Modular multilevel converter capacitor fault diagnosis method based on neural network

Abstract

This invention discloses a method for capacitor fault diagnosis in a modular multilevel converter based on neural networks. The invention includes acquiring data such as switching signals of submodule switches, bridge arm currents, and DC voltages in the modular multilevel converter system; processing the acquired data, integrating the switching signals to obtain the differences in switching signals between modules, adding the proportion of the integrated switching signals in the bridge arm as a feature, extracting the current component at twice the frequency of the bridge arm current, and using wavelet packet decomposition to extract the low-frequency fault features of this current; building a regression model based on a neural network, training the model with the obtained data, and finally obtaining a model that can monitor the health status of submodule capacitors in the modular multilevel converter. This enables online calculation of module capacitor values, monitoring the health status of module capacitors, and ensuring the stable operation of the modular multilevel converter system.

Description

Technical Field

[0001] This invention belongs to the field of multilevel converter capacitor fault diagnosis technology, and in particular relates to a modular multilevel converter capacitor fault diagnosis method based on neural networks. Background Technology

[0002] Against the backdrop of energy and environmental issues, distributed generation of new energy sources is currently receiving widespread attention worldwide. Due to the inherent intermittency and instability of distributed power sources, flexible transmission and transformation technologies are needed to address this intermittency when transmitting electricity. These technologies require high-quality output current, and modular multilevel converters are widely used due to their numerous output voltage levels, excellent output current quality, and ease of control.

[0003] Modular multilevel converters consist of numerous sub-modules, which contain a large number of power devices and capacitors, significantly affecting the stability of the system. Therefore, the problem of fault diagnosis for modular multilevel converters has received widespread attention. Summary of the Invention

[0004] The purpose of this invention is to provide a method for diagnosing capacitor faults in modular multilevel converters based on neural networks. This method decomposes the current parameters in the multilevel converter and uses them as one of the fault parameter features of the level converter, thereby reducing the computational difficulty of capacitor monitoring.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0006] This invention is a method for diagnosing capacitor faults in a modular multilevel converter based on neural networks. It includes obtaining the current component of the bridge arm current with respect to twice the fundamental frequency of the reference voltage frequency, performing wavelet packet decomposition on the current component, and extracting low-frequency energy as fault features.

[0007] Furthermore, wavelet packet decomposition is performed on the input current signal, and the energy of at least the first two frequency bands is decentered and used as capacitor fault characteristics.

[0008] Furthermore, when the input current signal is decomposed into wavelet packets, the wavelet packet coefficient matrix is ​​used to represent the degree of fit between the wavelet basis and the input signal at the nodes in that frequency band. The energy information of each node in the corresponding frequency band is calculated using the following formula.

[0009]

[0010] Where E(i) represents the energy of the i-th node, pi,j is the coefficient of the j-th node in the i-th node, and k is the total number of node coefficients of node i.

[0011] Furthermore, the switching signals of the switching transistors are integrated, and the integrated switching signals and the proportion of the integrated switching signals of the module in the corresponding bridge arm are used as fault characteristics. That is, the difference between the switching signals of each module is obtained by integrating the switching signals, and the proportion of the integrated switching signals in the bridge arm is added as one of the fault characteristics.

[0012] Furthermore, during a cycle, when integrating the switching signal, the integral result is multiplied.

[0013] Furthermore, a regression model of a neural network was built and trained using the acquired fault features.

[0014] Furthermore, the model is constructed using a single hidden layer, with fully connected layers between each other, and activation is performed using the sigmoid function after the hidden layer.

[0015] Furthermore, the trained model is installed into a modular multilevel converter.

[0016] The present invention has the following beneficial effects:

[0017] 1. The present invention proposes a method for diagnosing capacitor faults in a modular multilevel converter based on a neural network, which can effectively monitor the health of capacitors in the circuit and ensure the normal and stable operation of the modular multilevel converter.

[0018] 2. This invention applies intelligent algorithms to the traditional field of fault diagnosis, providing a new diagnostic method for this type of fault problem.

[0019] 3. In addressing the problem of extracting fault features, this invention incorporates wavelet packet decomposition, which solves the problem of unclear fault features in this type of fault.

[0020] 4. This invention incorporates a simulated annealing strategy during model training, avoiding the problem of easily getting trapped in local optima during traditional optimization.

