IGBT life prediction method and device, equipment and storage medium

Abstract

The invention provides an IGBT life prediction method and device, equipment and a storage medium. The IGBT life prediction method comprises the steps of obtaining historical voltage parameters of an insulated gate bipolar transistor (IGBT) to construct a sample set; the voltage parameter is a voltage parameter related to the service life; constructing a BP neural network for predicting the service life of the IGBT; taking the initial weight and the bias parameter of the BP neural network as population individuals, taking a mean square error function of the BP neural network as a fitness function, and performing global search through a pelican optimization algorithm to obtain an initial optimal weight and a bias parameter; training a BP neural network based on the initial optimal weight, the bias parameter and the sample set, and iterating forward propagation and back propagation processes to obtain a trained BP neural network; and inputting a voltage parameter of an IGBT to be predicted into the trained BP neural network, and predicting the service life of the IGBT through the trained BP neural network. The method can improve the accuracy of IGBT life prediction.

Description

Technical Field

[0001] The present invention belongs to the field of life prediction of power electronic equipment, and in particular relates to a life prediction method, device, equipment and storage medium for an IGBT. Background Art

[0002] With the rapid development of power electronics technology, the reliability of IGBTs (Insulated Gate Bipolar Transistors), core components in power electronics, has a significant impact on overall system performance. Predicting the life of IGBTs is crucial for improving the reliability of power electronics and reducing maintenance costs.

[0003] Currently, IGBT lifespan prediction is mainly based on machine learning algorithms such as BP (Back Propagation) neural networks. Traditional BP neural networks are highly dependent on data quality, sensitive to the selection of initial values ​​and the distribution of training data, and the training process may fall into a local optimal solution, resulting in an inability to provide accurate prediction results. Summary of the Invention

[0004] In order to solve the problem of low accuracy in IGBT life prediction, the present invention provides an IGBT life prediction method, device, equipment and storage medium.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for predicting the life of an IGBT, comprising:

[0007] Acquire historical voltage parameters of an insulated gate bipolar transistor (IGBT) to construct a sample set; the voltage parameters are voltage parameters related to the life of the IGBT;

[0008] Construct a BP neural network for predicting IGBT lifespan;

[0009] The initial weights and bias parameters of the BP neural network are used as population individuals, and the mean square error function of the BP neural network is used as the fitness function. The Pelican optimization algorithm is used to perform a global search to obtain the initial optimal weights and bias parameters.

[0010] The BP neural network is trained based on the initial optimal weight and bias parameters and the sample set, and the forward propagation and back propagation processes are iterated to obtain the trained BP neural network;

[0011] The voltage parameters of the IGBT to be predicted are input into the trained BP neural network, and the life of the IGBT is predicted by the trained BP neural network.

[0012] Optionally, a global search is performed using the Pelican optimization algorithm to obtain the initial optimal weight and bias parameters, including:

[0013] Randomly generate multiple initial weight and bias parameter combinations within the global range of weight and bias parameters as the initial population of the Pelican optimization algorithm;

[0014] For each combination in the population, the prediction value and mean square error value are calculated through forward propagation of the BP neural network, and the mean square error value is used as the fitness value for fitness evaluation;

[0015] Update the population according to the update mechanism of the Pelican optimization algorithm;

[0016] The process of iterative fitness evaluation and population update is repeated until the maximum number of iterations is reached, and the combination with the smallest fitness value is used as the initial optimal weight and bias parameters.

[0017] Optionally, the BP neural network is trained based on the initial optimal weight and bias parameters and the sample set, and the forward propagation and back propagation processes are iterated to obtain a trained BP neural network, including:

[0018] Assign the initial optimal weights and bias parameters to the BP neural network, input the sample set into the BP neural network, calculate the output error through forward propagation, and optimize the weights and bias parameters in the network through back propagation to minimize the error;

[0019] The forward propagation and back propagation processes are iterated until the BP neural network reaches the preset performance standard on the test set.

[0020] Optionally, the BP neural network includes an input layer, a hidden layer and an output layer;

[0021] Among them, the input layer receives the voltage parameter, the hidden layer uses the S-shaped activation function to perform nonlinear mapping on the input voltage parameter, and the output layer outputs the predicted value, which is the collector-emitter voltage.

