V2G inverter system single-tube open-circuit fault diagnosis method based on deep learning
Through deep learning-based methods, the three-phase current output signal of the V2G inverter system is simulated and converted into RGB images, and the CNN model is constructed for fault diagnosis, which solves the inaccurate fault diagnosis of single-tube open circuit of the V2G inverter system in the prior art. The problem of the problem has achieved efficient and accurate fault diagnosis, and improved the reliability and user experience of the system operation.
Patent Information
- Application Number
- CN202411473789.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-02-11
AI Technical Summary
The prior art is difficult to accurately diagnose single-tube open circuit failures in V2G inverter systems, resulting in unstable system operation and affecting energy conversion efficiency and user experience.
Using a deep learning method, by establishing a simulation model of the V2G inverter system, simulating the three-phase current output signal in different states, and converting it into two-dimensional RGB images, and constructing a convolutional neural network (CNN) model for failure. diagnosis. The method includes local diagnosis and collaborative diagnosis, which improves the accuracy and real-timeness of the diagnosis.
It realizes accurate diagnosis of single-tube open circuit faults of V2G inverter system, improves system operation reliability and energy conversion efficiency, reduces unnecessary energy losses, and improves user experience and economic benefits.
Smart Images
Figure CN119418110A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of neural network fault diagnosis, and in particular to a single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning. Background Art
[0002] The Vehicle-to-Grid (V2G) system is an innovative energy management system that allows two-way energy conversion between electric vehicles (EVs) and the grid. The core of this technology is that electric vehicles can not only obtain energy from the grid, but also feed the power in the on-board battery back to the grid, thus achieving a two-way flow of energy. Figure 2 , 3 As shown in the figure, the inverter plays a vital role, responsible for converting the direct current in the electric vehicle into alternating current to achieve bidirectional energy flow with the power grid. As the core power device, the insulated gate bipolar transistor (IGBT) in the inverter has a direct impact on the efficiency and stability of the entire system. Single-tube open-circuit fault, that is, the open-circuit fault of a single IGBT, may cause the inverter performance to deteriorate, thereby affecting the reliability of the entire V2G system. Due to multiple factors such as overheating, overvoltage, short circuit, gate drive problems, mechanical stress, and design material defects, single-tube open-circuit faults in V2G inverters are very likely to occur, and the diagnostic accuracy of this fault is still to be improved. Therefore, it is very important to diagnose the inverter single-tube open-circuit fault in a timely and accurate manner, which can not only ensure the stable operation of the V2G system, but also improve the energy conversion efficiency and reduce unnecessary energy loss, thereby improving the user experience and economic benefits.
[0003] However, there are not many studies on single-tube open-circuit faults of V2G inverters at home and abroad. These studies have some limitations. First, methods based on signal analysis often rely on specific sensors and signal processing technologies, which may require high hardware costs and complex signal processing processes. Second, these methods may not be robust enough under noise interference, signal distortion or complex working conditions. Then, with the diversification of fault types and the increase in system complexity, traditional methods may find it difficult to meet increasingly stringent industrial requirements in terms of diagnostic accuracy and real-time performance. At the same time, the fault information of the inverter may be contained in the current signals of different phases. In most cases, the single-phase current signal cannot accurately reflect the specific fault of the inverter. In addition, for the local convolutional neural network (CNN) fault diagnosis method, if the method of simply increasing the number of samples to improve the accuracy is extremely limited, further improving the diagnostic accuracy is also an important research direction. Summary of the invention
[0004] 1. Technical problems to be solved:
[0005] In view of the above technical problems, the present invention provides a single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning to improve the operating reliability of the V2G system.
