IGBT life prediction method and device based on CNN-LSTM model

Through the IGBT lifetime prediction method based on the CNN-LSTM model, the feature extraction is automatically extracted by turning off peak voltage data, and the problem of low accuracy of IGBT lifetime prediction in the prior art is solved, achieving higher prediction accuracy and confidence.

CN118917188APending Publication Date: 2024-11-08WUHAN UNIV
View PDF 0 Cites 11 Cited by

Patent Information

Application Number
CN202410972518.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing IGBT lifetime prediction methods based on statistical regression and machine learning are not very accurate and fail to fully utilize the relevant features of IGBT for accurate prediction.

Method used

Using the IGBT lifetime prediction method based on the CNN-LSTM model, the shutdown peak voltage data in the historical time period is obtained for pre-processing and input into the pre-trained CNN-LSTM model to automatically extract features and predict the remaining service life of the IGBT .

Benefits of technology

It improves the accuracy and confidence of IGBT life prediction, and can more accurately predict the aging state of IGBT, which is of great significance to the operation of IGBT power modules and systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917188A_ABST
    Figure CN118917188A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of power electronic reliability, and particularly discloses an IGBT service life prediction method and device based on a CNN-LST M model, and the method comprises the steps: obtaining a turn-off peak voltage data sequence of an IGBT in a historical time period; preprocessing the turn-off peak voltage data sequence, inputting the preprocessed turn-off peak voltage data sequence to a pre-trained CNN-LSTM model, and obtaining a turn-off peak voltage prediction sequence of the IGBT; and predicting the remaining service life of the IGBT based on the turn-off peak voltage prediction sequence. In an IGBT life prediction scene, the turn-off peak voltage is selected as the aging characteristic parameter of the IGBT, and the CNN-LSTM hybrid network is selected as the IGBT life prediction network, so that characteristic extraction and deep mining of parameter characteristics can be automatically performed from the acquired turn-off peak voltage data sequence, and the IGBT aging state with high confidence is obtained. The residual service life of the IGBT is predicted, and the method is of great significance to the operation of the IGBT power module and the whole system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of power electronics reliability technology, and more specifically, to an IGBT life prediction method and device based on a CNN-LSTM model. Background Art

[0002] Insulated Gate Bipolar Transistor (IGBT) has been widely used in power electronic application systems such as power transmission and high-speed rail systems due to its superior performance. During long-term operation, IGBT is subjected to electrical, thermal and mechanical stresses and wear, which seriously affects its service life. Accurately predicting the life of IGBT modules, formulating maintenance and repair plans in a timely manner, and replacing power modules in advance when they are damaged and about to fail can avoid a series of safety accidents and economic losses caused by the failure of the power modules themselves.

[0003] At present, the life prediction methods of IGBT can be divided into two categories: prediction methods based on mathematical and physical models and prediction methods based on data-driven. In the prediction methods based on mathematical and physical models, preset mathematical models such as exponential equations are usually used to fit the data. However, this method often relies on experience and has low reliability under complex operating conditions. In order to improve the accuracy of model prediction, additional parameters are added, which makes the prediction model more complicated. In contrast, prediction methods based on data-driven, including adaptive fuzzy systems, neural networks, support vector machines, Gaussian process regression, particle filters, etc., have the advantages of simple process, small amount of calculation, and no need to consider the complex internal mechanism of power devices. With the development of artificial intelligence today, people expect to make new breakthroughs in constructing prediction methods using artificial intelligence methods. With the help of machine learning algorithms, fitting equations that flexibly characterize the relationship between monitoring data and predictions can be learned or updated as degradation develops.

[0004] Data-driven prediction methods are mostly concentrated in applications with junction temperature and collector voltage as parameters. In some technical solutions, based on the measurement results of thermal imaging equipment, the log-normal distribution is used to describe the life distribution of IGBTs. It is believed that as the junction temperature difference and the average junction temperature increase, the life of the IGBT module gradually decreases; optical fiber temperature sensors are used to monitor the chip temperature and observe the internal degradation failure of power semiconductor devices under power cycle tests in real time; however, the acquisition of junction temperature is either directly measured at high cost, or the thermal model calculation is complex and has errors.

