IGBT remaining service life prediction method, system, device, medium and product

Abstract

The invention discloses an IGBT residual service life prediction method, system and device, a medium and a product, and relates to the field of residual service life prediction, and the method comprises the steps: obtaining a saturation voltage drop value of an IGBT at a current moment; inputting the saturation voltage drop value at the current moment into a trained residual service life network prediction model for prediction to obtain a saturation voltage drop value of the IGBT at the next moment; wherein the residual service life network prediction model comprises a convolutional neural network and a converter model, the converter model comprises an embedded layer, an encoder and a decoder, the output of the embedded layer is the input of the convolutional neural network, and the output of the convolutional neural network is the input of the encoder; and determining the residual service life of the IGBT according to the saturation voltage drop value at the next moment. According to the invention, the prediction precision of the residual service life of the IGBT can be improved.

Description

Technical Field

[0001] The present application relates to the field of remaining useful life prediction, and in particular to an IGBT remaining useful life prediction method, system, device, medium and product. Background Art

[0002] Because the insulated-gate bipolar transistor (IGBT) is affected by electrical and thermal stress when working under long-term high-voltage and high-current conditions, it may cause aging over time and seriously affect its maximum service life. Due to the different thermal expansion coefficients of various layers of the IGBT chip, aging usually occurs in the aluminum bonding wire and solder layer, which leads to an increase in semiconductor loss and package thermal resistance, and ultimately manifests as the saturation voltage drop of the IGBT gradually increasing with the aging process. Due to the easy measurement of the saturation voltage drop, it has been identified as an effective online aging feature.

[0003] Based on the saturation voltage drop as the aging characteristic of IGBT, the deep learning model has better learning and adaptive capabilities than traditional data-driven models, and has advantages in life prediction. Currently, in the field of IGBT remaining useful life (RUL) prediction, the long short-term memory network (LSTM) is a mainstream time series prediction model and is often used as a benchmark model. However, LSTM has the problems of gradient vanishing and gradient explosion, which seriously affect the prediction effect and lead to low prediction accuracy of the remaining useful life of IGBT. Summary of the invention

[0004] The purpose of this application is to provide a method, system, device, medium and product for predicting the remaining useful life of an IGBT, which can improve the prediction accuracy of the remaining useful life of the IGBT.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for predicting the remaining useful life of an IGBT, the method comprising:

[0007] Get the saturation voltage drop value of IGBT at the current moment;

[0008] The saturation voltage drop value at the current moment is input into the trained remaining service life network prediction model for prediction, so as to obtain the saturation voltage drop value of the IGBT at the next moment; wherein the remaining service life network prediction model includes a convolutional neural network and a converter model, the converter model includes an embedding layer, an encoder and a decoder, the output of the embedding layer is the input of the convolutional neural network, and the output of the convolutional neural network is the input of the encoder;

[0009] The remaining service life of the IGBT is determined according to the saturation voltage drop value at the next moment.

[0010] Furthermore, the convolutional neural network is a four-layer one-dimensional convolutional layer.

[0011] Furthermore, the training process of the remaining useful life network prediction model includes:

[0012] Construct a training set, which includes the IGBT measured from time t1 to time t k+1 The actual value of the saturation voltage drop at the moment;

[0013] t i-a to i The actual value of the saturation pressure drop at time t is input into the remaining service life network prediction model to obtain i+1 The predicted value of the saturation voltage drop at the moment; where a is the time step, the value range is 1≤a≤k, t i is the i-th moment, the value range of i is 1≤i≤k, and i is greater than a;

[0014] According to t i+1 The predicted value of the saturation voltage drop at time t i+1 The actual value of the saturated voltage drop at the moment is used to calculate the loss, and according to the loss, the parameters of the remaining useful life network prediction model are optimized until the loss reaches a preset range, thereby obtaining a trained remaining useful life network prediction model.

[0015] Furthermore, the loss is calculated as follows:

[0016]

[0017] in, For the tth i+1 The actual value of the saturation voltage drop at the moment, y i+1 For the tth i+1 The saturation pressure drop prediction value at time t, k is the number of actual saturation pressure drop values.