[0021] 5. Fault diagnosis is a highly valued problem in practical engineering applications. Compared with traditional fault diagnosis methods, the fault characteristics of this invention are more obvious, which improves the accuracy of existing fault diagnosis methods and has good practicality in engineering applications.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the modular multilevel converter capacitor fault diagnosis method of the present invention;

[0025] Figure 2 Wavelet waveform diagram using db4 wavelet basis as an example;

[0026] Figure 3 This is a diagram of the neural network structure of the present invention;

[0027] Figure 4 This is a graph of activation functions in the neural network of this invention;

[0028] Figure 5 The monitoring results of the model are given when the capacitance of the submodules on the bridge arm is consistently 9mF.

[0029] Figure 6 The monitoring results of the model are given when the capacitance of the submodules on the bridge arm is inconsistent, specifically 8.8mF, 8.6mF, 8.4mF, 8.2mF and 9mF. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] In the description of this invention, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inner", "around", etc., which indicate orientation or positional relationship, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting this invention.

[0032] Please see Figure 1As shown, this invention is a method for diagnosing capacitor faults in a modular multilevel converter based on a neural network. It includes collecting data such as switching signals and bridge arm currents in the modular multilevel converter, extracting fault features from the collected data, building a neural network regression model, training the model using the previously obtained fault features, and finally obtaining a model that can be used for capacitor fault diagnosis.

[0033] Specifically:

[0034] Step 1: Acquire the switching signals, bridge arm currents, and DC bus voltages from the modular multilevel converter. All information is collected in one-cycle increments, for example, 1 second per cycle.

[0035] The second step is to extract features from the collected data. Since the DC bus voltage is relatively stable and basically unchanged, it is directly collected, and the average value within one period is calculated.

[0036] In processing the switching signal, during one cycle, when integrating the switching signal, the integral result is multiplied.

[0037] Integrating the signal according to the acquisition cycle transforms the original 0-1 variation signal into a numerical feature. Due to the inconsistency in capacitance values ​​between different modules, the differences between switching signals gradually increase and become more pronounced during integration. Further removing the average value of the switching signals of each module on the bridge arm and then multiplying them by a weight to increase their effective value further amplifies the differences between signals and improves the computational efficiency and accuracy of subsequent models. The processed feature describes the differences between modules on the entire bridge arm, but it does not clearly indicate which submodule's capacitance the model calculates. To determine which submodule the calculated capacitance belongs to, the proportion of the signal integral value of the switching transistor of that module in the total signal integral value of all modules on the entire bridge arm is added.

[0038] The processing of the bridge arm current involves first passing it through a bandpass filter to obtain the current component of the bridge arm current with respect to twice the fundamental frequency of the reference voltage frequency. This component is then subjected to wavelet packet decomposition to extract the low-frequency energy.

[0039] The input current signal is decomposed using wavelet packets. At least the first two frequency bands are selected and their energy is decentered to serve as capacitor fault features. Taking db4 as the wavelet basis as an example, the steps are as follows:

[0040] The db4 wavelet function is attached. Figure 2 As shown, the input current signal is decomposed into 8-level wavelet packets to obtain the corresponding wavelet packet coefficient matrix, which represents the degree of fit between the wavelet basis and the input signal at the nodes in this frequency band. The energy information of each node in the corresponding frequency band is calculated using the following formula.

[0041]

[0042] Where E(i) represents the energy of the i-th node, pi,j is the coefficient of the j-th node in the i-th node, and k is the total number of node coefficients of node i.

[0043] After decentralizing the energy of the first five frequency bands, this characteristic is used as a capacitor fault feature. When the module capacitance changes, the trend of this feature changes significantly.

[0044] Step 3: Build a neural network regression model and train the model using the extracted features.

[0045] The neural network model construction includes: a shallow BP neural network, consisting of an input layer, a hidden layer, and an output layer. Each layer has at least 10 neurons. Neurons in the same layer are not connected to each other. Neural networks in adjacent layers are fully connected through weights and thresholds. After the hidden layer, a nonlinear transformation is performed through the sigmoid activation function.

[0046] Specifically, the neural network structure is shown in the attached figure. Figure 3 As shown, the model is constructed using a single hidden layer, with fully connected layers between each other. The hidden layer is then activated using a sigmoid function to induce a non-linear transformation. The formula for the sigmoid function is:

[0047]

[0048] In the formula, S(x) is the output value after activation, and x is the value input to the activation function after the hidden layer. Figure 4 This is the graph of the function.

[0049] In the input layer, the input dimension is related to the number of sub-modules in the modular multilevel converter. The input data includes the DC bus voltage, the bridge arm current energy extracted in the second step, the integral of the switching signal, and the proportion of the integral of the switching signal.