[0006] 2. Technical solution:
[0007] A single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning is used to diagnose whether multiple inverters in a V2G system have single-tube open-circuit faults, and is characterized in that it includes the following steps:
[0008] Step 1: According to the mathematical model of the V2G inverter system, a simulation model of the V2G inverter system is established to simulate the three-phase current output signal of the V2G inverter system in different states; the different states are a healthy state in which the inverter has a single-tube open-circuit fault, and a fault state in which a single-tube open-circuit fault exists and the IGBT tube with a specific corresponding sequence number fails; the corresponding sequence number comes from the number of 6 IGBT tubes included in the inverter, which is a positive number less than or equal to 6;
[0009] Step 2: Design a method to map the one-dimensional three-phase current signal obtained by simulation to a two-dimensional RGB image;
[0010] Step 3: The simulation model of the V2G inverter system simulates the three-phase output current of the system under different states, and converts the current into a two-dimensional RGB image using the method of step 2, thereby constructing a state data set of the inverter; the state data set includes 6 different IGBT tube fault data subsets and health data subsets corresponding to different states;
[0011] Step 4: Construct an inverter single-tube open-circuit fault diagnosis model based on a convolutional neural network architecture; in this model, the input two-dimensional RGB image is stacked through multiple convolutional layers and pooling layers and classified by a fully connected layer to achieve image recognition, and finally the probability score of the state corresponding to the image is output through the Softmax layer, that is, the local diagnosis of the single-tube open-circuit fault of a single inverter is achieved; the model is trained using the state data set obtained in step 3 to obtain a trained inverter single-tube open-circuit fault diagnosis model;
[0012] Step 5: Use collaborative diagnosis methods to comprehensively diagnose single-tube open-circuit faults in the system; connect multiple inverters in the V2G inverter system in a star shape using multi-agents, and use directed graphs to represent the communication between the central agent and other agents; train different inverter single-tube open-circuit fault diagnosis CNN models for multiple inverters to achieve local diagnosis of the central inverter, and combine multi-agent and directed graph collaboration to achieve comprehensive diagnosis of the inverter.
[0013] Furthermore, the mathematical model of the V2G inverter system in step 1 is as shown below:
[0014]
[0015] In the above formula, i=1,2,...,n represents the i-th inverter, where the V2G inverter system includes n inverters; R i ,L i ,C i are the resistance, inductance and capacitance corresponding to the i-th inverter; I i.t ,I i ,U i.t ,U i The three-phase variables I represent the three-phase current output by the i-th inverter at time t. i.tabc ,I i.abc ,U i.tabc ,U i.abc The space vector of
[0016] The different states in step one specifically include a healthy state in which there is no single tube open circuit fault, and a single tube open circuit fault state of the first to seventh IGBT tubes.
[0017] Furthermore, in step 2, the one-dimensional three-phase current signal is converted into a two-dimensional RGB image, as shown in the following formula:
[0018]
[0019] Where: P represents the pixel of the converted image; m and n represent the row and column of the pixel P in the two-dimensional RGB image after conversion, respectively; k represents the color channel, that is, k = R, G, B. In this formula, R, G, B correspond to the stator current of phase a, b, and c respectively; Round{} represents the rounding function; S k (j), j∈(1,M 2 ) represents the converted three-phase current signal sample queue, where M*M is the pixel of the converted image, and its size corresponds to the length of the sampling period of the converted three-phase current signal.
[0020] Furthermore, in step three, the six subsets of fault data corresponding to each IGBT tube specifically include: a first IGBT open circuit fault subset; a second IGBT open circuit fault subset; a third IGBT open circuit fault subset; a fourth IGBT open circuit fault subset; a fifth IGBT open circuit fault subset; and a sixth IGBT open circuit fault subset.
[0021] Furthermore, the V2G system inverter fault model based on the convolutional neural network architecture constructed in step 4 includes a feature extraction module composed of multiple convolutional layers and pooling layers, an image classification module composed of fully connected layers, and an output module that uses a Softmax layer to predict the probability corresponding to each state.
