitness function values, sort each mouse in the nth generation mouse population in descending order, and denote the position vector of the mouse with the largest fitness function value as the position vector of the nth generation mouse king. Rank the stress function values ​​first The position vector of each individual mouse is denoted as the position vector of the nth generation of adult mice. The remaining Let the position vector of each individual mouse be denoted as the nth generation of young mice. ,in, In the nth generation of the mouse colony, the nth generation represents the first generation of the mouse colony. The position vectors of an adult mouse. In the nth generation of the mouse colony, the nth generation represents the first generation of the mouse colony. The position vectors of each individual young mouse, and ; Step 3.6: Update the position vector of the nth generation mouse group using position update strategies for different levels of mouse groups, thereby obtaining the position vector of the (n+1)th generation mouse group set. ,in, This represents the position vector of the j-th mouse in the (n+1)-th generation mouse swarm. Step 3.7: Update the energy of the j-th mouse in the (n+1)-th generation mouse population using equation (6). Then proceed to step 3.9; (6) In equation (6), This represents the energy decay value, and the energy values ​​of the (n+1)th generation mouse population are all the same; Step 3.8: Use equation (7) to obtain the position vector of the j-th mouse individual in the (n+1)-th generation. and the energy of the (n+1)th generation mouse colony Restore to : (7) In equation (7), Indicates in The j-th D-dimensional random number of the nth generation uniformly distributed in the middle; Step 3.9: After assigning n+1 to n, if n>N, then the position vector of the mouse king with the highest fitness among the N generations of mouse kings is taken as the optimal penalty coefficient. and kernel function parameters Otherwise, proceed to step 3.2 sequentially.

[0008] Furthermore, step 3.6 includes the following steps: Step 3.6.1: Use equation (3) to obtain the position vector of the (n+1)th generation mouse king individual. : (3) In equation (3), Let represent the nth generation of random numbers following a D-dimensional standard normal distribution, where D is the dimension of the search space and D=2. The disturbance coefficient; As the lower bound of the parameter, This is the upper bound of the parameter; Step 3.6.2: Use equation (4) to obtain the (n+1)th generation. The position vector of an adult mouse individual : (4) In equation (4), , Representing two nth generation nth terms respectively A random number; Denotes the development coefficient of the nth generation, and , This represents the initial development coefficient; Step 3.6.3: Use equation (5) to obtain the (n+1)th generation. The position vector of each young mouse Random exploration is performed using the following formula: (5) In equation (5), , Representing two nth generation nth terms respectively A random number; Let h be the position vector of the h-th adult mouse individual randomly selected in the nth generation; Denotes the exploration coefficient of the nth generation, and , This represents the initial exploration coefficient.

[0009] The present invention provides an electronic device, including a memory and a processor, characterized in that the memory is used to store a program supporting the processor in performing the method described therein, and the processor is configured to execute the program stored in the memory.

[0010] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program is executed by a processor to perform the steps of the method described thereon.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention uses an improved mouse swarm algorithm to adaptively optimize the key parameters of a support vector machine, effectively solving the problems of traditional support vector machine parameters relying on empirical settings, being prone to getting trapped in local optima, and having limited diagnostic accuracy, thereby significantly improving the accuracy of fault diagnosis and the model's generalization ability.

[0012] 2. This invention designs a wavelet energy feature extraction method for rectifier fault characteristics. It focuses on the zero-value intervals of the positive and negative half-cycles and extracts their low-frequency signal energy as fault feature vectors. Compared to methods using the entire signal or conventional statistical features, this feature extraction strategy can more sensitively and directly capture the current distortion characteristics introduced by faults such as open-circuit switching transistors, effectively suppressing background noise interference, enhancing the distinguishability of fault modes, and laying a solid foundation for subsequent high-precision classification.