[0005] In some technical solutions, characteristic parameters are extracted from the historical data of the collector-emitter voltage to predict the life of the IGBT module, making it possible to make more accurate predictions through historical data collected by sensors without understanding the internal failure mechanism of the IGBT. For example, Zhao et al. established a neural network (NN) and adaptive neuro-fuzzy inference system (ANFIS) model to predict the remaining service life of the IGBT. For example, Ahsan et al. proposed a reliability assessment fusion algorithm to predict the life of the IGBT using the long-short-term memory network (LSTM) algorithm. For example, Lu and Christo used the LSTM prediction model to predict the early performance degradation trend. For example, Hu et al. used the collector on-state voltage, Miller platform voltage and Miller platform time as input to establish an artificial neural network to predict the life of the IGBT module.

[0006] It can be seen that most of the research in related technologies is based on traditional statistical regression and machine learning to predict the life of IGBT, and the relevant features that can be collected by IGBT are not fully used for prediction. Therefore, a model that can automatically extract features and more accurately predict the life of IGBT is needed. Summary of the invention

[0007] In view of the defects existing in the related technology, the embodiment of the present application provides an IGBT life prediction method and device based on the CNN-LSTM model, aiming to solve the problem of low prediction accuracy of IGBT life prediction based on traditional statistical regression and machine learning.

[0008] In a first aspect, an embodiment of the present application provides an IGBT life prediction method based on a CNN-LSTM model, comprising:

[0009] Obtain the IGBT turn-off peak voltage data sequence within the historical time period;

[0010] Preprocess the turn-off peak voltage data sequence and input it into the pre-trained CNN-LSTM model to obtain the turn-off peak voltage prediction sequence of the IGBT;

[0011] Based on the turn-off peak voltage prediction sequence, the remaining service life of the IGBT is predicted.

[0012] In a second aspect, the embodiment of the present application further provides an IGBT life prediction device based on a CNN-LSTM model, comprising:

[0013] The first acquisition module is used to acquire the IGBT turn-off peak voltage data sequence within a historical time period;

[0014] The second acquisition module is used to pre-process the turn-off peak voltage data sequence, and input the pre-processed data sequence into a pre-trained CNN-LSTM model to obtain the turn-off peak voltage prediction sequence of the IGBT;

[0015] The prediction module is used to predict the remaining service life of the IGBT based on the turn-off peak voltage prediction sequence.

[0016] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0017] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0018] In a fifth aspect, an embodiment of the present application further provides a computer program product, which, when executed on a processor, enables the processor to execute the method described in the first aspect or any possible implementation of the first aspect.

[0019] The IGBT life prediction method and device based on the CNN-LSTM model provided in the embodiment of the present application, in the IGBT life prediction scenario, selects the turn-off peak voltage as the aging characteristic parameter of the IGBT, and selects the CNN-LSTM hybrid network as the IGBT life prediction network. It can automatically extract features from the collected turn-off peak voltage data sequence, deeply mine parameter features, and thus obtain the IGBT aging state with high confidence, and predict the remaining service life of the IGBT, which is of great significance to the operation of the IGBT power module and the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present application or related technologies, the following is a brief introduction to the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 This is one of the flow charts of the IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application;

[0022] Figure 2 It is a schematic diagram of voltage change at the moment of IGBT shutdown provided by an embodiment of the present application;

[0023] Figure 3 is a schematic diagram of the change of the turn-off peak voltage during the aging process of the IGBT provided in an embodiment of the present application;

[0024] Figure 4 It is a schematic diagram of the framework of the CNN-LSTM model provided in the embodiment of the present application;

[0025] Figure 5 This is the second flow chart of the IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application;

[0026] Figure 6 It is a structural schematic diagram of an IGBT life prediction device based on a CNN-LSTM model provided in an embodiment of the present application;

[0027] Figure 7 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0029] Figure 1 This is one of the flow charts of the IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application, such as Figure 1 As shown, the method at least includes the following steps:

[0030] S101, obtaining a turn-off peak voltage data sequence of the IGBT in a historical time period.

[0031] S102, preprocessing the turn-off peak voltage data sequence, and inputting the preprocessed data sequence into a pre-trained CNN-LSTM model to obtain a turn-off peak voltage prediction sequence of the IGBT.