[0022] Optionally, after obtaining the historical voltage parameters of the IGBT, the method further includes:

[0023] Clean historical voltage parameters to identify and remove outliers;

[0024] Use filters to remove high-frequency noise and perform normalization.

[0025] The present invention also provides an IGBT life prediction device, comprising:

[0026] An acquisition module is used to acquire historical voltage parameters of an insulated gate bipolar transistor (IGBT) to construct a sample set; the voltage parameters are voltage parameters related to the life of the IGBT;

[0027] Model building module, used to build a BP neural network for predicting IGBT life;

[0028] The optimization module is used to use the initial weights and bias parameters of the BP neural network as population individuals, the mean square error function of the BP neural network as the fitness function, and perform a global search through the Pelican optimization algorithm to obtain the initial optimal weights and bias parameters;

[0029] The training module is used to train the BP neural network based on the initial optimal weight and bias parameters and the sample set, iterate the forward propagation and back propagation processes to obtain the trained BP neural network;

[0030] The prediction module is used to input the voltage parameters of the IGBT to be predicted into the trained BP neural network, and predict the life of the IGBT through the trained BP neural network.

[0031] The present invention also 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 implements the above-mentioned IGBT life prediction method when executing the program.

[0032] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the life prediction method of the IGBT is implemented.

[0033] The IGBT life prediction method provided by the present invention has the following beneficial effects:

[0034] The initial weights and bias parameters of the BP neural network are used as population individuals, and the mean square error function of the BP neural network is used as the fitness function. A global search is performed through the Pelican optimization algorithm to obtain the initial optimal weights and bias parameters; the BP neural network is trained based on the initial optimal weights, bias parameters and sample set, and the forward propagation and backpropagation processes are iterated to obtain a trained BP neural network; the present invention optimizes the initial weights and bias parameters of the BP neural network through the global search capability of the Pelican optimization algorithm to avoid the BP network from falling into a local optimal solution; for the initial optimal weights and bias parameters obtained by the Pelican optimization algorithm, a local search is performed again through the BP neural network, and the parameter settings of the BP network are dynamically adjusted during the entire training process to flexibly adapt to changes in input voltage parameters, thereby improving the accuracy and reliability of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] To more clearly illustrate the embodiments of the present invention and its design, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort.

[0036] Figure 1 A schematic flow chart of a method for predicting the life of an IGBT provided in an embodiment of the present invention;

[0037] Figure 2 A circuit platform design diagram for a DC power cycling test provided by an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of an interpolation process of a multiple interpolation method provided by an embodiment of the present invention;

[0039] Figure 4 A BP network structure diagram provided by an embodiment of the present invention;

[0040] Figure 5 A comparison chart of prediction results provided by an embodiment of the present invention; wherein, Figure 5 Figure (a) shows the prediction result of the original BP model. Figure 5 Figure (b) shows the prediction results of the POA-BP model;

[0041] Figure 6 A schematic structural diagram of an IGBT life prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solution of the present invention and to be able to implement it, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention.

[0043] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "axial", "radial", "circumferential", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the technical solution of the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0044] In addition, the terms "first" and "second" and the like are used for descriptive purposes only and are not to be understood as indicating or implying relative importance. In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In the description of the present invention, unless otherwise specified, "plurality" means two or more, which will not be described in detail here.

[0045] Before introducing the IGBT life prediction method provided by the present invention, the terminology involved in the embodiments of the present invention is first introduced.

[0046] Pelican Optimization Algorithm (POA): The POA algorithm solves the optimization problem by simulating the three stages of searching, foraging, and capturing during the hunting process of pelicans. In the search stage, the algorithm randomly generates a certain number of initial solutions (representing pelicans) and moves randomly in the solution space to explore potential solutions. After entering the foraging stage, the algorithm selects a part of the solutions based on the fitness of the solutions (that is, the objective function value) and performs local searches near the optimal solution to try to improve the quality of the solutions. Finally, in the capture stage, when the algorithm confirms that the optimal solution has been found or the stopping condition is met, it stops searching and outputs the optimal solution. The entire process solves the optimization problem by continuously iterating the search, foraging, and capturing steps. Due to its good search and local optimization capabilities, the POA algorithm is particularly suitable for solving high-dimensional, nonlinear, and multimodal optimization problems.