[0050] Furthermore, for example, 85 neurons are set in the hidden layer, which are fully connected to the input layer through weights and thresholds. After linear calculation, they are activated by an activation function to complete the nonlinear transformation.

[0051] The activated value is linearly calculated using weights and thresholds and output to the output layer to obtain the calculated module capacitance value.

[0052] The training process of the model uses the collected data, which is divided into training and test sets in a 7:3 ratio. The training process uses the traditional gradient descent backpropagation method, and the root mean square value is chosen as the loss function. First, the predicted output value is obtained based on the forward propagation result. Then, the root mean square value is calculated with respect to the actual sample value to obtain the error as the loss function. Then, the partial derivative of the loss function with respect to the weights and thresholds is calculated to solve for the gradient. Finally, the weights and thresholds are updated according to the direction of gradient descent.

[0053] During the parameter update process, it is common to get stuck in a local optimum, where the loss function no longer changes significantly, but the result is not ideal. Therefore, simulated annealing strategy was added.

[0054] A new temperature parameter T is set, which decreases linearly with the number of iterations, and the decay coefficient can be set to 0.95. During the gradient descent calculation, when encountering a gradient ascent, a number is randomly generated from 0 to 1 and compared with a probability function. If the number is greater than the probability function, the parameter is updated in the direction of gradient ascent. The probability function is calculated as follows:

[0055]

[0056] Where P is the probability, and loss(n+1) and loss(n) are the loss functions before and after the forward propagation in the nth iteration, respectively.

[0057] After training is complete, the training results are tested using the test set.

[0058] Step 4: After training, the model is added to the control of the modular multilevel converter, which enables online monitoring of the health of its module capacitors.

[0059] To demonstrate the effectiveness of the modular multilevel converter capacitor fault diagnosis method proposed in this invention, the system was simulated in MATLAB / Simulink. Figure 5 The monitoring effect of this method on the submodule capacitance is presented when the submodule capacitances on the bridge arm are consistent. Figure 6 The effectiveness of this method in monitoring submodule capacitors when the capacitors on the bridge arms are inconsistent is presented. Therefore, the modular multilevel converter capacitor fault diagnosis method proposed in this invention can effectively monitor the health status of submodule capacitors in modular multilevel converters under different conditions, provide early warning of possible system faults, and ensure the stability and reliability of the converter system.

[0060] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0061] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A neural network-based modular multilevel converter capacitor fault diagnosis method, comprising obtaining a current component of a bridge arm current with respect to a reference voltage frequency of a double fundamental frequency, wavelet packet decomposing the current component, and extracting low-frequency band energy as a fault feature; integrating a switching signal of a switching tube, and taking the integrated switching signal and a proportion of the switching signal integral on the bridge arm as fault features.

2. The neural network based modular multilevel converter capacitor fault diagnosis method according to claim 1, characterized in that, Wavelet packet decomposing an input current signal, and taking energy of at least the first two frequency bands after decentralization as a capacitor fault feature.

3. The neural network based modular multilevel converter capacitor fault diagnosis method of claim 2, wherein, When wavelet packet decomposing the input current signal, a wavelet packet coefficient matrix is used to represent a fitting degree of a wavelet base at a frequency band node and the input signal, and each node calculates energy information of a corresponding frequency band through the following formula: where E(i) represents the energy of the ith node, p i,j is the jth node coefficient in the ith node, and k is the total number of node coefficients that node i has.

4. The neural network based modular multilevel converter capacitor fault diagnosis method of claim 1, wherein, In a period, the integral result is multiplied when integrating the switching signal.

5. The neural network based modular multilevel converter capacitor fault diagnosis method of claim 1, wherein, A regression model of a neural network is built and the model is trained with the obtained fault features.

6. The neural network based modular multilevel converter capacitor fault diagnosis method of claim 5, wherein, The neural network model building comprises a shallow BP neural network composed of an input layer, a hidden layer and an output layer, each layer having at least 10 neurons, the neurons in the same layer being not connected to each other, the neural networks in adjacent layers being fully connected through weights and thresholds, and the hidden layer being nonlinearly transformed through a sigmoid activation function. In the training and optimization process of the model, gradient descent back propagation and simulated annealing strategies are combined for training.

7. The neural network based modular multilevel converter capacitor fault diagnosis method according to claim 6, characterized in that, The trained model is installed into a modular multilevel converter.

Images