[0032] S103 . Predict the remaining service life of the IGBT based on the turn-off peak voltage prediction sequence.

[0033] Specifically, in the relevant solutions for IGBT life prediction, IGBT aging / degradation data will be used, including collector voltage, gate-emitter voltage, collector-emitter voltage, turn-off peak voltage, case temperature, junction temperature, etc.

[0034] Figure 2Schematic diagram of voltage change at the moment of IGBT shutdown provided by the embodiment of the present application, such as Figure 2 As shown in the figure, at the moment of turning off, the IGBT will form a peak voltage, namely the turn-off peak voltage V ce-peak . Figure 3 Schematic diagram of the change of the turn-off peak voltage during the aging process of the IGBT provided in the embodiment of the present application. Figure 3 As shown, the IGBT turn-off peak voltage V ce-peak As the IGBT ages, the resistance between the collector and emitter increases due to the voids in the solder layer and the fall of the bonding wires. This increases the collector-emitter saturation voltage. At the same time, the resistance component in the parasitic inductance load increases, the flowing current decreases, and the IGBT's turn-off peak voltage V ce-peak Therefore, in the embodiment of the present application, the turn-off peak voltage V ce-peak As the IGBT aging characteristic parameter.

[0035] Convolutional Neural Networks (CNN) is a type of feedforward neural network with deep structure and convolutional calculation. By using convolutional calculation in at least one layer of the network to replace the general matrix multiplication operation, its operation speed is greatly improved. It generally includes input layer, convolution layer, pooling layer, fully connected layer and output layer. The alternating use of convolution layer and pooling layer can effectively extract local features of data and reduce the dimension of local features. Weight sharing in CNN can effectively reduce the number of networks and reduce the complexity of the model.

[0036] LSTM is a time recurrent neural network designed to solve the long-term dependency problem of recurrent neural networks. All recurrent neural networks have repeated neural network modules and are chain-like. LSTM can remember values ​​of indefinite time lengths. There is a gate structure in the block that determines whether the input is important enough to be remembered and whether it can be output. By adding the gate structure, the gradient vanishing and gradient exploding problems of recurrent neural networks are effectively solved.

[0037] In the IGBT life prediction scenario, the LSTM network is generally selected in related technologies because the LSTM network can effectively capture the time dependency of sequence data. However, this ignores the spatial features in the sequence data of the IGBT aging characteristic parameters. The structure of the CNN-LSTM model combines the characteristics of the CNN convolutional neural network and the long short-term memory network LSTM, and is designed to process data that contains both spatial features and time dependencies. The CNN part is used to extract the spatial features of the input data, and the LSTM part is used to process the time dependency in the feature sequence. The two parts are connected through a fully connected layer to achieve the conversion from spatial features to time dependencies. Compared with single networks such as LSTM networks, CNN networks, and GRU (Gated Recurrent Unit, GRU) networks, the prediction results of the CNN-LSTM network have higher stability and confidence.

[0038] The IGBT turn-off peak voltage data sequence is obtained in the historical time period, and the sequence data is preprocessed and input into the CNN-LSTM model to obtain the turn-off peak voltage prediction sequence output by the CNN-LSTM model, thereby predicting the remaining service life of the IGBT. The input of the CNN-LSTM model can be multiple turn-off peak voltage data sequences. The input data is feature extracted through CNN, and then the information to be retained and discarded is determined through LSTM, and finally the turn-off peak voltage prediction sequence of the IGBT is output.

[0039] The IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application selects the turn-off peak voltage as the aging characteristic parameter of the IGBT in the IGBT life prediction scenario, and selects the CNN-LSTM hybrid network as the IGBT life prediction network. It can automatically extract features from the collected turn-off peak voltage data sequence, deeply mine parameter features, and thus obtain the IGBT aging status with high confidence, and predict the remaining service life of the IGBT, which is of great significance to the operation of the IGBT power module and the overall system.