[0018] Furthermore, a training set is constructed, including:

[0019] During the accelerated aging test of the IGBT, when the IGBT is in the aging stage, the actual values ​​of the saturation voltage drop corresponding to three working conditions are collected respectively to obtain multiple data sets; wherein the three working conditions are extreme working conditions, severe working conditions and standard working conditions;

[0020] Multiple data sets are preprocessed and downsampled by a preset multiple, and any data set that has been downsampled is used as a training set.

[0021] Furthermore, multiple data sets are preprocessed, including

[0022] For any data set, the abnormal data in the data set is deleted, and the data set after the abnormal data is deleted is subjected to noise reduction processing.

[0023] In a second aspect, the present application provides an IGBT remaining service life prediction system, the IGBT remaining service life prediction system comprising:

[0024] An acquisition module is used to obtain the saturation voltage drop value of the IGBT at the current moment;

[0025] A prediction module, used for inputting the saturation voltage drop value at the current moment into the trained remaining service life network prediction model for prediction, so as to obtain the saturation voltage drop value of the IGBT at the next moment; wherein the remaining service life network prediction model comprises a convolutional neural network and a converter model, the converter model comprises an embedding layer, an encoder and a decoder, the output of the embedding layer is the input of the convolutional neural network, and the output of the convolutional neural network is the input of the encoder;

[0026] The life determination module determines the remaining life of the IGBT according to the saturation voltage drop value at the next moment.

[0027] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-described IGBT remaining useful life prediction methods.

[0028] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the above-mentioned methods for predicting the remaining useful life of an IGBT.

[0029] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned IGBT remaining useful life prediction methods.

[0030] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0031] The present application provides a method, system, device, medium and product for predicting the remaining useful life of an IGBT. The remaining useful life of the IGBT can be obtained by inputting the saturation voltage drop value of the IGBT at the current moment into a remaining useful life network prediction model to predict the remaining useful life of the IGBT. The remaining useful life network prediction model embeds a convolutional neural network into a converter model. The convolutional neural network collects local features through convolution operations. The converter model avoids the problem of gradient explosion or gradient disappearance by adopting a self-attention mechanism, and can effectively extract the global features of long sequence data, making full use of local and global features, significantly improving the generalization ability and prediction accuracy, thereby improving the prediction accuracy of the remaining useful life of the IGBT. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 This is an application environment diagram of a method for predicting the remaining useful life of an IGBT in one embodiment of the present application;

[0034] Figure 2 A schematic flow chart of a method for predicting the remaining useful life of an IGBT provided in one embodiment of the present application;

[0035] Figure 3 A curve diagram of aging cycles of different data sets provided in an embodiment of the present application;

[0036] Figure 4 A structural diagram framework diagram of a remaining useful life network prediction model provided in an embodiment of the present application;

[0037] Figure 5 A schematic diagram of functional modules of a system for predicting the remaining useful life of an IGBT provided in one embodiment of the present application;

[0038] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0040] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0041] The IGBT remaining service life prediction method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the saturation voltage drop value of the IGBT at the current moment to the server 104, and the server 104 receives the saturation voltage drop value of the IGBT at the current moment. For the saturation voltage drop value of the IGBT at the current moment, the server 104 predicts the saturation voltage drop value of the IGBT at the next moment based on the saturation voltage drop value of the IGBT at the current moment, and obtains the remaining service life of the IGBT according to the saturation voltage drop value of the IGBT at the next moment. The server 104 can feedback the remaining service life of the IGBT obtained to the terminal 102. In addition, in some embodiments, the IGBT remaining useful life prediction method can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform the remaining useful life prediction processing of the IGBT according to the IGBT remaining useful life prediction method, or the server 104 can obtain the saturation voltage drop value of the IGBT at the current moment from the data storage system, and perform the IGBT remaining useful life prediction processing according to the saturation voltage drop value of the IGBT at the current moment.