[0022] Furthermore, the collaborative diagnosis of faults using the method in step 5 specifically includes:
[0023] S51: Acquire a three-phase current output signal of a V2G inverter system to be diagnosed;
[0024] S52: Convert the three-phase current signal into an RGB image based on step 2;
[0025] S53: The V2G inverter system is represented by a multi-agent system and a star-shaped distribution in a directed graph is adopted; one of the inverters is defined as the central agent in the star;
[0026] S54: inputting the three-phase current signal to be diagnosed into the inverter of the central agent, and outputting the three-phase current signal to other agents based on the inverter of the central agent in the directed graph, and the other agents output the three-phase current signal;
[0027] S55: input the three-phase current signals output by other intelligent agents into the trained model respectively to obtain the state probability score of each IGBT tube of the corresponding inverter;
[0028] S56: Performing a comprehensive diagnosis on the status of the inverter in the V2G inverter system by taking an average of the probability scores in step S57 or the majority rule in step S58;
[0029] S57: Probability score averaging method: The probability scores of each state class of the central agent CNN model and other agent CNN models are averaged and calculated, and the state with the highest average value is regarded as the current state. The comprehensive probability score of the i-th inverter in each state class is calculated as:
[0030]
[0031] In the formula, subscript f c =f 1-7 Respectively represent 7 different status classes of the inverter; a i.θ Indicates the communication direction between the ith inverter and other inverters; d i =∑ θ a i.θ represents the number of inverters that establish direct communication with the i-th inverter, where θ represents the number of inverters that establish direct communication with the i-th inverter; and They represent the state class f in the i-th inverter respectively. c The probability scores evaluated by the central agent CNN and other agent CNN#θ respectively; Yes c The comprehensive probability score of the i-th inverter under the state class; after collecting the comprehensive scores of the 7 state classes, the state class with the highest comprehensive score is defined as the current state of the i-th inverter.
[0032] S58: Majority principle method; after each intelligent CNN with different diagnostic experience diagnoses a sample, if a certain state class is the result of the diagnosis of most CNNs and the state class is unique, then the state class is considered to be the current state; if the state is not unique, the diagnosis is performed by taking the average value of the probability score in step S57.
[0033] 3. Beneficial effects:
[0034] (1) The V2G inverter system single-tube open-circuit fault diagnosis method provided by this method is based on deep learning. The simulation model of the V2G inverter system can simulate the healthy state of the inverter in the V2G and the three-phase current signal of the single-tube open-circuit state, providing a targeted and more physically realistic data set. In addition, the corresponding state data set is constructed according to the different states of the fault, which helps to better determine the type of fault.
[0035] (2) The single-tube open-circuit fault diagnosis method for the V2G inverter system based on deep learning provided by this method converts the three-phase current signal into a two-dimensional RGB image, thereby converting the fault diagnosis problem into an image classification problem, while improving the signal processing rate.
[0036] (3) The single-tube open-circuit fault diagnosis method for the V2G inverter system provided by this method is based on deep learning. The convolutional neural network architecture used in this method performs image recognition on RGB images, which is optimized in terms of diagnostic accuracy and real-time performance.
[0037] (4) The single-tube open-circuit fault diagnosis method for the V2G inverter system provided by this method based on deep learning can improve the accuracy of diagnosis by performing local diagnosis and comprehensive diagnosis on the V2G inverter system. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 This is the overall flow chart of this method;
[0039] Figure 2 The circuit structure of a single inverter in the V2G system involved in this method;
[0040] Figure 3 A schematic diagram of a V2G system and one of the V2G inverter systems involved in the method;
[0041] Figure 4 The schematic diagram of converting the one-dimensional three-phase current signal into an RGB image is shown in step 2 of the method;
[0042] Figure 5 Schematic diagram of the inverter single-tube open-circuit fault diagnosis model constructed in this method;
[0043] Figure 6 It is a schematic diagram of the process of training the inverter single-tube open-circuit fault diagnosis model in a specific embodiment;
[0044] Figure 7 It is a topological structure of collaborative fault diagnosis of four inverters in a specific embodiment. DETAILED DESCRIPTION
[0045] The present invention will be described in detail below with reference to the accompanying drawings.