[0040] In some embodiments, the CNN-LSTM model is trained based on the following steps:

[0041] Extracting the initial data of the turn-off peak voltage from the IGBT aging data as the IGBT aging characteristic parameter;

[0042] Preprocess the initial turn-off peak voltage data to generate IGBT aging training samples;

[0043] Divide the IGBT aging training samples into training set, validation set and test set;

[0044] The CNN-LSTM model is trained based on the training set. During the training process, the performance of the CNN-LSTM model is evaluated based on the validation set. After the training is completed, the performance of the trained CNN-LSTM model is evaluated based on the test set to obtain a trained CNN-LSTM model.

[0045] Specifically, the CNN-LSTM model needs to be pre-trained before IGBT life prediction. The initial data of the turn-off peak voltage is extracted from the IGBT aging data as the IGBT aging characteristic parameter; the initial turn-off peak voltage data is pre-processed to generate IGBT aging training samples. The turn-off peak voltage is selected as the IGBT aging characteristic parameter, and the corresponding training samples are created.

[0046] The IGBT aging training samples are divided into training set, validation set and test set, which are independent of each other. The training set is used to train the model, that is, the CNN-LSTM model updates the weights and model parameters by learning the data in the training set; during the training process, the model will continuously adjust the model parameters according to the difference between the training results on the training set and the true label to minimize the prediction error; the training samples contain a large amount of sample data to ensure that the model can learn the inherent laws and characteristics of the IGBT turn-off peak voltage data.

[0047] The validation set is used to evaluate the performance of the CNN-LSTM model during training and to help select the best hyperparameter configuration. Evaluating the performance of the model through the validation set during training helps identify overfitting or underfitting and adjust the model structure or hyperparameters accordingly. At the same time, the validation set can also be used to select the maximum number of model iterations to avoid over-optimization on the training set.

[0048] The test set is used to evaluate the performance of the CNN-LSTM model after training is completed, and to evaluate the generalization ability of the trained CNN-LSTM model on new data. The validation set is frequently used during the training process, which leads to an overestimation of the model's true performance. Therefore, an independent test set is used to more accurately evaluate the model. Applying the trained CNN-LSTM model to the test set and calculating the performance indicators of the model on new data can more objectively reflect the generalization ability and actual application effect of the model.

[0049] The IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application ensures that the CNN-LSTM model is fully optimized during the training process by reasonably dividing and using the training set, validation set and test set, thereby improving the accuracy and robustness of the model, thereby being able to demonstrate good generalization ability in the IGBT life prediction process.

[0050] In some embodiments, pre-processing comprises:

[0051] A filtering method is used to smooth the turn-off peak voltage data to obtain smoothed turn-off peak voltage data;

[0052] The split window width is set to split the smoothed turn-off peak voltage data to obtain split turn-off peak voltage data;

[0053] The segmented off-peak voltage data are reversibly normalized.

[0054] Specifically, the training and application of the CNN-LSTM model involve the preprocessing of the IGBT turn-off peak voltage data obtained within a certain period of time. For the application of known models in different application scenarios, the preprocessing method will directly affect the effect of the model.

[0055] In the IGBT life prediction scenario, the input of the CNN-LSTM model is the IGBT turn-off peak voltage data within a certain period of time. In order to reduce the overall computational complexity and improve the accuracy and confidence of IGBT life prediction, the preprocessing method specifically includes: ce-peak The data is filtered, denoised, segmented and normalized.

[0056] First, a filtering method is used to filter V ce-peak The data is smoothed (such as removing abnormal data, sampling at different time intervals, etc.) to obtain the smoothed V ce-peak Data. The noise in the data will interfere with the authenticity and accuracy of the data, resulting in inaccurate IGBT life prediction structure. Filtering and noise reduction can effectively improve data quality and ensure the reliability of IGBT life prediction results.

[0057] After filtering and noise reduction, V ce-peak The data is segmented, and the segmentation window width is set to L. The segmented V ce-peak The input of data in the CNN-LSTM model is represented as:

[0058]

[0059] The output is represented as:

[0060] Y i T = {V cepeak,i+1 ,V cepeak,i+2 ,…,V cepeak,m-L+i},i=1,2,…,L

[0061] Furthermore, in order to reduce the overall computational complexity and the amount of computational effort, as well as to ce-peakThe data were analyzed and compared. ce-peak The data is normalized and V ce-peak The data is mapped to the interval [0, 1]. The selection of normalization processing methods needs to adapt to different data types and scenarios. In the IGBT life prediction scenario, considering that the final output sequence of the CNN-LSTM model needs to be restored to the original data, a reversible normalization method is used: Min-Max normalization.

