Charging pile fault diagnosis method and device based on convolutional neural network and wavelet packet decomposition, equipment and medium

Abstract

The invention belongs to the technical field of charging device fault diagnosis, and particularly provides a charging pile fault diagnosis method, device and equipment based on a convolutional neural network and wavelet packet decomposition, and a medium, and the method comprises the following steps: obtaining a charging pile fault current waveform; using a wavelet packet decomposition method to decompose a fault current waveform, and extracting an energy feature vector to generate a feature data set; dividing the generated data set into a training set and a test set; establishing a CNN neural network, and training the CNN neural network by using the feature data of the training set as input to generate a trained CNN model; testing the trained CNN model by using the feature data of the test set; when it is judged that a set precision threshold value is reached according to the test result, a charging pile fault diagnosis model is output; and using the charging pile fault diagnosis model to carry out fault diagnosis on the DC charging pile. And rapid real-time diagnosis of the fault of the charging pile can be realized. Therefore, the safe operation of the charging pile is ensured.

Description

Background Art

[0002] With the rapid development of new energy vehicles, the number of charging piles in operation and the amount of charging are also increasing year by year. Charging piles can be divided into two categories: AC charging piles and DC charging piles. However, the transmission power of DC charging piles is much greater than that of AC charging piles. Therefore, DC charging piles have great advantages in fast charging and are the main type of charging piles currently developed in the market.

[0003] Failure of a DC charging pile can cause great damage to the grid and user load. The failure mainly occurs in the charging module, and its operating status needs to be diagnosed in a timely manner. Due to the complex circuit of the charging module, manual step-by-step troubleshooting is not only time-consuming and laborious, but may also fail to accurately locate the fault, resulting in excessive replacement of components or even replacement of the entire charging pile equipment, increasing maintenance costs.

[0004] At present, diagnosis and prediction methods based on deep learning and expert systems have been proposed for different types of charging pile faults at home and abroad. For example, fuzzy neural algorithms are used to diagnose DC charging piles, but this diagnostic method cannot distinguish between switch tube faults between different bridge arms in the DC chopper circuit. Summary of the invention

[0005] In response to the fault diagnosis problem of a DC charging pile charging module, the present application provides a charging pile fault diagnosis method, device, equipment and medium based on convolutional neural network and wavelet packet decomposition.

[0006] In a first aspect, the technical solution of the present invention provides a charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition, comprising the following steps: Obtain the charging pile fault current waveform; Decompose the fault current waveform using wavelet packet decomposition method, extract energy feature vectors to generate feature data sets; and divide the generated data sets into training sets and test sets; Establish a CNN neural network, use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; Use the feature data of the test set to test the trained CNN model; When the test results determine that the set accuracy threshold is reached, the charging pile fault diagnosis model is output; The charging pile fault diagnosis model is used to perform fault diagnosis of DC charging piles.

[0007] As a further limitation of the technical solution of the present invention, the step of obtaining the charging pile fault current waveform includes: Build the DC charging pile charging module circuit; Set various operating status fault types to perform fault simulation on the charging module circuit and obtain the fault current waveform of the charging pile charging module.

[0008] As a further limitation of the technical solution of the present invention, the steps of using wavelet packet decomposition method to decompose the fault current waveform, extracting energy feature vectors to generate feature data sets; and dividing the generated data sets into training sets and test sets include: The fault current waveform data is classified into labels according to the fault type to generate a fault type label set.

[0009] As a further limitation of the technical solution of the present invention, the steps of decomposing the fault current waveform using wavelet packet decomposition method, extracting energy feature vectors to generate feature data sets; and dividing the generated data sets into training sets and test sets include: Perform three-layer wavelet packet decomposition on the fault current waveform of the DC charging pile charging module; Select the sum of squares of the signal decomposition coefficients in each frequency band as the energy characteristic value of the frequency band; The energy eigenvalues ​​of all decomposed frequency bands are used to generate an energy feature matrix, i.e., a feature data set; The feature data set is divided into a training set and a test set according to a set ratio.

[0010] As a further limitation of the technical solution of the present invention, the energy characteristic value of each frequency band after the three-layer wavelet packet decomposition is: ; Energy signature matrix .