[0042] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, and tablet computers. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0043] In an exemplary embodiment, Figure 2 As shown, a method for predicting the remaining useful life of an IGBT is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1The server 104 in the example is used as an example to illustrate, including the following steps 201 to 203. Among them:

[0044] Step 201, obtaining the saturation voltage drop value of the IGBT at the current moment.

[0045] Step 202: input the saturation voltage drop value at the current moment into the trained remaining service life network prediction model for prediction, and obtain the saturation voltage drop value of the IGBT at the next moment; wherein the remaining service life network prediction model includes a convolutional neural network and a converter model, and the converter model includes an embedding layer, an encoder and a decoder, the output of the embedding layer is the input of the convolutional neural network, and the output of the convolutional neural network is the input of the encoder.

[0046] Convolutional Neural Network (CNN) collects local features through convolution operations. The specific operation is to gradually extract local features through three layers of one-dimensional convolution layers, and then reduce the number of channels through a 1*1 convolution layer. The advantage of replacing the pooling layer with a 1*1 convolution layer is that it can greatly reduce the amount of calculation compared to the pooling layer.

[0047] The transformer model includes an embedding layer, an encoder, and a decoder. The transformer model is used to enhance the global perception of the model. The encoder in the transformer model contains multiple encoder layers, and the decoder contains decoder layers corresponding to the encoder layers. Each encoder layer contains a feedforward network and a self-attention layer. The multiple self-attention layers form the multi-head attention in the transformer model. Figure 4 As shown in Figure 1, the local features extracted by the convolutional neural network are input to the encoder. In the Transformer model, position encoding is used to describe the position of the data point in the sequence, thereby assigning a unique representation between 0 and 1 to each position. Next, the encoder layer is responsible for generating the encoding, which contains information about which parts of the input are related to each other. The output of the last encoder layer is the input of the decoder. Finally, the decoder receives the output of the encoder and the output of the decoder at the previous time step to generate an output sequence, which is input to the flat layer. The structural parameters of the remaining service life network prediction model are shown in Table 1.

[0048] Table 1

[0049] Types of layers Activate Layer Output size enter (32，1) Embedding layer (32，16) One-dimensional convolutional layer ReLU (16，14) One-dimensional convolutional layer ReLU (32，12) One-dimensional convolutional layer ReLU (64，10) One-dimensional convolutional layer (16，10) Multi-head attention (16，10) Flat Layer (160) Dense Layer (1)

[0050] Among them, the linear rectification function (ReLU) is a commonly used activation function in artificial neural networks, also known as the rectified linear unit.

[0051] The flatten layer combines all multi-dimensional inputs into one dimension so that all data can be efficiently passed to each neuron of the model. The neurons in the dense layer receive the output from each neuron in the previous layer. This application uses the dropout layer to randomly ignore a group of neurons to prevent the model from overfitting.

[0052] Step 203: determining the remaining service life of the IGBT according to the saturation voltage drop value at the next moment.

[0053] By implementing the above steps 201 to 203, the present application embeds the convolutional neural network into the converter model to improve the prediction accuracy of the remaining useful life of the IGBT.

[0054] In another exemplary embodiment of the present application, the training process of the above-mentioned remaining useful life network prediction model includes steps 301 to 303:

[0055] Step 301: construct a training set, wherein the training set includes the IGBT measured from time t1 to time t k+1 The actual value of the saturation voltage drop at the moment.

[0056] The IGBT was controlled to undergo accelerated aging tests when the junction temperature fluctuated at 80k but the average junction temperature was different. Specifically, the IGBT module model selected for the test was Infineon FF100R12KS4, with a maximum operating temperature of 150°C and a rated current of 100A. Accelerated aging tests were conducted under three working conditions, and the failure form of bond wire breakage occurred in all of them. IGBT1, IGBT2, and IGBT3 are data sets collected under three working conditions, corresponding to standard working conditions, severe working conditions, and extreme working conditions, respectively. The data sets under different test conditions and the corresponding test results are shown in Table 2. ΔT j is the junction temperature difference, T jm is the average junction temperature, and N is the number of aging cycles when the IGBT fails.