[0046] As attached Figure 1 As shown, a single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning is used to diagnose whether multiple inverters of a V2G system have a single-tube open-circuit fault, and is characterized in that it includes the following steps:
[0047] Step 1: According to the mathematical model of the V2G inverter system, a simulation model of the V2G inverter system is established to simulate the three-phase current output signal of the V2G inverter system in different states; the different states are a healthy state in which the inverter has a single-tube open-circuit fault, and a fault state in which a single-tube open-circuit fault exists and the IGBT tube with a specific corresponding sequence number fails; the corresponding sequence number comes from the number of 6 IGBT tubes included in the inverter, which is a positive number less than or equal to 6;
[0048] Step 2: Design a method to map the one-dimensional three-phase current signal obtained by simulation to a two-dimensional RGB image;
[0049] Step 3: The simulation model of the V2G inverter system simulates the three-phase output current of the system under different states, and converts the current into a two-dimensional RGB image using the method of step 2, thereby constructing a state data set of the inverter; the state data set includes 6 different IGBT tube fault data subsets and health data subsets corresponding to different states;
[0050] Step 4: Construct an inverter single-tube open-circuit fault diagnosis model based on a convolutional neural network architecture; in this model, the input two-dimensional RGB image is stacked through multiple convolutional layers and pooling layers and classified by a fully connected layer to achieve image recognition, and finally the probability score of the state corresponding to the image is output through the Softmax layer, that is, the local diagnosis of the single-tube open-circuit fault of a single inverter is achieved; the model is trained using the state data set obtained in step 3 to obtain a trained inverter single-tube open-circuit fault diagnosis model;
[0051] Step 5: Use collaborative diagnosis methods to comprehensively diagnose single-tube open-circuit faults in the system; connect multiple inverters in the V2G inverter system in a star shape using multi-agents, and use directed graphs to represent the communication between the central agent and other agents; train different inverter single-tube open-circuit fault diagnosis CNN models for multiple inverters to achieve local diagnosis of the central inverter, and combine multi-agent and directed graph collaboration to achieve comprehensive diagnosis of the inverter.
[0052] Furthermore, the mathematical model of the V2G inverter system in step 1 is as shown below:
[0053]
[0054] In the above formula, i=1,2,...,n represents the i-th inverter, where the V2G inverter system includes n inverters; R i ,L i ,C i are the resistance, inductance and capacitance corresponding to the i-th inverter; I i.t ,I i ,U i.t ,U i The three-phase variables I represent the three-phase current output by the i-th inverter at time t. i.tabc ,I i.abc ,U i.tabc ,U i.abc The space vector of
[0055] The different states in step one specifically include a healthy state in which there is no single tube open circuit fault, and a single tube open circuit fault state of the first to seventh IGBT tubes.
[0056] Furthermore, in step 2, the one-dimensional three-phase current signal is converted into a two-dimensional RGB image, as shown in the following formula:
[0057]
[0058] Where: P represents the pixel of the converted image; m and n represent the row and column of the pixel P in the two-dimensional RGB image after conversion, respectively; k represents the color channel, that is, k = R, G, B. In this formula, R, G, B correspond to the stator current of phase a, b, and c respectively; Round{} represents the rounding function; S k (j), j∈(1,M 2 ) represents the converted three-phase current signal sample queue, where M*M is the pixel of the converted image, and its size corresponds to the length of the sampling period of the converted three-phase current signal.
[0059] Furthermore, in step three, the six subsets of fault data corresponding to each IGBT tube specifically include: a first IGBT open circuit fault subset; a second IGBT open circuit fault subset; a third IGBT open circuit fault subset; a fourth IGBT open circuit fault subset; a fifth IGBT open circuit fault subset; and a sixth IGBT open circuit fault subset.
[0060] Furthermore, the V2G system inverter fault model based on the convolutional neural network architecture constructed in step 4 includes a feature extraction module composed of multiple convolutional layers and pooling layers, an image classification module composed of fully connected layers, and an output module that uses a Softmax layer to predict the probability corresponding to each state.