[0011] As a further limitation of the technical solution of the present invention, the steps of establishing a CNN neural network and using the feature data of the training set as input to train the CNN neural network to generate a trained CNN model include: Build a CNN neural network; The training set data with different fault current characteristics are used as input, and the convolution kernel in the convolution layer is used to extract the features of the training set data. The extracted feature matrix is ​​activated and pooled, and the matrix is ​​flattened. The flattened data and the corresponding fault type label set are used as the target, and the absolute value of the difference between the output and the label value is used as the loss function of the error; By calculating the gradient of the loss function, new weight and bias parameters are obtained that make the gradient decrease; The loss function value is recalculated using the new parameters until the loss function value reaches the set requirements to complete the training and generate the trained CNN model.

[0012] As a further limitation of the technical solution of the present invention, the loss function

[0013] In the formula, For input and weight and bias The output product of The weight parameters are updated as follows:

[0014] Where ε is the learning rate.

[0015] In a second aspect, the technical solution of the present invention provides a charging pile fault diagnosis device based on convolutional neural network and wavelet packet decomposition, including a fault data acquisition module, a wavelet packet decomposition processing module, a training module, a testing module, a model output module and an execution diagnosis module; A fault data acquisition module is used to obtain the fault current waveform of the charging pile; A wavelet packet decomposition processing module is used to decompose the fault current waveform using the wavelet packet decomposition method, extract the energy feature vector to generate a feature data set; and divide the generated data set into a training set and a test set; The training module is used to establish a CNN neural network and use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; The testing module is used to test the trained CNN model using the feature data of the test set; A model output module is used to output a charging pile fault diagnosis model when the test result determines that the set accuracy threshold has been reached; The execution diagnosis module is used to perform fault diagnosis of the DC charging pile using the charging pile fault diagnosis model.

[0016] As a further limitation of the technical solution of the present invention, the step of obtaining the charging pile fault current waveform includes: Build the DC charging pile charging module circuit; Set various operating status fault types to perform fault simulation on the charging module circuit and obtain the fault current waveform of the charging pile charging module.

[0017] As a further limitation of the technical solution of the present invention, the device also includes a label classification module, which classifies labels for fault current waveform data according to the fault type and generates a fault type label set.

[0018] As a further limitation of the technical solution of the present invention, the wavelet packet decomposition processing module includes a decomposition unit, a frequency band eigenvalue calculation unit, a characteristic matrix generation unit and a division unit; A decomposition unit, used for performing three-layer wavelet packet decomposition on the fault current waveform of the DC charging pile charging module; A frequency band eigenvalue calculation unit, used for selecting the sum of squares of signal decomposition coefficients in each frequency band as the energy eigenvalue of the frequency band; A characteristic matrix generating unit, used for generating an energy characteristic matrix, i.e., a characteristic data set, from the energy characteristic values ​​of all the frequency bands after decomposition; The partitioning unit is used to partition the feature data set into a training set and a test set according to a set ratio.

[0019] As a further limitation of the technical solution of the present invention, the energy characteristic value of each frequency band after the three-layer wavelet packet decomposition is: ; Energy signature matrix .

[0020] As a further limitation of the technical solution of the present invention, the training module includes a model building unit and a training execution unit; A model building unit, used to build a CNN neural network; The training execution unit is used to take the training set data of different fault current characteristics as input, use the convolution kernel in the convolution layer to extract features of the training set data, activate and pool the extracted feature matrix and flatten the matrix; take the flattened data and the corresponding fault type label set as the target, and use the absolute value of the difference between the output and the label value as the error loss function; calculate the gradient of the loss function to obtain new weights and bias parameters that make the gradient decrease; use the new parameters to recalculate the loss function value until the loss function value reaches the set requirements to complete the training and generate the trained CNN model.

[0021] As a further limitation of the technical solution of the present invention, the loss function

[0022] In the formula, For input and weight and bias The output product of The weight parameters are updated as follows:

[0023] Where ε is the learning rate.

[0024] In a third aspect, the technical solution of the present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; the memory stores computer program instructions executable by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition as described in the first aspect.

[0025] In a fourth aspect, the technical solution of the present invention also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable the computer to execute the charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition as described in the first aspect.