[0057] Table 2

[0058] Dataset <![CDATA[ΔT j / ℃]]> <![CDATA[T j_max / ℃]]> <![CDATA[T j_min / ℃]]> <![CDATA[T jm / ℃]]> N / times IGBT1 80 150 70 110 129080 IGBT2 80 160 80 120 104878 IGBT3 80 170 90 130 100998

[0059] When a cycle starts, I is a large current, the IGBT is turned on, and the junction temperature T j Continue to rise to the set maximum value T j_max ; At this time, the current will change from large to small, the IGBT will be turned off, and the junction temperature T j Will drop to the set minimum value T j_min Repeat this process until VCE (ON)If the VCE exceeds 5% of the initial value, the IGBT is considered to have failed. In this application, the device aging is divided into two stages, namely the healthy stage and the aging stage. (ON) The first time a step jump occurs, the IGBT device enters the aging stage. In the aging stage, the actual value of the saturation voltage drop is collected. Specifically, the actual values ​​of the saturation voltage drop corresponding to the three working conditions are recorded during each cycle, and a time series is formed according to the time sequence of the records to form multiple data sets.

[0060] Preprocessing is performed on multiple data sets, specifically including deleting abnormal data in the data sets, and performing noise reduction processing on the data sets after deleting the abnormal data.

[0061] Abnormal data is mostly caused by abnormal cycle temperature when the test is not within the set range. For example, when the test bench is started, the junction temperature starts to rise from the ambient temperature, so the maximum junction temperature cannot reach the predetermined maximum temperature; another case is when VCE (ON) When a step jump occurs, this is because the aging and breakage of the bonding wires cause the resistance of the IGBT module to increase. In order to keep the junction temperature difference constant, the aging platform will adjust the current to meet this situation. Therefore, some discrete values ​​will appear in the current regulation process.

[0062] After removing the abnormal data, in order to avoid the influence of environmental noise on the data, the data set after deleting the abnormal data is processed by Kalman filtering to achieve noise reduction.

[0063] Taking IGBT1 as an example, the preprocessed data is as follows Figure 3 Observation Figure 3 It can be seen that VCE (ON) It grows slowly in the early stage, but after about 118,000 power cycles, the VCE increases due to the breakage of the bonding wire. (ON) It is growing exponentially.

[0064] Specifically, the preprocessed data set is downsampled by a preset multiple, and any data set that has been downsampled is used as a training set. The original data is still used in the test set. The original data is the data that is only preprocessed and not downsampled. In this application, the data that is downsampled 10 times is used as the training set to improve the training efficiency of the model.

[0065] This application uses the original data as the training set after 10 times downsampling instead of directly using the original data as the training set. Taking the single-step prediction of IGBT1 as an example, the epoch is 1, and the training time is 0.214 and 1.682 respectively. Therefore, this method can greatly reduce the training time while ensuring the training effect.

[0066] The present application uses data obtained under any of the three working conditions to train the remaining service life network prediction model, and uses data obtained under another working condition for testing, thereby realizing cross-condition prediction of the remaining service life of the IGBT.

[0067] Step 302: t i-a to i The actual value of the saturation pressure drop at time t is input into the remaining service life network prediction model to obtain i+1 The predicted value of the saturation voltage drop at the moment; where a is the time step, the value range is 1≤a≤k, t i is the i-th moment, the value range of i is 1≤i≤k, and i is greater than a. In this application, a is 32, the optimizer is Adam (Adaptive Moment Estimation, Adam), the learning rate is 0.01, and the epoch is 100.

[0068] Step 303, according to t i+1 The predicted value of the saturation voltage drop at time t i+1 The actual value of the saturated voltage drop at the moment is used to calculate the loss, and according to the loss, the parameters of the remaining useful life network prediction model are optimized until the loss reaches a preset range, thereby obtaining a trained remaining useful life network prediction model.