[0013] 3. This invention makes key improvements to the traditional mouse swarm optimization algorithm by introducing an energy decay and recovery mechanism based on individual fitness, and a differentiated position update strategy combined with hierarchical partitioning. The energy mechanism dynamically balances the algorithm's "exploration" and "development" behaviors, avoiding premature convergence; the three-level hierarchical strategy simulates the social structure of natural populations, making the search process directional, diverse, and random. These improvements significantly enhance the algorithm's global search capability and convergence speed, enabling more efficient and reliable acquisition of the optimal parameter combination of the support vector machine, ultimately achieving more accurate and robust rectifier fault diagnosis. Attached Figure Description

[0014] Figure 1 This is a diagram of the main circuit and control structure of a three-phase voltage-type PWM rectifier in an embodiment of the present invention; Figure 2 This is a flowchart of the open-circuit fault diagnosis algorithm for a three-phase voltage-source PWM rectifier according to an embodiment of the present invention; Figure 3 This is a diagram of the L-layer wavelet decomposition structure for extracting fault feature values ​​according to the present invention. Figure 4 The low-frequency signal obtained by wavelet decomposition after an open-circuit fault in the switching transistor T1 of a three-phase PWM rectifier. Figure 5 The low-frequency signal obtained by wavelet decomposition after an open-circuit fault in the switching transistors T1 and T2 of a three-phase PWM rectifier. Figure 6 The low-frequency signal obtained by wavelet decomposition after an open-circuit fault in the switching transistors T1 and T4 of a three-phase PWM rectifier. Figure 7 The graph shows the classification accuracy results of the support vector machine without optimized fault feature extraction. Figure 8 The graph shows the accuracy results of support vector machine classification for optimizing fault feature extraction. Figure 9 The image shows the classification accuracy results of the support vector machine optimized based on the conventional mouse swarm algorithm. Figure 10 The figure shows the classification accuracy results of the support vector machine based on the improved mouse swarm algorithm. Detailed Implementation

[0015] In this embodiment, a method for diagnosing open-circuit faults in the switching transistors of a three-phase voltage-source PWM rectifier is described, such as... Figure 2 As shown, this includes: 1. Collecting three-phase current data of the rectifier under different fault states and operating conditions. 2. Extracting the low-frequency components of each phase current using discrete wavelet decomposition, and specifically calculating the signal energy within the zero-value intervals of the positive and negative half-cycles, constructing a feature vector that can sensitively characterize open-circuit faults of the switching transistors, and performing normalization processing. The decomposition structure diagram is shown below. Figure 3 As shown. 3. An improved mouse swarm algorithm is introduced to adaptively optimize the penalty coefficient C and kernel parameter γ of the support vector machine. This improved algorithm dynamically adjusts the search behavior through individual energy decay and recovery mechanisms, and adopts a three-level hierarchical strategy based on fitness ranking to achieve population division of labor, effectively balancing the algorithm's global exploration and local development capabilities, thereby quickly and stably obtaining the optimal parameter combination of the support vector machine model. The final optimal parameter combination is then used to train the final RBF kernel support vector machine model. 4. The rectifier operating current is collected online, and after the same feature extraction process, it is input into the trained optimized support vector machine model for real-time fault diagnosis. Specifically, it is carried out according to the following steps: Step 1: Obtain the M-type fault states of the switching transistors in the three-phase rectifier and the D-type operating states under each fault state, thereby obtaining the three-phase current database of the three-phase rectifier. , This represents the S-phase current data of the three-phase rectifier, and , This represents the s-phase current data of the three-phase rectifier under the i-th type of operating state, and , express middle The current value of phase s at time t, I represents the overall operating state of the rectifier, and I = M × D, T represents the total duration of one discrete cycle; for example Figure 1 As shown, SimLink simulation was used to collect 22 types of fault states of the switching transistors in the three-phase rectifier and 24 types of operating states under each type of fault state, for a total of 528 sets of fault data.

[0016] Step 2, for Wavelet analysis is performed to obtain the s-phase fault feature vector. Step 2.1, as follows Figure 3 As shown, for Perform L-level wavelet analysis to obtain the s-phase low-frequency signal of the L-th layer under the i-th type of operating state. ,in, The s-phase low-frequency signal of the Lth layer under the i-th type of operating state is represented by the s-phase low-frequency signal of the L-th layer. A low-frequency signal, This represents the signal quantity of the S-phase low-frequency signal, and In actual fault diagnosis, the db3 wavelet is used to... Three-level wavelet analysis is performed to improve the problem of high-frequency noise interference and low-frequency fault features overlapping in the original current signal, so that the extracted zero-value interval energy features can accurately reflect the fault state of the rectifier.