[0026] It can be seen from the above technical solutions that the present invention has the following advantages: By combining wavelet packet decomposition with convolutional neural network (CNN), this method can deeply explore the complex features in the fault current waveform, especially the energy feature vector extracted by wavelet packet decomposition, which effectively improves the accuracy of fault diagnosis. This feature extraction method can capture more subtle signal changes, making the diagnosis results more accurate and reliable. When training the CNN model, a training set that covers a wide range of different fault types is used for training, so that the model can learn rich fault features, thereby showing strong generalization ability in practical applications. Whether it is a known fault type or a potential unknown fault, the model can give a relatively accurate diagnosis result. The entire fault diagnosis process is highly automated and intelligent, and there is no need for excessive human intervention from the acquisition of fault current waveform, feature extraction, model training to the final fault diagnosis. This not only improves the diagnostic efficiency, but also reduces the errors caused by human factors. Due to the high efficiency of the CNN model in processing large-scale data, combined with the fast feature extraction capability of wavelet packet decomposition, this method can realize fast and real-time diagnosis of charging pile faults. This is of great significance for ensuring the safe operation of charging piles and improving the user charging experience.

[0027] When building the DC charging pile charging module circuit and performing fault simulation, multiple operating states and fault types can be flexibly set to obtain rich fault current waveform data. This flexibility enables the method to adapt to charging piles of different brands and models, increasing its application scope and practical value.

[0028] In addition, the invention has a reliable design principle, a simple structure and a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention.

[0031] Figure 2is a schematic block diagram of an apparatus according to an embodiment of the present invention.

[0032] Figure 3 is a three-layer wavelet packet decomposition tree.

[0033] Figure 4 It is a schematic diagram of the two-dimensional convolution process of the CNN network.

[0034] Figure 5 It is the confusion matrix diagram of the test set prediction results.

[0035] Figure 6 This is a diagram of the model prediction effect under different numbers of convolutional layers.

[0036] Figure 7 This is the model prediction effect diagram under the 4-layer convolutional layer network.

[0037] Figure 8 This is a schematic diagram of the fault partitioning of the DC charging pile charging module.

[0038] Fig. 9 This is a typical three-phase rectification circuit diagram of a DC charging pile.

[0039] Fig.10 This is a typical DC charging pile phase-shifted full-bridge circuit diagram. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0041] like Figure 1 As shown, an embodiment of the present invention provides a charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition, comprising the following steps: Step 1: Obtain the charging pile fault current waveform; In some embodiments, the charging pile fault current waveform is obtained from the historical data. Since the fault data obtained from the historical data may have the problem of incomplete fault types, in the embodiment of the present invention, a DC charging pile charging module circuit is built; multiple operating state fault types are set to perform fault simulation on the charging module circuit to obtain the charging pile charging module fault current waveform.

[0042] It should be noted that the charging module of a DC charging pile is generally divided into two-stage circuits. The front stage is a PWM three-phase rectifier circuit, which uses IGBT switching tubes to achieve full control and is a voltage and current dual closed-loop circuit for controlling the power factor stability; the rear stage is a phase-shifted full-bridge DC chopper circuit, which outputs the rectified DC power as a power supply with a suitable DC voltage to charge the car battery.

[0043] A simulation circuit model of the charging module was built using a simulation tool, and then 16 operating state types were set for the charging module for identification. The waveform data of each operating state under different working conditions were obtained. The basic parameters of the circuit simulation are shown in Table 1.

[0044] Table 1

[0045] When a DC charging pile charging module fails, its three-phase current waveform will be distorted, and the fault waveform is different for different types of faults. The operating status of the charging module can be diagnosed based on this fault characteristic. According to the composition structure of the DC charging pile equipment, the main faults of the DC charging pile charging module are divided into four types: rectifier AC side fault, rectifier fault, DC network fault (DC bus, feeder, chopper circuit fault, etc.), and load fault. Figure 8 shown.

[0046] S1-2: Classify the fault current waveform data into labels according to the fault type and generate a fault type label set.

[0047] As one of the most complex circuit modules in the DC charging pile, the charging module contains a large number of small devices such as diodes and power devices. Among the faults that occur in the DC charging pile, the open circuit fault and short circuit fault of the power device have the most serious consequences, and the short circuit fault will be protected by the hardware circuit in a short time, disconnecting the faulty device to an open circuit state. According to statistics, aluminum electrolytic capacitors are also one of the components that are most prone to failure. Therefore, no fault, grid-side phase loss, AC side switch tube S1, S3, S5 open circuit fault, AC side output voltage stabilizing capacitor open circuit fault, phase shift circuit switch tube T1, T3 open circuit fault, phase shift circuit diode D1 open circuit fault, DC side output capacitor open circuit, rectifier circuit same bridge arm or different bridge arm switch tube fault are selected. There are 16 operating states in total, named F1-F16, the simulation time is 0.2s, and the fault occurs at 0.16s.