[0069] During model training, the two style variables of CNN and Transformer are coupled together using mean-square error (MSE) loss. This fusion process can significantly enhance the global perception ability of local variables and the local details of global attributes.

[0070] The calculation formula for the mean square error loss is:

[0071]

[0072] in, For the tth i+1 The actual value of the saturation voltage drop at the moment, y i+1 For the tth i+1 The saturation pressure drop prediction value at time t, k is the number of actual saturation pressure drop values.

[0073] This application compares the prediction effects of Transformer, CNN-LSTM and multi-layer LSTM. Hyperparameter setting: The CNN-LSTM network consists of three convolutional layers (the convolution kernel is 2, the number of channels is 8, 16, and 32) and three LSTM layers (the number of hidden layer neurons is 128). The LSTM network consists of three LSTM layers (the number of hidden layer neurons is 128).

[0074] In order to intuitively compare the prediction performance of each model, the correlation coefficient (R 2 ), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) evaluation indicators were compared. 2 The calculation formulas of MAE and RMSE are:

[0075]

[0076]

[0077] Among them, x j is the actual value, is the predicted value, For x j The average value of , m is the number of test samples.

[0078] Taking single-step prediction as an example, this application uses the IGBT2 and IGBT3 data that have been downsampled 10 times as the training set and validation set, and then uses the original data in IGBT1 as the test set. Table 3 shows the single-step prediction results of each model. The remaining service life network prediction model proposed in this application has the highest degree of fit with the real data. Compared with Transformer, the addition of the CNN module enhances the model's ability to extract local features. CNN-Transformer in R 2 The average improvement was 2.74%, and the MAE and RMSE decreased by 60.70% and 59.47% respectively. 2 The average improvement was 3.03%, and the MAE and RMSE decreased by 61.76% and 66.54% respectively. It can be seen from Table 4 and Table 5 that Table 4 is the two-step prediction result and Table 5 is the three-step prediction result. In the two-step prediction and three-step prediction, the remaining useful life network prediction model proposed in this application is better than other models. Among them, the single-step prediction is to use the trained remaining useful life network prediction model in this application, and the input is t i-a to i The actual value of the saturation voltage drop at time t i+1 When performing two-step prediction and three-step prediction, the network parameters of the trained remaining service life network prediction model in this application are modified respectively. After the modification, the inputs of the two-step prediction and the three-step prediction remain unchanged as t i-a to i The actual value of the saturation voltage drop at the moment; the two-step prediction output becomes two, namely t i+1 Time and t i+2 The saturation voltage drop prediction value at time t; the three-step prediction output becomes three, namely ti+1 Time, t i+2 Time and t i+3 The predicted value of saturation pressure drop at time.

[0079] Further analysis found that the four models have better prediction effects on the remaining useful life using the IGBT1 data set than those using IGBT2 and IGBT3, indicating that the migration effect from severe working conditions to standard working conditions is better. The analysis shows that the degradation information is more obvious in a short period of time under severe working conditions, the signal-to-noise ratio is higher, the noise impact is smaller, and the model can learn more effective features.

[0080] Table 3

[0081]

[0082] Table 4

[0083]

[0084] Table 5

[0085]

[0086] Based on the same inventive concept, the embodiment of the present application also provides an IGBT remaining useful life prediction system for implementing the above-mentioned IGBT remaining useful life prediction method. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the one or more IGBT remaining useful life prediction system embodiments provided below can refer to the limitations of the IGBT remaining useful life prediction method above, and will not be repeated here.

[0087] In an exemplary embodiment, Figure 5 As shown, a prediction system for the remaining useful life of an IGBT is provided, comprising:

[0088] An acquisition module 21 is used to acquire the saturation voltage drop value of the IGBT at the current moment;

[0089] A prediction module 22, used for inputting the saturation voltage drop value at the current moment into the trained remaining service life network prediction model for prediction, to obtain the saturation voltage drop value of the IGBT at the next moment; wherein the remaining service life network prediction model includes a convolutional neural network and a converter model, the converter model includes an embedding layer, an encoder and a decoder, the output of the embedding layer is the input of the convolutional neural network, and the output of the convolutional neural network is the input of the encoder;

[0090] The life determination module 23 determines the remaining life of the IGBT according to the saturation voltage drop value at the next moment.