[0047] BP neural network: A neural network with high flexibility and nonlinear approximation capabilities. It calculates the output value through forward propagation, then performs backpropagation based on the error between the output value and the expected value. Using gradient descent, it adjusts the weights and biases in the network to minimize the error, thereby achieving accurate prediction or classification of input data.

[0048] Typically, traditional BP neural networks are often used as a method for predicting the life of IGBTs. However, this method is highly dependent on data quality and is sensitive to the selection of initial values ​​and the distribution of training data. In addition, the training process may fall into a local optimal solution rather than a global optimal solution, resulting in an inability to provide accurate prediction results.

[0049] In response to the above technical problems, the present invention provides an IGBT life prediction method that can utilize the global search capability of POA to optimize the initial weights and biases of the BP neural network, thereby preventing the BP network from falling into a local optimal solution and improving the network's generalization ability and prediction accuracy.

[0050] Figure 1 This is a schematic flow chart of a method for predicting the life of an IGBT provided by an embodiment of the present invention. Optionally, the method may be executed by a power electronic device.

[0051] Exemplarily, the method includes the following steps:

[0052] S1, obtaining historical voltage parameters of the insulated gate bipolar transistor (IGBT) to construct a sample set.

[0053] Among them, the voltage parameters are voltage parameters related to the life of the IGBT. Optionally, based on the accelerated aging test of the IGBT and according to the working principle and failure mechanism of the IGBT, key voltage parameters closely related to its life are selected for collection and acquisition.

[0054] It should be understood that the IGBT accelerated aging test mainly turns on and off the IGBT device under test periodically. During the on-time, the power loss causes the device under test to heat up rapidly; during the off-time, heat is dissipated through the heat dissipation device, causing the junction temperature to drop rapidly. The power cycle aging test is a heating method that actively applies excitation, so it can simulate the aging and failure process of IGBT in actual industrial field environments.

[0055] The accelerated aging test of the IGBT in this embodiment specifically includes:

[0056] Using DC power cycle test, by designing a simple circuit platform, such as Figure 2 As shown, Figure 2 This is a circuit platform design diagram for a DC power cycling test provided by an embodiment of the present invention. In the test, an IGBT is directly connected to a load and a constant DC source. The temperature of the module under test is increased by conduction loss. When the module junction temperature rises to the required maximum junction temperature T jmax When the power is cut off, the external cooling system starts to work and lowers the module temperature to the minimum junction temperature T jmin In addition, the key IGBT voltage parameters collected are the corresponding IGBT drain-source voltage when the constant current is 15A and the temperature is 150°C.

[0057] Furthermore, corresponding labels can be added to the voltage parameters, which are performance indicators related to the life of the IGBT, such as the collector-emitter saturation voltage, the collector-emitter turn-off voltage peak value, etc. The voltage parameter data set is divided into a sample set and a test set according to a certain preset ratio (for example, 7:1). The sample set is used to train the neural network, and the test set is used to test the specific performance of the neural network.

[0058] In order to improve the quality of the sample set and test set, after obtaining the historical voltage parameters of the IGBT, the voltage parameter data set is cleaned, denoised and normalized.

[0059] In order to prevent outliers from affecting the prediction accuracy of the model, the Laida criterion can be used to remove outliers in the data set. As one of the data detection methods, the Laida criterion has the advantages of simple calculation and wide adaptability.

[0060] In order to supplement the missing values ​​in the data, the multiple imputation method is used. The multiple imputation method refers to the interpolation method that uses the existing data in the incomplete data set to interpolate the missing values ​​multiple times, thereby generating multiple complete data sets. In order to solve the uncertainty problem of missing data, the multiple imputation method divides the interpolation process of the incomplete data set into three stages: interpolation, analysis, and merging. By performing multiple interpolations, several groups of interpolated values ​​are obtained. On this basis, multiple analysis results are solved and these results are effectively merged. The interpolation process is as follows: Figure 3 As shown, Figure 3 A schematic diagram of an interpolation process of a multiple interpolation method provided by an embodiment of the present invention. Figure 3 In the example, S represents the number of attributes in the dataset; A j (j=1,2,…,S) represents the name of the jth attribute; black boxes represent missing values; white boxes represent existing values; m represents the number of interpolation times.