[0061] Furthermore, the collaborative diagnosis of faults using the method in step 5 specifically includes:
[0062] S51: Acquire a three-phase current output signal of a V2G inverter system to be diagnosed;
[0063] S52: Convert the three-phase current signal into an RGB image based on step 2;
[0064] S53: The V2G inverter system is represented by a multi-agent system and a star-shaped distribution in a directed graph is adopted; one of the inverters is defined as the central agent in the star;
[0065] S54: inputting the three-phase current signal to be diagnosed into the inverter of the central agent, and outputting the three-phase current signal to other agents based on the inverter of the central agent in the directed graph, and the other agents output the three-phase current signal;
[0066] S55: input the three-phase current signals output by other intelligent agents into the trained model respectively to obtain the state probability score of each IGBT tube of the corresponding inverter;
[0067] S56: Performing a comprehensive diagnosis on the status of the inverter in the V2G inverter system by taking an average of the probability scores in step S57 or the majority rule in step S58;
[0068] S57: Probability score averaging method: The probability scores of each state class of the central agent CNN model and other agent CNN models are averaged and calculated, and the state with the highest average value is regarded as the current state. The comprehensive probability score of the i-th inverter in each state class is calculated as:
[0069]
[0070] In the formula, subscript f c =f 1-7 Respectively represent 7 different status classes of the inverter; ai.θ Indicates the communication direction between the ith inverter and other inverters; d i =∑ θ a i.θ represents the number of inverters that establish direct communication with the i-th inverter, where θ represents the number of inverters that establish direct communication with the i-th inverter; and They represent the state class f in the i-th inverter respectively. c The probability scores evaluated by the central agent CNN and other agent CNN#θ respectively; Yes c The comprehensive probability score of the i-th inverter under the state class; after collecting the comprehensive scores of the 7 state classes, the state class with the highest comprehensive score is defined as the current state of the i-th inverter.
[0071] S58: Majority principle method; after each intelligent CNN with different diagnostic experience diagnoses a sample, if a certain state class is the result of the diagnosis of most CNNs and the state class is unique, then the state class is considered to be the current state; if the state is not unique, the diagnosis is performed by taking the average value of the probability score in step S57. Specific embodiment 1:
[0073] As attached Figure 4 As shown in the figure, each sampling value of the three-phase current signal is mapped to each pixel of the two-dimensional RGB image. The three phases of the three-phase current correspond to the three channels of R, G, and B of the RGB image respectively. Each channel is represented by a number of 0-255, which is similar to the data format of each sampling point of the three-phase current. Define M×M sampling data as a sampling cycle, then there are M×M×3 data points in the three-phase current, and each sampling data corresponds to a pixel P in the RGB image.
[0074] In step three, the six data subsets corresponding to each IGBT tube fault include: the first IGBT open circuit fault subset; the second IGBT open circuit fault subset; the third IGBT open circuit fault subset; the fourth IGBT open circuit fault subset; the fifth IGBT open circuit fault subset; the sixth IGBT open circuit fault subset, whose state distribution is the name of the corresponding subset, combined with the healthy state data set, i.e., without IGBT open circuit fault, a total of 7 data subsets constitute the data set for training and verifying the model.
[0075] Before training the model, the obtained RGB image needs to be preprocessed, mainly including the selection of the size of the RGB image, based on the following factors: (1) If the RGB image size is too small, it cannot capture the important details contained in the fault current waveform; (2) If the RGB image size is too large, more sampling samples are required, thereby reducing the fault diagnosis efficiency. Therefore, in this embodiment, the size of the RGB image is selected to be 50×50, that is, one RGB image is generated for every 2500 sampling samples, corresponding to M taking 50 in formula (3).
[0076] In order to train the inverter single-tube open-circuit fault diagnosis model, the number of samples collected for each state mode in this specific embodiment is 4 million, totaling 28 million. For each state mode, the training set and the validation set are divided in a ratio of 8:2, and then this ratio is applied to each state mode. The construction of the entire data set and the model training process are shown in the attached figure. Figure 6 shown. Figure 6 In the figure, #1-7 on the left represent the three-phase current signals in the seven data subsets in step three.