[0062] In some embodiments, S103 specifically includes:

[0063] After the turn-off peak voltage prediction sequence is denormalized, the remaining service life of the IGBT is predicted.

[0064] Specifically, in the training and application process of the CNN-LSTM model, the preprocessing stage performs reversible normalization on the turn-off peak voltage data sequence before inputting it into the CNN-LSTM model. The data in the model output sequence is in the interval [0, 1]. Therefore, it is necessary to perform denormalization on the turn-off peak voltage prediction sequence output by the model to obtain a prediction sequence in the original data range, so as to judge the aging status of the IGBT and predict the remaining service life of the IGBT.

[0065] The IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application sets a preprocessing method for the IGBT turn-off peak voltage data in the IGBT life prediction scenario, including: filtering and denoising, segmentation and reversible normalization processing, to improve data quality and model prediction accuracy while ensuring that the predicted data output by the model can be restored to the original data.

[0066] In some embodiments, the CNN-LSTM model includes CNN, LSTM and a fully connected layer; CNN extracts the local spatial features of the input preprocessed turn-off peak voltage data sequence through a convolutional layer; LSTM takes the feature sequence extracted by CNN as input, controls the input data through a gate structure, processes the timing relationship of the turn-off peak voltage sequence data and makes a prediction; the fully connected layer is used to output the turn-off peak voltage data prediction sequence.

[0067] Specifically, the CNN-LSTM model includes: CNN, LSTM and fully connected layers. Among them, the CNN part extracts the local spatial features (i.e., aging features) of the input preprocessed turn-off peak voltage data sequence through the convolution layer. Each convolution layer contains multiple convolution kernels / receptive fields. These convolution kernels slide on the input sequence and calculate the dot product to obtain the feature sequence. The LSTM part is mainly used for extracting the time features of sequence processing. It can remember historical information and predict future outputs based on current input and historical information. The gate unit in LSTM can perform information control to avoid the gradient vanishing problem that occurs when RNN processes long sequences. The long-term dependency of the turn-off peak voltage data sequence is captured through the gate unit and memory unit.

[0068] The connection between CNN and LSTM can be achieved through the fully connected layer, which converts the features extracted by CNN into a feature sequence that can be processed by LSTM. At the same time, the data output by LSTM is passed to the fully connected layer, which converts the output of LSTM into the final prediction result.

[0069] Figure 4 : is a schematic diagram of the framework of the CNN-LSTM model provided in the embodiment of the present application, such as Figure 4 As shown, CNN includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a Flatten layer, and an output layer.

[0070] Among them, the convolution layer is used to extract the aging characteristics of the input turn-off peak voltage data sequence. The aging feature extraction formula is:

[0071]

[0072] Among them, M j Represents the input turn-off peak voltage data sequence, is the input of the lth layer, is the output of layer l, is the convolution kernel of the lth layer, is the bias parameter of the lth layer, and f() represents the activation function;

[0073] The pooling layer is used to reduce the dimension of the aging feature vector extracted by the convolution layer, while retaining important spatial information, which can reduce the computational complexity of the model and prevent overfitting. The pooling layer uses maximum pooling, and the maximum pooling operation satisfies the following formula:

[0074]

[0075] Where W is the width of the pooling area, is the value of the tth neuron in the ith eigenvector in the lth layer.

[0076] The aging feature vector after dimensionality reduction is input to LSTM. The LSTM part includes two LSTM units. Each LSTM unit contains a memory unit and a gating unit. The gating unit can selectively decide which information to pass. The gating unit includes three gates: input gate, forget gate and output gate. These three gates are used to control the flow of information and the state update of the memory unit. The input gate, forget gate and output gate meet the following conditions:

[0077] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0078] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0079]

[0080] θ t =σ(W o ·[h t-1 ,x t ]+b o )

[0081] h t =o t *tanh(C t )

[0082] Among them, W f , W i , W C , W o is the weight matrix, b f 、b i 、b C 、b o is the corresponding offset, * is the dot product, f t is the retention degree value, i t is the added degree value of the current state, h t-1 is the output of the previous state, x t is the input of the current state, is the intermediate state, C t-1 , C t are the memory state of the previous state and the current state respectively, o t is the output degree value.