[0061] Optionally, the voltage parameters can be denoised by a filter.

[0062] Normalization adjusts the numerical range of the data to a specific interval [0,1] through scaling operations to eliminate the impact of different feature dimensions or numerical range differences on data analysis, making all features numerically comparable, thereby improving the convergence speed of the algorithm and the accuracy of the model.

[0063] S2, construct a BP neural network for predicting IGBT life.

[0064] The BP neural network specifically consists of an input layer, a hidden layer, and an output layer. The input layer receives voltage parameters. The hidden layer uses an S-shaped activation function to perform nonlinear mapping on the input data. The output layer outputs the predicted collector-emitter voltage value. The topology of the BP neural network can be determined by selecting an appropriate number of hidden layer nodes based on an empirical formula. The formula is as follows:

[0065]

[0066] Where w is the number of hidden layer nodes, n is the number of input layer nodes, v is the number of output layer nodes, and z is a constant between 1 and 10.

[0067] When constructing the BP neural network structure, the parameters of the BP neural network are initialized at the same time, including the number of network layers, activation function, learning rate, and number of iterations.

[0068] It should be understood that increasing the number of hidden layers enhances nonlinear mapping capabilities and increases learning time. However, a single-hidden-layer BP neural network can approximate continuous functions within any closed interval. Therefore, a single-layer network is selected in this embodiment. Learning rates that are too low or too high can have negative consequences. In this embodiment, a learning rate of 0.01 is selected to ensure system stability while achieving the minimum error approximation.

[0069] It should be understood that the principle of BP neural network is to imitate the synaptic connection between neurons in the human brain to process data. The three network layers are responsible for receiving external signals, connecting input and output, and outputting predicted data. Figure 4 As shown, Figure 4 A diagram of a BP network structure provided by an embodiment of the present invention. The BP neural network first calculates the error between the actual output and the expected output. By continuously adjusting the parameter weights and bias parameters, the actual output value is continuously approached to the expected output value, stopping when the mean square error (MSE) is minimized. Its basic concept is generally divided into a forward information transmission part and an error backpropagation part. Specifically, using the gradient descent method, the weight deviation changes are continuously calculated along the descending direction of the error function slope. Each change is propagated layer by layer through backpropagation, gradually approximating the objective function.

[0070] S3, taking the initial weights and bias parameters of the BP neural network as population individuals, and the mean square error function of the BP neural network as the fitness function, a global search is performed through the Pelican optimization algorithm to obtain the initial optimal weights and bias parameters.

[0071] It should be understood that the Pelican Optimization Algorithm simulates the behavior and strategy of pelicans when attacking and hunting prey to update candidate solutions. The hunting strategy is simulated in two phases: exploration and exploitation. Its mathematical model is as follows:

[0072] (1) Initialization

[0073] The mathematical description of the pelican population initialization is as follows:

[0074] x i,j =l j +rand(u j -l j ).

[0075] Where x i,j is the j-th dimension position of the i-th pelican, i = 1, 2, ... N, j = 1, 2, ... m; N is the population of pelicans; m is the dimension of the problem to be solved; rand is a random number in the range [0, 1]; u j and l j are the upper and lower bounds of the j-th dimension of the problem to be solved.

[0076] In the Pelican optimization algorithm, the objective function of the problem can be used to calculate the objective function value of the Pelican. The objective function value of the Pelican population is expressed as a vector:

[0077]

[0078] Where F is the objective function vector of the pelican population; X i is the position of the i-th pelican.

[0079] (2) Approaching the prey (exploration phase)

[0080] During the exploration phase, the pelican locates the prey and moves toward it. By modeling the pelican's strategy, it scans the search space and uses the Point of Arrival (POA) to discover different areas. The key to this algorithm is that the prey's location is randomly generated in the search space, enhancing the ability to accurately search. The pelican's movement strategy is expressed as follows:

[0081]

[0082] Where, is the new state of the i-th pelican in the j-th dimension based on the exploration phase; P j is the position of the prey in the jth dimension; F P is the objective function value of the prey; F i is the objective function value for the i-th Pelican; I is a random number of 1 or 2. This parameter is randomly selected for each iteration and each member. When the value of this parameter is equal to 2, it causes more displacement for a member, potentially leading it to a new area of ​​the search space. Therefore, this parameter affects the detection ability of the POA, thereby accurately scanning the search space.

