Power device life prediction method and device, equipment and storage medium

Abstract

The invention relates to the technical field of power device service life monitoring, and provides a power device service life prediction method, device and equipment and a storage medium, and the method comprises the steps: collecting the operation parameters, environment variables and other key parameters of a to-be-predicted power device in the operation process, respectively inputting the collected key parameters into a pre-trained classification prediction model and a pre-trained trend prediction model to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; and based on a Kalman filtering algorithm, fusing the first prediction result and the second prediction result to obtain a life prediction result of the power device. Through the key parameters such as the operation parameters and the environment variables, when the aging state and the remaining service life of the power device are predicted, the influence of the multi-dimensional stress factor can be comprehensively considered, the Kalman filtering algorithm is combined with the fusion of different model prediction results, and the accuracy of predicting the service life of the power device under the complex working condition is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power device life monitoring, and in particular to a power device life prediction method, device, equipment and storage medium. Background Art

[0002] With the rapid development of modern manufacturing, CNC (Computerized Numerical Control) machine tools are increasingly used in industrial manufacturing. CNC systems automatically control machine tool operations through numerical control programs, achieving high-precision and high-efficiency processing. As a crucial component of CNC machine tools, the driver's primary function is to control the servo motor or spindle motor, ensuring precise motion control during machining. One of the core components of a driver is power devices, such as semiconductor devices like insulated gate bipolar transistors (IGBTs) and silicon carbide (SiC). Power devices are primarily responsible for efficiently converting electrical energy into mechanical energy to drive machine tools to perform various complex machining tasks.

[0003] Power devices are susceptible to performance degradation during operation due to long-term exposure to complex operating conditions such as high-frequency switching, thermal stress, and electrical stress. This can affect the stability of the driver and the machining accuracy and reliability of the entire CNC system. Aging or failure of power devices can cause sudden shutdowns of CNC machine tools, and in severe cases, even damage to workpieces, resulting in production interruptions and significant losses. Therefore, effectively assessing and predicting the lifespan of power devices, and enabling proactive maintenance and replacement, is crucial to maintaining the performance of CNC machine tools.

[0004] Currently, power device lifespan assessment methods have limitations. Most rely on physical damage models or periodic manual inspections, making it difficult to accurately reflect the aging status of power devices in real time and inefficient. Especially in complex processing environments, where the aging rate of power devices is affected by a combination of factors, existing assessment methods fail to comprehensively consider the impact of these factors on different power devices, affecting the accuracy of the aging assessment. Summary of the Invention

[0005] The present invention provides a power device life prediction method, device, equipment and storage medium to solve the defects in the prior art that the aging state and remaining life of power devices cannot be accurately predicted and the life prediction of power devices is inaccurate.

[0006] The present invention provides a method for predicting the life of a power device, comprising the following steps: Collecting key parameters during the operation of the power device to be predicted; the key parameters include at least operating parameters and environmental variables of the power device; Inputting the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; Based on the Kalman filter algorithm, the first prediction result and the second prediction result are fused to obtain a life prediction result of the power device.

[0007] According to the power device life prediction method provided by the present invention, before inputting the keyword parameters into the pre-trained classification prediction model and trend prediction model respectively, the method further includes: Acquiring historical key parameters corresponding to the power device and constructing a sample data set based on the historical key parameters; the sample data set includes label data, and the label data is used to characterize the remaining service life of the power device; Based on a preset number of iterations, the preset classification prediction model and the preset trend prediction model are jointly trained using the sample data set to obtain a pre-trained classification prediction model and a pre-trained trend prediction model.

[0008] According to the power device life prediction method provided by the present invention, based on a preset number of iterations, the preset classification prediction model and the preset trend prediction model are jointly trained using the sample data set to obtain a pre-trained classification prediction model and a pre-trained trend prediction model, including: Iteratively training a preset classification prediction model using the sample data set based on a preset number of iterations to obtain a first prediction accuracy of the classification prediction model, and iteratively training a preset trend prediction model using the sample data set to obtain a second prediction accuracy of the trend prediction model; determining an initial measurement error according to the first prediction accuracy and the second prediction accuracy, and calculating a Kalman gain value of the Kalman filter algorithm based on the initial measurement error and a preset estimation error; The first prediction value and the second prediction value are fused using the Kalman gain value to obtain a fused prediction result, and the estimated error is updated according to the Kalman gain value; the first prediction value is a prediction value corresponding to the prediction result of the classification prediction model, and the second prediction value is a prediction value corresponding to the prediction result of the trend prediction model; The model parameters are adjusted based on the fusion prediction result, and the step of iteratively training the preset classification prediction model based on the preset number of iterations using the sample data set is returned and executed until the number of adjustments to the model parameters reaches the preset number of iteration cycles, thereby obtaining a pre-trained classification prediction model and a pre-trained trend prediction model; the model parameters include the first model parameters of the classification prediction model and the second model parameters of the trend prediction model.