[0017] Step 2.2, Extraction Low-frequency signals in the zero range of the positive half-cycle and low-frequency signals in the zero range of the negative half-cycle ,in, express The v-th low-frequency signal is located in the zero-value interval of the positive half-cycle. express The w-th low-frequency signal in the zero-value interval of the negative half-cycle, such as Figure 4 , Figure 5 and Figure 6 As shown, This represents the low-frequency signal quantity within the zero-value interval of the positive or negative half-cycle. By focusing on the zero-value intervals of the positive and negative half-cycles and extracting their low-frequency signal energy as fault feature vectors, this feature extraction strategy can more sensitively and directly capture the current distortion characteristics introduced by faults such as open circuits in switching transistors, effectively suppressing background noise interference and enhancing the distinguishability of fault modes.

[0018] Step 2.3: Use equation (1) to obtain the energy of the zero-value interval of the positive half-cycle of phase s under the i-th type of operating state. Energy in the zero-value interval of the negative half-cycle of phase s Thus, the energy in the zero-value interval of the positive half-cycle of phase s is obtained. Energy in the zero-value interval of the negative half-cycle of phase s : (1) Step 2.4, and The fault feature vector of phase s in the three-phase rectifier under the i-th type of operating state is normalized to obtain the normalized fault feature vector of phase s in the zero value interval of the positive half-cycle under the i-th type of operating state. and the normalized s-phase negative half-cycle zero value interval under the i-th type of operating state .

[0019] Step 3: Based on the s-phase fault feature vector, optimize the parameters of the support vector machine model using an improved mouse swarm algorithm to obtain the optimal penalty coefficient. and kernel function parameters The mouse swarm algorithm is an intelligent optimization algorithm. This invention mainly adopts a three-level hierarchical strategy and an energy update mechanism to improve the traditional mouse swarm algorithm, thereby improving its tendency to get trapped in local optima and enabling the improved mouse swarm algorithm to obtain the optimal support vector machine parameters.

[0020] Step 3.1: Define the current iteration count as n and initialize n=1; define the maximum iteration count as N. Randomly initialize the location set of the nth generation mouse colony. ,in, Let represent the position vector of the j-th mouse in the n-th generation mouse population, and let represent the position vector of each mouse individual, which is determined by the penalty coefficient in the support vector machine model. and RBF kernel parameters composition, The size of the rat colony; Initialize the energy set of the nth generation mouse colony ,in, Let represent the energy of the j-th mouse in the n-th generation mouse population, where the energy of each mouse is the initial energy value. .

[0021] Step 3.2: Determine the energy level of the nth generation mouse colony. Is it greater than the set energy value? If so, proceed to steps 3.3-3.7 and 3.9; otherwise, proceed to steps 3.8-3.9. Step 3.3: Use the position vector of the j-th mouse in the nth generation mouse population. The corresponding penalty coefficient and RBF kernel parameters Construct the j-th support vector machine model in the n-th generation mouse swarm, and... and The j-th support vector machine model from the nth generation mouse swarm is processed to obtain... The corresponding j-th predicted fault state under the nth generation and i-th type of operating state. Thus, the predicted fault state vector of the j-th mouse individual in the nth generation is obtained. .

[0022] Step 3.4: Calculate the fitness function value of the j-th mouse individual in the n-th generation mouse population using equation (2). ; (2) In equation (4), It is an indicator function; if the condition inside the parentheses is true, let... =1; otherwise, =0.

[0023] Step 3.5: Based on the fitness function values, sort each mouse in the nth generation mouse population in descending order, and denote the position vector of the mouse with the largest fitness function value as the position vector of the nth generation mouse king. Rank the stress function values ​​first The position vector of each individual mouse is denoted as the position vector of the nth generation of adult mice. The remaining Let the position vector of each individual mouse be denoted as the nth generation of young mice. ,in, In the nth generation of the mouse colony, the nth generation represents the first generation of the mouse colony. The position vectors of an adult mouse. In the nth generation of the mouse colony, the nth generation represents the first generation of the mouse colony. The position vectors of each individual young mouse, and .