[0083] The LSTM part takes the aging feature sequence extracted by CNN as input, processes these sequence data through a loop structure, captures the long-term dependencies in the sequence data, and predicts future outputs based on current input and historical information.

[0084] The connection part between CNN and LSTM is one or more fully connected layers, which convert the features extracted by CNN into a feature sequence that can be processed by LSTM and convert the output of LSTM into the final turn-off peak voltage prediction sequence.

[0085] The IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application selects the CNN-LSTM model for IGBT life prediction. The CNN-LSTM model combines CNN and LSTM to achieve simultaneous processing of spatial features and time dependencies.

[0086] In some embodiments, the training process of the CNN-LSTM model further includes:

[0087] Train the network in small batches through the Adam algorithm solver, backpropagate the training error, and update the model parameters layer by layer;

[0088] Specify the initial learning rate for training and reduce the learning rate in stages after iterations;

[0089] Specify regularized training options to add regularized training options to the damage function.

[0090] Specifically, during the training process of the CNN-LSTM model, in order to ensure that the model can be trained as expected, specify the training options. Specify the following training options:

[0091] The Adam algorithm solver is used to train a small batch of networks, back-propagate the training error, and update the model parameters layer by layer. The optimizer refers to the optimization algorithm used to update the model weights, such as stochastic gradient descent, Adam, etc.

[0092] Specify the initial learning rate for training, and then reduce the learning rate in stages after iterations. The initial learning rate determines the step size of weight updates at the beginning of training. Usually, a higher learning rate will result in a wider convergence at the beginning of training, but it may also cause unstable training, while a lower learning rate will result in a slow training process. Therefore, you can specify the initial learning rate at the beginning of training, and then reduce the learning rate in stages after iterations, while taking into account the convergence speed and stability of model training.

[0093] Specify regularized training options and add regularized training options to the damage function. For example, add a certain weighted L2 norm to the damage function. Specifying regularized training options can prevent overfitting in training.

[0094] The IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application specifies training options such as optimizer, initial learning rate and regularization during the model training process to ensure that the CNN-LSTM model can be trained as expected, thereby improving the subsequent IGBT life prediction accuracy.

[0095] The technical solution provided in the embodiment of the present application is further illustrated below by using a specific example.

[0096] Figure 5 This is the second flow chart of the IGBT life prediction method based on the CNN-LSTM model provided in the embodiment of the present application, such as Figure 5 As shown, the method comprises the following steps:

[0097] Step 1: Data preprocessing. The original V extracted from the IGBT aging data is ce-peak After filtering, denoising, segmentation and normalization of the data, the IGBT aging training samples are obtained, and the IGBT aging training samples are divided into training set, validation set and test set.

[0098] Step 2: CNN-LSTM model training. The CNN-LSTM model is trained using the training set. The CNN network in the CNN-LSTM model extracts V ce-peak The local spatial features of the data, LSTM network extracts V ce-peak The temporal characteristics of the data. During the model training process, the validation set is used to evaluate the model performance and adjust the model parameters. When the model training times reaches the preset number of iterations, the training is completed.

[0099] Step 3: CNN-LSTM model testing. After the model training is completed, the model performance is further evaluated using the test set, and performance evaluation indicators such as mean square error (MAE), mean absolute error (MAPE), mean square error (MSE), and root mean square error (RMSE) are calculated to evaluate the model performance. Finally, the trained CNN-LSTM model is used to predict the life of the IGBT.

[0100] Figure 6 is a schematic diagram of the structure of an IGBT life prediction device based on a CNN-LSTM model provided in an embodiment of the present application, such as Figure 6 As shown, the device at least includes:

[0101] The first acquisition module 601 is used to acquire the IGBT turn-off peak voltage data sequence in a historical time period;

[0102] The second acquisition module 602 is used to pre-process the turn-off peak voltage data sequence, and input the pre-processed data sequence into a pre-trained CNN-LSTM model to obtain the turn-off peak voltage prediction sequence of the IGBT;

[0103] The prediction module 603 is used to predict the remaining service life of the IGBT based on the turn-off peak voltage prediction sequence.