[0048] Abnormality on the AC side of the rectifier : The abnormality of the AC side of the charging pile rectifier generally refers to the disconnection or phase loss of the three-phase power supply from the grid side. The electrical characteristic of the phase loss fault is the lack of one-phase current waveform. Rectifier abnormality :Rectifier abnormality generally refers to a three-phase rectifier circuit (such as Fig. 9 As shown in the figure, some devices have failed, and the fault location is a switch failure, that is, or Open circuit fault; output side voltage stabilizing capacitor C fault, that is Fig. 9 The parallel capacitor fault on the load side; the inductor fault, that is Fig. 9 When different device faults occur, the inductor current presents different fault deformed waveforms.

[0049] DC network abnormality : DC network faults are divided into abnormalities in the DC bus or feeder branches and abnormalities in the DC chopper circuit. When an open circuit fault occurs in the DC bus or feeder, the power supply of the chopper circuit disappears and the load charging stops, which is equivalent to an open circuit on the load side of the rectifier circuit; the DC chopper circuit (such as Fig.10 The faults include open circuits of switch tubes Q1-Q4, load inductors, transformers, etc.

[0050] Load abnormality: Load abnormality can be manifested as an open circuit or short circuit on the load side, causing the three-phase current waveform on the grid side to be different from the normal working waveform.

[0051] Step 2: Decompose the fault current waveform using wavelet packet decomposition method, extract the energy feature vector to generate a feature data set; and divide the generated data set into a training set and a test set; It should be noted that wavelet packet decomposition is developed on the basis of conventional wavelet transform. Wavelet transform refers to the use of a series of wavelet basis functions to approximate the original signal, that is, to decompose the original signal into wavelet signals in different frequency domains, so that the spectral feature signals hidden in the time domain can be expressed. Figure 3 is a three-layer wavelet decomposition. Among them, S is the original signal, A represents the low-frequency signal, which can be called the scale space, and D represents the high-frequency signal, which is called the wavelet space. In this way, the original signal is decomposed into the superposition of low-frequency and high-frequency signals. Wavelet packet decomposition further decomposes each wavelet space based on wavelet decomposition, thus retaining more high-frequency signal features.

[0052] The wavelet packet decomposition method is used to decompose the fault current waveform into three layers of wavelet packets. Considering the time-frequency resolution and computational complexity, the sample data is decomposed into three layers of wavelet packets by the third-order Daubechies (db3) wavelet function. 24,000 points of the waveform data of the three cycles of the three-phase input current on the grid side are sampled as the original data for decomposition. The grid voltage fluctuation is set to be about ±5%, and the voltage-stabilizing capacitor fluctuation on the AC output side is about ±2.5%. 100 working conditions are determined, and 100 groups of data are measured for each operating state type, a total of 1,600 groups of data, each group of data contains a total of 24 eigenvalues ​​of the three-phase current, and the data eigenvalues ​​are exported. The circuit mode of each fault type is set separately, and then the code is run to obtain the data. Finally, the data is summarized and the feature types are labeled as 1-16, a total of 16 eigenvector groups, to form a data set. The data set is divided into a training set and a test set in a ratio of 7:3.

[0053] Taking the three-layer wavelet packet decomposition of the fault current data of the DC charging pile charging module as an example, the sum of the squares of the signal decomposition coefficients in each frequency band is selected as the energy representation. Therefore, the energy of each frequency band after wavelet packet decomposition can be expressed as: ; Energy signature matrix .

[0054] For different types of faults in DC charging piles, the energy in different frequency bands in the fault current waveform data is different. Therefore, the E vector can be used as the feature vector extracted from it and used as the input data of the convolutional neural network for training and recognition.

[0055] Step 3: Establish a CNN neural network, use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; CNN stands for Convolutional Neural Network, which means convolutional neural network. Convolutional neural network can be regarded as a black box. Its input is a matrix array representing the recognition object, and then convolution operation is performed on it (such as Figure 4 is a two-dimensional convolution process with a step size of 1), and the output is the result of object recognition.