[0083] (3) Flying over water to catch prey (development stage)

[0084] During the exploitation phase, when a pelican reaches the water's surface, it spreads its wings to drive prey upward and collects it in its throat pouch. This strategy allows more prey to be captured within the attack area. By modeling the pelican's behavior, the POA can better converge to high-quality points in the hunting area, improving local search and exploitation capabilities. The algorithm needs to examine points surrounding the pelican to find a better solution. The pelican's hunting behavior can be represented as follows:

[0085]

[0086] Where, is the new state of the i-th pelican in dimension j based on the development stage; R(1-t / T) is x i,j The neighborhood radius of each member is represented by the coefficient R(1-t / T), where R is a constant of 0.2, t is the current iteration number, and T is the maximum iteration number. The coefficient R(1-t / T) represents the neighborhood radius of each member and is used to perform local searches near each member to converge to a better solution. As the algorithm progresses, this coefficient gradually decreases, resulting in a smaller neighborhood radius for each member. This allows the area around each member to be scanned in smaller and more precise steps, allowing the algorithm to approach the global optimal solution.

[0087] After the aforementioned search and capture process, the pelicans will attack and hunt again, performing an iterative calculation. The optimal position is constantly updated, and the pelicans need to constantly update their positions. Based on the newly selected design variables and objective function values, the newly obtained optimal position replaces the original optimal position. The iterative calculation continues until the optimal solution is found. The iteration stops and the final calculation result is output.

[0088] Based on the optimization ideas of the above-mentioned Pelican optimization algorithm, the implementation process of S2 can be:

[0089] Initialization: Randomly generate multiple initial combinations of weights and bias parameters within the global range of weights and bias parameters as the initial population of the Pelican optimization algorithm.

[0090] It should be understood that the initialization process includes initialization of basic parameters such as the pelican population size N, the maximum number of iterations T, and the search space dimension m.

[0091] Fitness evaluation: For each combination (i.e., individual) in the population, the BP neural network is used to perform forward propagation to calculate the predicted value and mean square error value, and the mean square error value is used as the fitness value for fitness evaluation.

[0092] Specifically, for each combination, the mean square error output by the BP neural network is used as its fitness value, that is, the loss function (mean square error function) of the BP neural network is used as the fitness function of the Pelican optimization algorithm, where the expression of the fitness function is:

[0093]

[0094] F is the fitness function, T q is the actual value of the qth sample, q = 1, 2, …, e, e is the total number of training samples; is the predicted value output by the BP neural network after the pth iteration of the qth sample; H is the total number of iterations, p = 1, 2, …, H.

[0095] Exploration and Exploitation: Update the population according to the update mechanism of the Pelican optimization algorithm.

[0096] Specifically, according to the update mechanism of the Pelican optimization algorithm (such as foraging behavior, position update formula, etc.), the new parameter space is explored and the population is updated. This step aims to find a potential combination of weights and bias parameters with high fitness (low error). That is, according to the new state of the population and the value of the objective function, the optimal predation position is set. The plan is constantly updated to obtain the best positioning solution.

[0097] Iteration: The process of iterative fitness evaluation and population update until the maximum number of iterations is reached, and the combination with the smallest fitness value is used as the initial optimal weight and bias parameters.

[0098] S4, based on the initial optimal weight and bias parameters and the sample set, the BP neural network is trained, and the forward propagation and back propagation processes are iterated to obtain the trained BP neural network.

[0099] Specifically, the initial optimal weights and bias parameters found by the Pelican optimization algorithm are assigned to the BP neural network, and the BP algorithm is applied for training. The prediction value and error are calculated through forward propagation, and the network weights and bias parameters are updated through backpropagation to minimize the error.

[0100] Iterate the above forward propagation and back propagation process, and stop the iteration when the BP neural network reaches the preset performance standard on the test set or reaches the maximum number of iterations.