[0104] In some embodiments, the CNN-LSTM model is trained based on the following steps:

[0105] Extracting the initial data of the turn-off peak voltage from the IGBT aging data as the IGBT aging characteristic parameter;

[0106] Preprocess the initial turn-off peak voltage data to generate IGBT aging training samples;

[0107] Divide the IGBT aging training samples into training set, validation set and test set;

[0108] The CNN-LSTM model is trained based on the training set. During the training process, the performance of the CNN-LSTM model is evaluated based on the validation set. After the training is completed, the performance of the trained CNN-LSTM model is evaluated based on the test set to obtain a trained CNN-LSTM model.

[0109] In some embodiments, pre-processing comprises:

[0110] A filtering method is used to smooth the turn-off peak voltage data to obtain smoothed turn-off peak voltage data;

[0111] The split window width is set to split the smoothed turn-off peak voltage data to obtain split turn-off peak voltage data;

[0112] The segmented off-peak voltage data are reversibly normalized.

[0113] In some embodiments, the prediction module 603 is specifically used to:

[0114] After the turn-off peak voltage prediction sequence is denormalized, the remaining service life of the IGBT is predicted.

[0115] In some embodiments, the training process of the CNN-LSTM model further includes:

[0116] Train the network in small batches through the Adam algorithm solver, backpropagate the training error, and update the model parameters layer by layer;

[0117] Specify the initial learning rate for training and reduce the learning rate in stages after iterations;

[0118] Specify regularized training options to add regularized training options to the damage function.

[0119] In some embodiments, the CNN-LSTM model includes CNN, LSTM and a fully connected layer; CNN extracts the local spatial features of the input preprocessed turn-off peak voltage data sequence through a convolutional layer; LSTM takes the feature sequence extracted by CNN as input, and captures the long-term dependency of the turn-off peak voltage data sequence through memory units and gating units; the fully connected layer is used to output the turn-off peak voltage data prediction sequence.

[0120] In some embodiments, the CNN includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a Flatten layer, and an output layer;

[0121] Among them, the convolution layer is used to extract the aging characteristics of the input turn-off peak voltage data sequence. The aging feature extraction formula is:

[0122]

[0123] Among them, M j is the input turn-off peak voltage data sequence, is the input of the lth layer, is the output of layer l, is the convolution kernel of the lth layer, is the bias parameter of the lth layer, f() is the activation function,

[0124] The pooling layer is used to reduce the dimension of the aging feature vector extracted by the convolution layer. The maximum pooling operation satisfies the following formula:

[0125]

[0126] Where W is the width of the pooling area, is the value of the tth neuron in the ith eigenvector in the lth layer.

[0127] It should be understood that the above-mentioned device is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method, which will not be repeated here.

[0128] Based on the method in the above embodiment, an embodiment of the present application provides an electronic device. The device may include: at least one memory for storing programs and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the above embodiment.

[0129] Figure 7 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, such as Figure 7 As shown, the electronic device may include: a processor (processor) 701, a communication interface (Communications Interface) 702, a memory (memory) 703 and a communication bus 704, wherein the processor 701, the communication interface 702, and the memory 703 communicate with each other through the communication bus 704. The processor 701 can call the software instructions in the memory 703 to execute the method described in the above embodiment.

[0130] In addition, the logic instructions in the above-mentioned memory 703 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology 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 a number of 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 application.

[0131] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0132] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0133] It is understandable that the processor in the embodiment of the present application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0134] The method steps in the embodiments of the present application can be implemented by hardware, or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (Random Access Memory, RAM), flash memory, read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM, PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electrically erasable programmable read-only memory (Electrically EPROM, EEPROM), registers, hard disks, mobile hard disks, CD-ROMs or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0135] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integration. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0136] It should be understood that the various numerical numbers involved in the embodiments of the present application are only used for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0137] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for predicting IGBT life based on CNN-LSTM model, characterized in that: include: Obtain the IGBT turn-off peak voltage data sequence within the historical time period; Preprocessing the turn-off peak voltage data sequence, and inputting the preprocessed data sequence into a pre-trained CNN-LSTM model to obtain a turn-off peak voltage prediction sequence of the IGBT; Based on the turn-off peak voltage prediction sequence, the remaining service life of the IGBT is predicted.