[0056] A CNN network usually includes the following layers: input layer (for data input), convolution layer (using convolution operation to extract features of the object to be identified), activation layer (for nonlinear fitting, including activation functions such as relu, sigmoid, tanh, etc.), pooling layer (reducing or simplifying the extracted feature matrix), fully connected layer (transposing and flattening the data matrix), Softmax layer (outputting classification probability) and output layer, among which structures such as convolution, activation and pooling layers can appear multiple times in some combination when the network is constructed.

[0057] The convolution kernel in the convolution layer is used to extract features from the training set data, and the extracted feature matrix is ​​activated and pooled and the matrix is ​​flattened to calculate the loss function. The loss function is a certain difference between the output result and the actual value, which is a function of the weight and bias parameters. Assume that the data after wavelet packet decomposition feature extraction and transposition flattening is and the corresponding fault type label set As the target, the absolute value of the difference between the output and the label value is the loss function of the error L As shown below:

[0058] In the formula, For input and weight and bias The output product of By calculating the gradient of the loss function, we obtain new weight and bias parameters that make the gradient decrease. The weight parameters are updated according to the following formula:

[0059] Where ε is the learning rate.

[0060] The loss L is then recalculated using the new parameters. The above process is called forward propagation and back propagation. Repeating this process gradually reduces the loss function value, and finally obtains the desired output result. This is the model training process. After the model training is completed, the DC charging pile test set data is input into the model to test and judge the model's diagnostic accuracy.

[0061] When training the model, the data must first be normalized between 0 and 1 to improve the convergence speed of the model. This application constructs a network structure with two layers of convolution, where the two convolutional layer types are 2×1×1×16 (1 channel, 16 convolution kernels) and 2×1×16×32 (16 channels, 32 convolution kernels), and a batch normalization layer and a relu activation layer are used after each convolutional layer. Using the Adam algorithm, set the initial learning rate, the batch size of the training data to 100, the total number of iterations to 2000, and the learning descent rate to 0.5.

[0062] Step 4: Use the feature data of the test set to test the trained CNN model; Input the test set data into the trained model and output the sample recognition and classification results. For example, input a sample data into the model and output the probability of all class labels. Then, the denormalization method can be used to extract the label with the highest classification probability, and then compare it with the actual label of the sample data to determine whether it is correct or not, and output the diagnosis result. Figure 5 shown.

[0063] Step 5: When the test results determine that the set accuracy threshold is reached, the charging pile fault diagnosis model is output; The fault diagnosis effect after model training with different numbers of convolution layers, where 32 convolution kernels are set in each layer, and batch normalization and activation layers are set after each convolution layer. Six data sets of size 480 are randomly selected from 1600 sets of data. When the training parameters of the CNN model convolution network are the same, the models with different numbers of convolution layers are observed for the F7, F9, and F10 fault prediction results of the same randomly selected test set. Figure 6 This is the changing trend of the average values ​​of fault recognition accuracy and overall diagnosis rate of F7, F9, and F10 with the number of convolutional layers.

[0064] As the number of convolutional network layers increases, the overall fault diagnosis rate of the CNN model is also increasing. When the number of layers is 4, the accuracy is the highest, which is 97.1%. When the number of convolutional layers reaches 3, the recognition accuracy of the F10 fault type is 100%. Overall, the recognition rate of the CNN network model for F7 type faults also increases with the increase in the number of layers, and the recognition rate of F9 type faults reaches a maximum of 87.0% at around 4 layers. In summary, the CNN model with 4 convolutional layers has the best fault recognition effect in this example. The prediction effect of the randomly selected test set is shown in the following example. Figure 7 shown.

[0065] Step 6: Use the charging pile fault diagnosis model to perform fault diagnosis of the DC charging pile.

[0066] like Figure 2 As shown, an embodiment of the present invention provides a charging pile fault diagnosis device based on convolutional neural network and wavelet packet decomposition, including a fault data acquisition module, a wavelet packet decomposition processing module, a training module, a testing module, a model output module and an execution diagnosis module; A fault data acquisition module is used to obtain the fault current waveform of the charging pile; A wavelet packet decomposition processing module is used to decompose the fault current waveform using the wavelet packet decomposition method, extract the energy feature vector to generate a feature data set; and divide the generated data set into a training set and a test set; The training module is used to establish a CNN neural network and use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; The testing module is used to test the trained CNN model using the feature data of the test set; A model output module is used to output a charging pile fault diagnosis model when the test result determines that the set accuracy threshold has been reached; The execution diagnosis module is used to perform fault diagnosis of the DC charging pile using the charging pile fault diagnosis model.