[0024] Step 3.6: Update the position vector of the nth generation mouse group using position update strategies for different levels of mouse groups, thereby obtaining the position vector of the (n+1)th generation mouse group set. ,in, Let represent the position vector of the j-th mouse in the (n+1)-th generation mouse population. Since a single-position update strategy struggles to simultaneously balance convergence speed and search breadth, a three-level hierarchical strategy is employed to clarify the division of labor within the population, balancing the algorithm's global exploration and local exploitation capabilities. Based on fitness, the mouse population is dynamically divided into three levels: the alpha mouse, adult mice, and young mice. The alpha mouse guides the population's direction and performs fine-grained local searches; adult mice follow the alpha mouse to accelerate convergence towards the optimal solution; and young mice explore randomly in a wider space to maintain population diversity. This differentiated update mechanism effectively prevents the population from prematurely clustering and falling into local optima, thus ensuring that the algorithm maintains rapid convergence while possessing strong global optimization capabilities.

[0025] Step 3.6.1: Use equation (3) to obtain the position vector of the (n+1)th generation mouse king individual. : (3) In equation (3), Let represent the nth generation of random numbers following a D-dimensional standard normal distribution, where D is the dimension of the search space and D=2. The disturbance coefficient; As the lower bound of the parameter, This is the upper bound of the parameter.

[0026] Step 3.6.2: Use equation (4) to obtain the (n+1)th generation. The position vector of an adult mouse individual : (4) In equation (4), , Representing two nth generation nth terms respectively A random number; Denotes the development coefficient of the nth generation, and , The initial development coefficient, development coefficient It enables the mouse swarm to perform a detailed and in-depth search in the vicinity of the currently known excellent solutions, and can find the optimal solution in the region. It has strong development capabilities, can accelerate convergence, and improve the accuracy of the solution.

[0027] Step 3.6.3: Use equation (5) to perform random exploration to obtain the (n+1)th generation. The position vector of each young mouse : (5) In equation (5), , Representing two nth generation nth terms respectively A random number; Let h be the position vector of the h-th adult mouse individual randomly selected in the nth generation; Denotes the exploration coefficient of the nth generation, and , The initial exploration coefficient, exploration coefficient It allows the mouse swarm to conduct extensive and random searches in the solution space, enabling them to discover new and potentially excellent solution regions. It has strong exploratory capabilities and can prevent the algorithm from getting trapped in local optima too early.

[0028] Step 3.7: Update the energy of the j-th mouse in the (n+1)-th generation mouse population using equation (6). : (6) In equation (6), This represents the energy decay value, and the energy values ​​of the (n+1)th generation mouse population are all the same.

[0029] Step 3.8: Use equation (7) to conduct random exploration and obtain the position vector of the j-th mouse individual in the (n+1)-th generation. And after exploration, the energy of the (n+1)th generation mouse swarm will be... Restore to initial value : (7) In equation (7), D-dimensional The j-th random number of the nth generation is uniformly distributed. After this step, the process returns to step 3.2 for sequential execution. Addressing the difficulty in dynamically balancing global exploration and local exploitation weights in the mouse swarm algorithm during the search process, an energy decay mechanism is used to increase the algorithm's flexibility and improve its ability to escape local optima. Energy value, as a key indicator controlling the switching of individual mouse behavior patterns, decays with fitness calculations, simulating the energy consumption process of an organism. Dynamic energy changes force individual mice to change their search strategies within specific energy ranges, effectively preventing excessive stagnation in a single region. The cyclical decay and reset mechanism of energy enhances the ergonomics of the population, enabling the algorithm to find the global optimum with a higher probability.

[0030] Step 3.9: After assigning n+1 to n, if n>N, then the position vector of the mouse king with the highest fitness among the N generations of mouse kings is taken as the optimal penalty coefficient. and kernel function parameters Otherwise, proceed to step 3.2 sequentially. Step 4: Optimal penalty coefficient and kernel function parameters The support vector machine model is used as a fault diagnosis model to diagnose faults in the s-phase current data of the online-acquired three-phase rectifier. To verify the effectiveness of this invention, as follows... Figure 7 , Figure 8 , Figure 9 and Figure 10 As shown, comparing the SVM classification accuracy results of the unoptimized fault feature extraction method, the optimized fault feature extraction method, the SVM classification accuracy results optimized based on the conventional mouse swarm algorithm, and the SVM classification accuracy results optimized based on the improved mouse swarm algorithm, it can be found that both the optimized fault feature extraction method and the improved mouse swarm algorithm can effectively improve the accuracy of three-phase rectifier IGBT open circuit fault diagnosis.