2. The IGBT life prediction method based on the CNN-LSTM model according to claim 1 is characterized in that: The CNN-LSTM model is trained based on the following steps: Extracting the initial data of the turn-off peak voltage from the IGBT aging data as the IGBT aging characteristic parameter; Preprocessing the initial turn-off peak voltage data to generate an IGBT aging training sample; Dividing the IGBT aging training samples into a training set, a validation set and a test set; The CNN-LSTM model is trained based on the training set, the performance of the CNN-LSTM model is evaluated based on the validation set during the training process, and the performance of the trained CNN-LSTM model is evaluated based on the test set after the training is completed, so as to obtain a trained CNN-LSTM model.

3. The IGBT life prediction method based on the CNN-LSTM model according to claim 1 or 2, characterized in that: The preprocessing comprises: A filtering method is used to smooth the turn-off peak voltage data to obtain smoothed turn-off peak voltage data; The split window width is set to split the smoothed turn-off peak voltage data to obtain split turn-off peak voltage data; The segmented off-peak voltage data are reversibly normalized.

4. The IGBT life prediction method based on the CNN-LSTM model according to claim 1, characterized in that: The predicting the remaining service life of the IGBT based on the turn-off peak voltage prediction sequence includes: After performing a denormalization process on the turn-off peak voltage prediction sequence, the remaining service life of the IGBT is predicted.

5. The IGBT life prediction method based on the CNN-LSTM model according to claim 2 is characterized in that: The training process of the CNN-LSTM model also includes: Train the network in small batches through the Adam algorithm solver, backpropagate the training error, and update the model parameters layer by layer; Specify the initial learning rate for training and reduce the learning rate in stages after iterations; Specify regularized training options to add regularized training options to the damage function.

6. The IGBT life prediction method based on the CNN-LSTM model according to claim 1 is characterized in that: The CNN-LSTM model includes CNN, LSTM and a fully connected layer; the CNN extracts the local spatial features of the preprocessed turn-off peak voltage data sequence input through a convolution layer; the LSTM takes the feature sequence extracted by the CNN as input, and captures the long-term dependency of the turn-off peak voltage data sequence through a memory unit and a gating unit; the fully connected layer is used to output a turn-off peak voltage data prediction sequence.

7. The IGBT life prediction method based on the CNN-LSTM model according to claim 6 is characterized in that: The CNN includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a Flatten layer and an output layer; The convolution layer is used to extract the aging characteristics of the input off peak voltage data sequence, and the aging characteristic extraction formula is: Among them, M j is the input turn-off peak voltage data sequence, is the input of the lth layer, is the output of layer l, is the convolution kernel of the lth layer, is the bias parameter of the lth layer, f() is the activation function, The pooling layer is used to reduce the dimension of the aging feature vector extracted by the convolution layer. The maximum pooling operation satisfies the following formula: Where W is the width of the pooling area, is the value of the tth neuron in the ith eigenvector in the lth layer.

8. An IGBT life prediction device based on CNN-LSTM model, characterized in that: include: The first acquisition module is used to acquire the IGBT turn-off peak voltage data sequence within a historical time period; A second acquisition module is used to preprocess the turn-off peak voltage data sequence, and input the preprocessed data sequence into a pre-trained CNN-LSTM model to obtain a turn-off peak voltage prediction sequence of the IGBT; A prediction module is used to predict the remaining service life of the IGBT based on the turn-off peak voltage prediction sequence.

9. An electronic device / image signal generator / network device / transmitter / terminal / base station / industrial computer, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program runs on a processor, the processor is caused to execute the method according to any one of claims 1 to 7.

Citation Information

Cited By

  • Electro-hydrogen coupling system equipment life prediction method, equipment and medium

    CN119415896A

  • Method for predicting residual life of high-voltage power module based on optimized neural network and multi-parameter feature fusion and related device

    CN119578255A

  • High-voltage power module remaining life prediction method and related device based on optimized neural network and multi-parameter feature fusion

    CN119578255B

  • Signal identification method and system of optical fiber sensing network

    CN119743195A

  • IGBT life prediction method based on Gaussian process regression and LSTM neural network

    CN120541393A