[0101] In the above process, the POA algorithm is first used to perform a global optimization search on the initial weights and bias parameters of the BP neural network to find a set of potential high-quality parameter combinations. Subsequently, the BP algorithm is used to calculate the output error through forward propagation based on these potential high-quality parameters, and the network parameters are adjusted through backpropagation to minimize the error. This training process is repeated iteratively. Each iteration uses the POA update mechanism to explore the new parameter space and fine-tune the parameter values ​​in combination with the local optimization capability of BP until the model reaches the predetermined performance standard on the validation set or meets the stopping condition. The optimal solution is found by iteratively updating the position of the solution to address the problem of strong data quality dependence and easy fall into local optimal solutions when predicting IGBT life under traditional data-driven models. During the prediction process, the POA algorithm will continuously update the weights and bias parameters of the BP network to adapt to changes in input voltage parameters and improve the training speed of the algorithm.

[0102] S5, inputting the voltage parameter of the IGBT to be predicted into the trained BP neural network, and predicting the life of the IGBT through the trained BP neural network.

[0103] Through the global optimization of the above-mentioned Pelican optimization algorithm and the local optimization of the BP neural network, a more accurate prediction model can be obtained, the voltage parameters of the IGBT to be predicted are obtained, and the voltage parameters are input into the trained BP neural network. The BP neural network predicts the life of the IGBT based on the voltage parameters.

[0104] It should be understood that the failure criterion of the IGBT is failure when the collector-emitter voltage rises by 5%, so the number of cycles corresponding to the moment when the collector-emitter voltage rises to 5% is the life of the IGBT.

[0105] Next, the effectiveness of the present invention is verified through experiments.

[0106] Specifically, the performance of the BP neural network obtained by the present invention (which can be recorded as the POA-BP model) is evaluated on the test set by the prediction error RMSE and the goodness of fit R2. If the performance of the POA-BP model is not ideal, the parameters of the POA algorithm can be adjusted or the training data can be added to optimize the model. The prediction results of the POA-BP model can be compared with those of the traditional BP model. Figure 5 As shown, Figure 5 Figure (a) shows the prediction result of the original BP model. Figure 5 Figure (b) shows the prediction results of the POA-BP model.

[0107] RMSE is the square root of the mean square error (MSE), which is used to quantify the difference between the model's predicted value and the actual value (true value). Its calculation formula is:

[0108]

[0109] Where n is the number of samples; y i is the true value of the i-th sample; is the model's predicted value for the i-th sample.

[0110] R2 represents the degree of correlation between the variability of the model prediction value and the true value, and its calculation formula is:

[0111]

[0112] Where n is the number of samples; y i is the true value of the i-th sample; is the model's predicted value for the i-th sample; is the mean of all samples.

[0113] according to Figure 5 From the prediction results in FIG, it can be seen that the POA-BP prediction method of this embodiment effectively improves the prediction accuracy in IGBT life prediction compared with the traditional BP prediction method.

[0114] Based on the same inventive concept, the embodiment of the present invention also provides an IGBT life prediction device, such as Figure 6 As shown, the device includes:

[0115] The acquisition module 6-1 is used to acquire historical voltage parameters of the IGBT to construct a sample set; the voltage parameters are voltage parameters related to the life of the IGBT.

[0116] The model building module 6-2 is used to build a BP neural network for predicting the life of the IGBT.

[0117] The optimization module 6-3 is used to use the initial weights and bias parameters of the BP neural network as population individuals, the mean square error function of the BP neural network as the fitness function, and perform a global search through the Pelican optimization algorithm to obtain the initial optimal weights and bias parameters.

[0118] The training module 6-4 is used to train the BP neural network based on the initial optimal weight and bias parameters and the sample set, iterate the forward propagation and back propagation processes, and obtain a trained BP neural network.

[0119] The prediction module 6 - 5 is used to input the voltage parameter of the IGBT to be predicted into the trained BP neural network, and predict the life of the IGBT through the trained BP neural network.

[0120] Each module in the above-mentioned IGBT life prediction device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0121] The present invention further provides a computer device comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the embodiment of the method for predicting the life of an IGBT. The specific implementation method can be found in the method embodiment and will not be repeated here.

[0122] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions, wherein a computer program is stored on the storage medium. For example, a memory containing instructions, wherein the instructions can be executed by a processor of a computer device to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device. When the computer program is executed by the processor, the steps in the embodiment of the IGBT life prediction method can be implemented. The specific implementation method can be found in the method embodiment and will not be repeated here.