[0077] In this embodiment, the inverter single-tube open-circuit fault diagnosis model is constructed as shown in the attached figure. Figure 5 As shown in the figure, the network structure and depth design of CNN have great flexibility, but no matter how varied, the structure of CNN is usually composed of multiple convolutional layers and pooling layers stacked together, and finally connected to a fully connected layer for classification.
[0078] Each neuron in the convolution layer is connected to multiple neurons in an area close to the previous layer. The size of this area depends on the size of the convolution kernel. The features of the input image are scanned regularly by the convolution kernel, and the calculation between the matrix elements is:
[0079]
[0080] Where: b is the bias constant; Z l and Z l+1 are the feature maps of the lth layer and the l+1th layer respectively; L l+1 Denotes the feature map Z l+1 dimension; Z(m,n) represents the position of the pixel in the feature map; k is the number of channels in the feature map, and in RGB images, k = 3; f, s, and p represent the size, stride, and padding of the convolution kernel, respectively.
[0081] The pooling layer is controlled by the pooling window size, stride, and padding, which can be expressed as:
[0082]
[0083] Where: h is a positive constant. When h = 1, it is called average pooling, and the output is the average value in the pooling area; H→∞ is called Max pooling, and the output is the maximum value in the pooling area.
[0084] Activation functions, such as ReLU (Rectified Linear Unit), are usually added after convolutional and pooling layers to introduce nonlinear characteristics. Activation functions help convolutional neural networks learn complex features.
[0085] The fully connected layer is located at the end of the hidden layer of CNN. Each input neuron is connected to all neurons in the previous layer. The fully connected layer is used to combine the previous convolutional layer and pooling layer to fully process the extracted features.
[0086] The structure of the output layer depends on the specific task. For the image classification task proposed in this method, the Softmax layer is used to predict the probability of each working mode. It is combined with the previous fully connected layer to form a nonlinear model, which converts the output of the model into a probability form, thereby obtaining the probability score of the state class corresponding to the input image. Specific embodiment 2:
[0088] In order to select a CNN architecture suitable for image classification in this method, the accuracy comparison and result analysis of some high-precision CNN architectures in the current image classification field are carried out, including 2-D CNN methods (ShuffleNet, Xception, Inception-v3, VGG-16, AlexNet) and 1-D CNN methods.
[0089] The comparison of the accuracy of the above classification methods is shown in Table 1, where the evaluation accuracy is calculated as shown in the following formula:
[0090]
[0091] Where N T and N F The values represent the number of correctly classified and incorrectly classified samples, respectively.
[0092] Table 1
[0093]
[0094] The results in the table above show that the accuracy of the 2-D CNN method using two-dimensional RGB images is significantly higher than that of the 1-D CNN method in terms of fault diagnosis and image classification. Although the 1-D CNN fault diagnosis method also has a high accuracy, it is still not the most suitable method in this design. In addition, among all 2-D CNN methods, the CNN with the ShuffleNet architecture has the highest accuracy of 97.95% on the RGB image dataset. Therefore, the model in this method prefers the ShuffleNet architecture. Specific embodiment 3:
[0096] The design principle of the collaborative fault diagnosis method proposed in this method is to collect the three-phase current signal of the V2G inverter system to be diagnosed, convert it into an RGB image, and use the information collected through interaction with adjacent inverters to obtain the probability score of each state for local faults using the diagnostic experience of adjacent CNNs, so as to diagnose and classify them. Finally, the diagnostic information of local and adjacent CNNs is integrated to form the final diagnosis result.
[0097] Taking the four-inverter V2G system as an example, the topology of the collaborative diagnosis method is as follows: Figure 7 As shown. Assuming that the first inverter has established direct communication with the second, third, and fourth inverters, the first inverter sends a current signal to the second, third, and fourth inverters and requests a diagnosis result. Next, the second, third, and fourth inverters return the probability score of each state class of the first inverter. Finally, the first inverter combines the returned probability score with the local probability score to evaluate the final diagnosis result.