[0031] Step 4.1: Data on phase S current Perform the same wavelet analysis to obtain fault characteristic values. , , , , , Then substitute it into the three-phase fault characteristic vector of the three-phase rectifier. , , , , , After normalization, the result is obtained. , , , , , .

[0032] Step 4.2: Use the fault feature values ​​from the above steps as input values ​​for the trained RSO-SVM model to diagnose the actual fault, obtain the fault diagnosis results of the three-phase rectifier, and judge the fault diagnosis results. Step 4.3: Judge the fault diagnosis results obtained in Step 5.2. When an open circuit fault occurs, the system performs fault protection; when no fault occurs, it continues to run and performs fault diagnosis again in the next cycle.

[0033] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0034] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A method for diagnosing open-circuit faults in the switching transistors of a three-phase voltage-source PWM rectifier, characterized in that, Includes the following steps: Step 1: Obtain the M-type fault states of the switching transistors in the three-phase rectifier and the D-type operating states under each fault state, thereby obtaining the three-phase current database of the three-phase rectifier. , This represents the S-phase current data of the three-phase rectifier, and , This represents the s-phase current data of the three-phase rectifier under the i-th type of operating state, and , express middle The phase current value at time s, I represents the overall operating state of the rectifier, and I = M × D, T represents the total duration of a discrete cycle; Let the fault code of the i-th type of operating state be denoted as The fault state vector is obtained. ; Step 2, for Wavelet analysis is performed to obtain the s-phase fault feature vector. Step 3: Based on the s-phase fault feature vector, optimize the parameters of the support vector machine model using an improved mouse swarm algorithm to obtain the optimal penalty coefficient. and kernel function parameters ; Step 4: Optimal penalty coefficient and kernel function parameters The support vector machine model is used as a fault diagnosis model to perform fault diagnosis on the s-phase current data of the online-acquired three-phase rectifier.

2. The method for diagnosing open-circuit faults of switching transistors in a three-phase voltage-source PWM rectifier according to claim 1, characterized in that, Step 2 includes the following steps: Step 2.1, for Perform L-level wavelet analysis to obtain the s-phase low-frequency signal sequence of the L-th layer under the i-th operating state. ,in, The s-phase low-frequency signal of the Lth layer under the i-th type of operating state is represented by the s-phase low-frequency signal of the L-th layer. A low-frequency signal, This represents the signal quantity of the S-phase low-frequency signal, and ; Step 2.2, Extraction Low-frequency signal sequences in the zero range of the positive half-cycle and low-frequency signal sequences in the zero range of the negative half-cycle ,in, express The v-th low-frequency signal is located in the zero-value interval of the positive half-cycle. express The w-th low-frequency signal is located in the zero-value interval of the negative half-cycle. This represents the low-frequency signal quantity within the zero-value interval of the positive half-cycle or the zero-value interval of the negative half-cycle. Step 2.3: Use equation (1) to obtain the energy of the zero-value interval of the positive half-cycle of phase s under the i-th type of operating state. Energy in the zero-value interval of the negative half-cycle of phase s Thus, the energy sequence of the zero-value interval of the positive half-cycle of phase s is obtained. Energy sequence of the negative half-cycle zero value interval of phase s : (1) Step 2.4, and The fault feature vector of phase s in the three-phase rectifier under the i-th type of operating state is normalized to obtain the normalized fault feature vector of phase s in the zero value interval of the positive half-cycle under the i-th type of operating state. and the normalized s-phase negative half-cycle zero value interval under the i-th type of operating state .