[0067] As a further limitation of the technical solution of the present invention, the step of obtaining the charging pile fault current waveform includes: Build the DC charging pile charging module circuit; Set various operating status fault types to perform fault simulation on the charging module circuit and obtain the fault current waveform of the charging pile charging module.

[0068] As a further limitation of the technical solution of the present invention, the device also includes a label classification module, which classifies labels for fault current waveform data according to the fault type and generates a fault type label set.

[0069] As a further limitation of the technical solution of the present invention, the wavelet packet decomposition processing module includes a decomposition unit, a frequency band eigenvalue calculation unit, a characteristic matrix generation unit and a division unit; A decomposition unit, used for performing three-layer wavelet packet decomposition on the fault current waveform of the DC charging pile charging module; A frequency band characteristic value calculation unit, used for selecting the sum of squares of signal decomposition coefficients in each frequency band as the energy characteristic value of the frequency band; A characteristic matrix generating unit, used for generating an energy characteristic matrix, i.e., a characteristic data set, from the energy characteristic values ​​of all the frequency bands after decomposition; The partitioning unit is used to divide the feature data set into a training set and a test set according to a set ratio.

[0070] As a further limitation of the technical solution of the present invention, the energy characteristic value of each frequency band after the three-layer wavelet packet decomposition is: ; Energy signature matrix .

[0071] As a further limitation of the technical solution of the present invention, the training module includes a model building unit and a training execution unit; A model building unit, used to build a CNN neural network; The training execution unit is used to take the training set data of different fault current characteristics as input, use the convolution kernel in the convolution layer to extract features of the training set data, activate and pool the extracted feature matrix and flatten the matrix; take the flattened data and the corresponding fault type label set as the target, and use the absolute value of the difference between the output and the label value as the error loss function; calculate the gradient of the loss function to obtain new weights and bias parameters that make the gradient decrease; use the new parameters to recalculate the loss function value until the loss function value reaches the set requirements to complete the training and generate the trained CNN model.

[0072] As a further limitation of the technical solution of the present invention, the loss function

[0073] In the formula, For input and weight and bias The output product of The weight parameters are updated as follows:

[0074] Where ε is the learning rate.

[0075] The embodiment of the present invention also provides an electronic device, which includes: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The communication bus can be used for information transmission between the electronic device and the sensor. The processor can call the logic instructions in the memory to execute the following method: Step 1: Obtain the charging pile fault current waveform; Step 2: Decompose the fault current waveform using the wavelet packet decomposition method, extract the energy feature vector to generate a feature data set; and divide the generated data set into a training set and a test set; Step 3: Establish a CNN neural network, use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; Step 4: Use the feature data of the test set to test the trained CNN model; Step 5: When it is judged according to the test results that the set accuracy threshold is reached, output the charging pile fault diagnosis model; Step 6: Use the charging pile fault diagnosis model to perform fault diagnosis of the DC charging pile.

[0076] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0077] An embodiment of the present invention provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to execute the method provided by the above method embodiment, for example, including: Step 1: Obtain a charging pile fault current waveform; Step 2: Decompose the fault current waveform using a wavelet packet decomposition method, extract energy feature vectors to generate a feature data set; and divide the generated data set into a training set and a test set; Step 3: Establish a CNN neural network, and use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; Step 4: Use the feature data of the test set to test the trained CNN model; Step 5: When it is determined according to the test results that the set accuracy threshold is reached, output a charging pile fault diagnosis model; Step 6: Use the charging pile fault diagnosis model to perform fault diagnosis of the DC charging pile.

[0078] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0079] An embodiment of the charging pile fault diagnosis device based on convolutional neural network and wavelet packet decomposition provided in an embodiment of the present invention belongs to the same inventive concept as the charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition in the above-mentioned embodiments. For details not described in detail in the embodiment of the charging pile fault diagnosis device based on convolutional neural network and wavelet packet decomposition, reference can be made to the embodiment of the charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition.