[0098] The specific process is:
[0099] The probability score averaging method: the probability scores of each state class of the central agent CNN model and other agent CNN models are averaged and calculated, and the state with the highest average value is regarded as the current state. For example, the comprehensive probability score of inverter #1 in each state class is calculated as:
[0100]
[0101] In the formula, subscript f c =f 1-7 Respectively represent 7 different status classes; a 1.θ Indicates the communication direction between the first inverter and other inverters; d 1 =∑ θ＝2,3,4 a 1.θ represents the number of inverters that establish direct communication with the first inverter; and They represent the state class f in the first inverter respectively. c Probability scores evaluated by the central agent CNN and other agent CNNs respectively; Yes c Comprehensive probability score of the first inverter under state class. After collecting the comprehensive scores of the seven classes, the class with the highest comprehensive score is defined as the current state of the first inverter.
[0102] Majority principle method: After each intelligent CNN with different diagnostic experience diagnoses a sample, if a state class is the result of most CNN diagnosis and the state class is unique, then the state class is considered to be the current state; if the state is not unique, the diagnosis is performed by taking the average of the probability scores above.
[0103] From the above process, it can be seen that compared with the probability score averaging method, the majority principle method is equivalent to adding a pre-discrimination condition.
[0104] Although the present invention has been disclosed as above in terms of preferred embodiments, they are not intended to limit the present invention. Anyone skilled in the art can make various changes or modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection defined by the claims of this application.
Claims
1. A single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning is used to diagnose whether multiple inverters in a V2G system have single-tube open-circuit faults, and is characterized by: The following steps are involved: Step 1: According to the mathematical model of the V2G inverter system, a simulation model of the V2G inverter system is established to simulate the three-phase current output signal of the V2G inverter system in different states; the different states are a healthy state in which the inverter has a single-tube open-circuit fault, and a fault state in which a single-tube open-circuit fault exists and the IGBT tube with a specific corresponding sequence number fails; the corresponding sequence number comes from the number of 6 IGBT tubes included in the inverter, which is a positive number less than or equal to 6; Step 2: Design a method to map the one-dimensional three-phase current signal obtained by simulation to a two-dimensional RGB image; Step 3: The simulation model of the V2G inverter system simulates the three-phase output current of the system under different states, and converts the current into a two-dimensional RGB image using the method of step 2, thereby constructing a state data set of the inverter; the state data set includes 6 different IGBT tube fault data subsets and health data subsets corresponding to different states; Step 4: Construct an inverter single-tube open-circuit fault diagnosis model based on a convolutional neural network architecture; in this model, the input two-dimensional RGB image is stacked through multiple convolutional layers and pooling layers and classified by a fully connected layer to achieve image recognition, and finally the probability score of the state corresponding to the image is output through the Softmax layer, that is, the local diagnosis of the single-tube open-circuit fault of a single inverter is achieved; the model is trained using the state data set obtained in step 3 to obtain a trained inverter single-tube open-circuit fault diagnosis model; Step 5: Use collaborative diagnosis methods to comprehensively diagnose single-tube open-circuit faults in the system; connect multiple inverters in the V2G inverter system in a star shape using multi-agents, and use directed graphs to represent the communication between the central agent and other agents; train different inverter single-tube open-circuit fault diagnosis CNN models for multiple inverters to achieve local diagnosis of the central inverter, and combine multi-agent and directed graph collaboration to achieve comprehensive diagnosis of the inverter.