3. The method for diagnosing open-circuit faults of switching transistors in a three-phase voltage-source PWM rectifier according to claim 2, characterized in that, Step 3 includes the following steps: Step 3.1: Define the current iteration count as n and initialize n=1; define the maximum iteration count as N. Randomly initialize the location set of the nth generation mouse colony. ,in, Let represent the position vector of the j-th mouse in the n-th generation mouse population, and let represent the position vector of each mouse individual, which is determined by the penalty coefficient in the support vector machine model. and RBF kernel parameters composition, The size of the rat colony; Initialize the energy set of the nth generation mouse colony ,in, This represents the energy of the j-th mouse in the n-th generation mouse colony. This is the initial energy value; Step 3.2: Determine the energy level of the nth generation mouse colony. Is it greater than the set energy value? If yes, proceed to steps 3.3-3.7; otherwise, proceed to steps 3.8-3.

9. Step 3.3: Use the position vector of the j-th mouse in the n-th generation mouse population. The corresponding penalty coefficient and RBF kernel parameters Construct the j-th support vector machine model in the n-th generation mouse swarm, and... and The j-th support vector machine model from the nth generation mouse swarm is processed to obtain... The corresponding j-th predicted fault state under the nth generation and i-th type of operating state. Thus, the predicted fault state vector of the j-th mouse individual in the nth generation is obtained. ; Step 3.4: Calculate the fitness function value of the j-th mouse individual in the n-th generation mouse population using equation (2). ; (2) In equation (4), It is an indicator function; if the condition inside the parentheses is true, let... =1; otherwise, =0; Step 3.5: Based on the fitness function values, sort each mouse in the nth generation mouse population in descending order, and denote the position vector of the mouse with the largest fitness function value as the position vector of the nth generation mouse king. Rank the stress function values ​​first The position vector of each individual mouse is denoted as the position vector of the nth generation of adult mice. The remaining Let the position vector of each individual mouse be denoted as the nth generation of young mice. ,in, In the nth generation of the mouse colony, the nth generation represents the first generation of the mouse colony. The position vectors of an adult mouse. In the nth generation of the mouse colony, the nth generation represents the first generation of the mouse colony. The position vectors of each individual young mouse, and ; Step 3.6: Update the position vector of the nth generation mouse group using position update strategies for different levels of mouse groups, thereby obtaining the position vector of the (n+1)th generation mouse group set. ,in, This represents the position vector of the j-th mouse in the (n+1)-th generation mouse swarm. Step 3.7: Update the energy of the j-th mouse in the (n+1)-th generation mouse population using equation (6). Then proceed to step 3.9; (6) In equation (6), This represents the energy decay value, and the energy values ​​of the (n+1)th generation mouse population are all the same; Step 3.8: Use equation (7) to obtain the position vector of the j-th mouse individual in the (n+1)-th generation. and the energy of the (n+1)th generation mouse colony Restore to : (7) In equation (7), Indicates in The j-th D-dimensional random number of the nth generation uniformly distributed in the middle; Step 3.9: After assigning n+1 to n, if n>N, then the position vector of the mouse king with the highest fitness among the N generations of mouse kings is taken as the optimal penalty coefficient. and kernel function parameters Otherwise, proceed to step 3.2 sequentially.

4. The method for diagnosing open-circuit faults of switching transistors in a three-phase voltage-source PWM rectifier according to claim 3, characterized in that, Step 3.6 includes the following steps: Step 3.6.1: Use equation (3) to obtain the position vector of the (n+1)th generation mouse king individual. : (3) In equation (3), Let represent the nth generation of random numbers following a D-dimensional standard normal distribution, where D is the dimension of the search space and D=2. The disturbance coefficient; As the lower bound of the parameter, This is the upper bound of the parameter; Step 3.6.2: Use equation (4) to obtain the (n+1)th generation. The position vector of an adult mouse individual : (4) In equation (4), , Representing two nth generation nth terms respectively A random number; Denotes the development coefficient of the nth generation, and , This represents the initial development coefficient; Step 3.6.3: Use equation (5) to obtain the (n+1)th generation. The position vector of each young mouse Random exploration is performed using the following formula: (5) In equation (5), , Representing two nth generation nth terms respectively A random number; Let h be the position vector of the h-th adult mouse individual randomly selected in the nth generation; Denotes the exploration coefficient of the nth generation, and , This represents the initial exploration coefficient.

5. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports a processor in executing the method of any one of claims 1-4, the processor being configured to execute the program stored in the memory.

6. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the steps of the method according to any one of claims 1-4.

Images