[0080] The charging pile fault diagnosis device based on convolutional neural network and wavelet packet decomposition is a unit and algorithm step of each example described in combination with the embodiments disclosed in this article, which can be implemented by electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0081] Those skilled in the art will appreciate that various aspects of the charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition can be implemented as a system, method or program product. Therefore, various aspects of the present disclosure can be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which can be collectively referred to as "circuit", "module" or "system".

[0082] Although the present invention has been described in detail with reference to the accompanying drawings and in combination with the preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and these modifications or substitutions shall be within the scope of the present invention. Any person of ordinary skill in the art may easily think of changes or substitutions within the technical scope disclosed by the present invention, and these shall be within the scope of protection of the present invention.

Claims

1. A charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition, characterized in that: The steps include: Obtain the charging pile fault current waveform; Decompose the fault current waveform using wavelet packet decomposition method, extract energy feature vectors to generate feature data sets; and divide the generated data sets into training sets and test sets; Establish a CNN neural network, use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; Use the feature data of the test set to test the trained CNN model; When the test results determine that the set accuracy threshold is reached, the charging pile fault diagnosis model is output; The charging pile fault diagnosis model is used to perform fault diagnosis of DC charging piles.

2. The charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition according to claim 1 is characterized in that: The steps of obtaining the charging pile fault current waveform include: Build the DC charging pile charging module circuit; Set various operating status fault types to perform fault simulation on the charging module circuit and obtain the fault current waveform of the charging pile charging module.

3. The charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition according to claim 2 is characterized in that: The fault current waveform is decomposed using wavelet packet decomposition method, and the energy feature vector is extracted to generate a feature data set; The steps of dividing the generated dataset into training and test sets include: The fault current waveform data is classified into labels according to the fault type to generate a fault type label set.

4. The charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition according to claim 3 is characterized in that: The fault current waveform is decomposed using wavelet packet decomposition method, and the energy feature vector is extracted to generate a feature data set; The steps of dividing the generated data set into training set and test set include: Perform three-layer wavelet packet decomposition on the fault current waveform of the DC charging pile charging module; Select the sum of squares of the signal decomposition coefficients in each frequency band as the energy characteristic value of the frequency band; The energy eigenvalues ​​of all decomposed frequency bands are used to generate an energy feature matrix, i.e., a feature data set; The feature data set is divided into a training set and a test set according to a set ratio.

5. The charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition according to claim 4 is characterized in that: The energy eigenvalue of each frequency band after three-layer wavelet packet decomposition is: ； Energy signature matrix .

6. The charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition according to claim 5 is characterized in that: The steps of establishing a CNN neural network and using the feature data of the training set as input to train the CNN neural network and generate a trained CNN model include: Build a CNN neural network; The training set data with different fault current characteristics are used as input, and the convolution kernel in the convolution layer is used to extract the features of the training set data. The extracted feature matrix is ​​activated and pooled, and the matrix is ​​flattened. The flattened data and the corresponding fault type label set are used as the target, and the absolute value of the difference between the output and the label value is used as the loss function of the error; By calculating the gradient of the loss function, new weight and bias parameters are obtained that make the gradient decrease; The loss function value is recalculated using the new parameters until the loss function value reaches the set requirements to complete the training and generate the trained CNN model.

7. The charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition according to claim 6 is characterized in that: Loss Function In the formula, For input and weight and bias The output product of The weight parameters are updated as follows: Where ε is the learning rate.

8. A charging pile fault diagnosis device based on convolutional neural network and wavelet packet decomposition, characterized in that: It includes a fault data acquisition module, a wavelet packet decomposition processing module, a training module, a testing module, a model output module and an execution diagnosis module; A fault data acquisition module is used to obtain the fault current waveform of the charging pile; A wavelet packet decomposition processing module is used to decompose the fault current waveform using the wavelet packet decomposition method, extract the energy feature vector and generate a feature data set; And divide the generated data set into training set and test set; The training module is used to establish a CNN neural network and use the feature data of the training set as input to train the CNN neural network to generate a trained CNN model; The testing module is used to test the trained CNN model using the feature data of the test set; A model output module is used to output a charging pile fault diagnosis model when the test result determines that the set accuracy threshold has been reached; The execution diagnosis module is used to perform fault diagnosis of the DC charging pile using the charging pile fault diagnosis model.

9. An electronic device, characterized in that: The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; the memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the charging pile fault diagnosis method based on convolutional neural network and wavelet packet decomposition as described in any one of claims 1 to 7.

Images