2. The single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning according to claim 1 is characterized in that: The mathematical model of the V2G inverter system in step 1 is shown in the following formula: In the above formula, i=1,2,...,n represents the i-th inverter, where the V2G inverter system includes n inverters; R i ,L i ,C i are the resistance, inductance and capacitance corresponding to the i-th inverter; I i.t ,I i ,U i.t ,U i The three-phase variables I represent the three-phase current output by the i-th inverter at time t. i.tabc ,I i.abc ,U i.tabc ,U i.abc The space vector of The different states in step one specifically include a healthy state in which there is no single tube open circuit fault, and a single tube open circuit fault state of the first to seventh IGBT tubes.
3. The single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning according to claim 2 is characterized in that: In step 2, the one-dimensional three-phase current signal is converted into a two-dimensional RGB image, as shown in the following formula: Where: P represents the pixel of the converted image; m and n represent the row and column of the pixel P in the two-dimensional RGB image after conversion, respectively; k represents the color channel, that is, k = R, G, B. In this formula, R, G, B correspond to the stator current of phase a, b, and c respectively; Round{} represents the rounding function; S k (j), j∈(1,M 2 ) represents the converted three-phase current signal sample queue, where M*M is the pixel of the converted image, and its size corresponds to the length of the sampling period of the converted three-phase current signal.
4. The single-tube open-circuit fault diagnosis method for a V2G inverter system based on deep learning according to claim 3 is characterized in that: In step three, the six subsets of fault data corresponding to each IGBT tube specifically include: the first IGBT open circuit fault subset; the second IGBT open circuit fault subset; the third IGBT open circuit fault subset; the fourth IGBT open circuit fault subset; the fifth IGBT open circuit fault subset; and the sixth IGBT open circuit fault subset.
5. The deep learning-based single-tube open-circuit fault diagnosis method for a V2G inverter system according to claim 4 is characterized in that: The V2G system inverter fault model based on the convolutional neural network architecture constructed in step 4 includes a feature extraction module composed of multiple convolutional layers and pooling layers, an image classification module composed of fully connected layers, and an output module that uses a Softmax layer to predict the probability corresponding to each state.
6. The deep learning-based single-tube open-circuit fault diagnosis method for a V2G inverter system according to claim 5, characterized in that: The collaborative diagnosis of faults using the method in step 5 specifically includes: S51: Acquire a three-phase current output signal of a V2G inverter system to be diagnosed; S52: Convert the three-phase current signal into an RGB image based on step 2; S53: The V2G inverter system is represented by a multi-agent system and a star-shaped distribution in a directed graph is adopted; one of the inverters is defined as the central agent in the star; S54: inputting the three-phase current signal to be diagnosed into the inverter of the central agent, and outputting the three-phase current signal to other agents based on the inverter of the central agent in the directed graph, and the other agents output the three-phase current signal; S55: input the three-phase current signals output by other intelligent agents into the trained model respectively to obtain the state probability score of each IGBT tube of the corresponding inverter; S56: Performing a comprehensive diagnosis on the status of the inverter in the V2G inverter system by taking an average of the probability scores in step S57 or the majority rule in step S58; S57: Probability score averaging method: The probability scores of each state class of the central agent CNN model and other agent CNN models are averaged and calculated, and the state with the highest average value is regarded as the current state. The comprehensive probability score of the i-th inverter in each state class is calculated as: In the formula, subscript f c =f 1-7 Respectively represent 7 different status classes of the inverter; a i.θ Indicates the communication direction between the ith inverter and other inverters; d i =∑ θ a i.θ represents the number of inverters that establish direct communication with the i-th inverter, where θ represents the number of inverters that establish direct communication with the i-th inverter; and Respectively represent the state class f in the i-th inverter c The probability scores evaluated by the central agent CNN and other agent CNN#θ respectively; Yes c The comprehensive probability score of the i-th inverter under the state class; after collecting the comprehensive scores of the 7 state classes, the state class with the highest comprehensive score is defined as the current state of the i-th inverter. S58: Majority principle method; after each intelligent CNN with different diagnostic experience diagnoses a sample, if a certain state class is the result of the diagnosis of most CNNs and the state class is unique, then the state class is considered to be the current state; if the state is not unique, the diagnosis is performed by taking the average value of the probability score in step S57.