[0123] Those skilled in the art will appreciate that embodiments of the present invention may provide methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0125] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0127] It should be pointed out that the specific implementation methods described above can enable those skilled in the art to understand the invention more comprehensively, but do not limit the invention in any way. Therefore, although the present specification and examples have described the invention in detail, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the scope of protection of the patent for the invention. Any figure mark in the claims should not be regarded as limiting the claims involved. Any simple change or equivalent replacement of the technical solution that can be obviously obtained by any person familiar with the art within the technical scope disclosed in the present invention belongs to the scope of protection of the present invention.

Claims

1. A method for predicting the life of an IGBT, characterized in that: include: Acquire historical voltage parameters of an insulated gate bipolar transistor (IGBT) to construct a sample set; the voltage parameters are voltage parameters related to the life of the IGBT; Construct a BP neural network for predicting IGBT life; The initial weights and bias parameters of the BP neural network are used as population individuals, and the mean square error function of the BP neural network is used as the fitness function. The Pelican optimization algorithm is used to perform a global search to obtain the initial optimal weights and bias parameters. The BP neural network is trained based on the initial optimal weight and bias parameters and the sample set, and the forward propagation and back propagation processes are iterated to obtain the trained BP neural network; The voltage parameters of the IGBT to be predicted are input into the trained BP neural network, and the life of the IGBT is predicted by the trained BP neural network.

2. The IGBT life prediction method according to claim 1, characterized in that: The Pelican optimization algorithm is used to perform a global search to obtain the initial optimal weight and bias parameters, including: Randomly generate multiple initial weight and bias parameter combinations in the global range of weight and bias parameters as the initial population of the Pelican optimization algorithm; For each combination in the population, the prediction value and mean square error value are calculated by forward propagation through the BP neural network, and the mean square error value is used as the fitness value for fitness evaluation; Update the population according to the update mechanism of the Pelican optimization algorithm; The process of iterative fitness evaluation and population update is continued until the maximum number of iterations is reached, and the combination with the smallest fitness value is used as the initial optimal weight and bias parameters.

3. The IGBT life prediction method according to claim 1 or 2, characterized in that: Based on the initial optimal weights and bias parameters and sample sets, the BP neural network is trained, and the forward propagation and back propagation processes are iterated to obtain the trained BP neural network, including: Assign the initial optimal weights and bias parameters to the BP neural network, input the sample set into the BP neural network, calculate the output error through forward propagation, and optimize the weights and bias parameters in the network through back propagation to minimize the error; The forward propagation and back propagation processes are iterated until the BP neural network reaches the preset performance standard on the test set.

4. The IGBT life prediction method according to claim 1 or 2, characterized in that: BP neural network includes input layer, hidden layer and output layer; Among them, the input layer receives the voltage parameter, the hidden layer uses the S-shaped activation function to perform nonlinear mapping on the input voltage parameter, and the output layer outputs the predicted value, which is the collector-emitter voltage.

5. The IGBT life prediction method according to claim 1 or 2, characterized in that: After obtaining the historical voltage parameters of the IGBT, it also includes: Clean historical voltage parameters, identify and remove abnormal values; A filter is used to remove high frequency noise and normalize the image.

6. An IGBT life prediction device, characterized in that: include: An acquisition module, used for acquiring historical voltage parameters of an insulated gate bipolar transistor (IGBT) to construct a sample set; the voltage parameters are voltage parameters related to the life of the IGBT; Model building module, used to build BP neural network for predicting IGBT life; The optimization module is used to use the initial weights and bias parameters of the BP neural network as population individuals, the mean square error function of the BP neural network as the fitness function, and perform a global search through the Pelican optimization algorithm to obtain the initial optimal weights and bias parameters; The training module is used to train the BP neural network based on the initial optimal weight and bias parameters and the sample set, iterate the forward propagation and back propagation processes, and obtain the trained BP neural network; The prediction module is used to input the voltage parameter of the IGBT to be predicted into the trained BP neural network, and predict the life of the IGBT through the trained BP neural network.

7. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the life prediction method of the IGBT according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the life prediction method of the IGBT according to any one of claims 1 to 5 is implemented.

Images