[0091] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for a method for predicting the remaining service life of an IGBT. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting the remaining service life of an IGBT is implemented.

[0092] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0093] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0094] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0095] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0096] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0097] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0098] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0099] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for predicting the remaining useful life of an IGBT, characterized in that: The IGBT remaining service life prediction method comprises: Get the saturation voltage drop value of IGBT at the current moment; The saturation voltage drop value at the current moment is input into the trained remaining service life network prediction model for prediction, so as to obtain the saturation voltage drop value of the IGBT at the next moment; wherein the remaining service life network prediction model includes a convolutional neural network and a converter model, the converter model includes an embedding layer, an encoder and a decoder, the output of the embedding layer is the input of the convolutional neural network, and the output of the convolutional neural network is the input of the encoder; The remaining service life of the IGBT is determined according to the saturation voltage drop value at the next moment.

2. The IGBT remaining service life prediction method according to claim 1, characterized in that: The convolutional neural network has four one-dimensional convolutional layers.

3. The IGBT remaining service life prediction method according to claim 1, characterized in that: The training process of the remaining useful life network prediction model includes: Construct a training set, which includes the IGBT measured from time t1 to time t k+1 The actual value of saturation pressure drop at the moment; t i-a to i The actual value of the saturation pressure drop at time t is input into the remaining service life network prediction model to obtain i+1 The predicted value of the saturation voltage drop at the time; where a is the time step, the value range is 1≤a≤k, k is the number of actual values ​​of the saturation voltage drop, t i is the i-th moment, the value range of i is 1≤i≤k, and i is greater than a; According to t i+1 The predicted value of the saturation voltage drop at time t i+1 The actual value of the saturated voltage drop at the moment is used to calculate the loss, and according to the loss, the parameters of the remaining useful life network prediction model are optimized until the loss reaches a preset range, thereby obtaining a trained remaining useful life network prediction model.

4. The IGBT remaining service life prediction method according to claim 3, characterized in that: The loss is calculated as follows: Where L is the loss, For the tth i+1 The actual value of the saturation voltage drop at the moment, y i+1 For the tth i+1 The saturation pressure drop prediction value at time t, k is the number of actual saturation pressure drop values.

5. The IGBT remaining service life prediction method according to claim 3, characterized in that: Construct a training set, including: During the accelerated aging test of the IGBT, when the IGBT is in the aging stage, the actual values ​​of the saturation voltage drop corresponding to three working conditions are collected respectively to obtain multiple data sets; wherein the three working conditions are extreme working conditions, severe working conditions and standard working conditions; Multiple data sets are preprocessed and downsampled by a preset multiple, and any data set after the downsampling process is used as a training set.

6. The IGBT remaining service life prediction method according to claim 5, characterized in that: Preprocess multiple data sets, including For any data set, the abnormal data in the data set is deleted, and the data set after the abnormal data is deleted is subjected to noise reduction processing.

7. An IGBT remaining useful life prediction system, characterized in that: The IGBT remaining service life prediction system comprises: An acquisition module is used to obtain the saturation voltage drop value of the IGBT at the current moment; A prediction module, used for inputting the saturation voltage drop value at the current moment into the trained remaining service life network prediction model for prediction, so as to obtain the saturation voltage drop value of the IGBT at the next moment; wherein the remaining service life network prediction model comprises a convolutional neural network and a converter model, the converter model comprises an embedding layer, an encoder and a decoder, the output of the embedding layer is the input of the convolutional neural network, and the output of the convolutional neural network is the input of the encoder; The life determination module determines the remaining life of the IGBT according to the saturation voltage drop value at the next moment.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the IGBT remaining useful life prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the remaining useful life of an IGBT according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the IGBT remaining service life prediction method described in any one of claims 1 to 6 are implemented.

Images