[0009] According to the power device life prediction method provided by the present invention, the iterative training of a preset classification prediction model using the sample data set includes: Inputting the sample data set into a preset classification prediction model, and iteratively training the classification prediction model; In each round of iteration, the function value of the objective function is minimized as the optimization goal, and the first model parameter of the classification prediction model is iteratively optimized based on the preset population optimization algorithm to obtain the optimal parameter combination; In which, the objective function is constructed based on a regularization parameter and a slack variable, and the first model parameters include the regularization parameter and the slack variable; the slack variable is used to characterize that some predicted values ​​in the prediction results of the classification prediction model are allowed to exceed a preset error range.

[0010] According to the power device life prediction method provided by the present invention, the first model parameter of the classification prediction model is iteratively optimized based on a preset population optimization algorithm to obtain the optimal parameter combination, including: Based on a preset population size, randomly initializing population particles within a parameter space of a first model parameter of the classification prediction model; a particle position of any population particle corresponds to a parameter combination of the first model parameter of the classification prediction model; According to the function value of the objective function corresponding to each of the swarm particles, minimizing the function value of the objective function is used as an optimization goal to determine the individual optimal position and the swarm optimal position of the swarm particles; Based on the individual optimal position and the population optimal position, the particle position and particle velocity of the population particles are iteratively updated, and the step of determining the individual optimal position and the population optimal position of the population particles according to the function value of the objective function corresponding to each population particle is returned and executed until the number of iterative updates of the population particles reaches a preset maximum number of iterations, and the optimal parameter combination of the first model parameters is obtained according to the individual optimal position and the population optimal position.

[0011] According to the power device life prediction method provided by the present invention, the trend prediction model includes a hidden layer and an output layer; the iterative training of the preset trend prediction model using the sample data set includes: Inputting the sample data set into a preset trend prediction model, and performing iterative training on the trend prediction model; In each round of iteration, the trend prediction model performs nonlinear mapping on the sample data in the sample data set by randomly generating the weight matrix, bias vector and preset activation function of the hidden layer to form the output of the hidden layer; The output weight matrix is ​​obtained by minimizing the target error between the output weight matrix and the output matrix of the trend prediction model; The output matrix is ​​calculated based on the weight matrix, the bias vector and the preset real-time environmental variables. The prediction result of the trend prediction model is the product of the output weight matrix and the output matrix; the real-time environmental variables are variables among the environmental variables, which are used to characterize the dynamic change factors of the environment in which the power device is located.

[0012] According to the power device life prediction method provided by the present invention, after collecting the key parameters during the operation of the power device to be predicted, the method further includes: The key parameters are preprocessed to clean abnormal data in the key parameters; wherein the preprocessing includes normalization processing and filtering processing, and the abnormal data includes noise data and abnormal value data.

[0013] The present invention also provides a power device life prediction device, comprising the following modules: A data acquisition module, configured to acquire key parameters during the operation of the power device to be predicted; the key parameters at least include operating parameters and environmental variables of the power device; a prediction module, configured to input the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; A fusion module is used to fuse the first prediction result and the second prediction result based on a Kalman filter algorithm to obtain a life prediction result of the power device.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the power device life prediction method as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for predicting the life of a power device.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for predicting the life of a power device as described above is implemented.

[0017] The power device life prediction method, apparatus, device, and storage medium provided by the present invention collect key parameters, such as operating parameters and environmental variables, during the operation of the power device and use them as inputs to a classification prediction model and a trend prediction model. The method predicts the aging state of the power device and its changing trend, respectively. Finally, based on a Kalman filter algorithm, the prediction results are integrated to obtain a life prediction result for the power device. By using key parameters, such as the operating parameters and environmental variables of the power device, the impact of multidimensional stress factors on the aging state of the power device can be comprehensively considered when predicting the aging state and remaining useful life of the power device. By combining the Kalman filter algorithm with the integration of prediction results from different models, the accuracy of life prediction for power devices under complex operating conditions is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 It is a flow chart of the power device life prediction method provided by the present invention.

[0020] Figure 2 It is a schematic diagram of the process of power device life prediction provided by the present invention.

[0021] Figure 3 It is a schematic diagram of the model training process for power device life prediction provided by the present invention.

[0022] Figure 4 It is a structural schematic diagram of the power device life prediction device provided by the present invention.

[0023] Figure 5It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0025] An embodiment of the present invention provides a power device life prediction method. By real-time monitoring of the operating parameters of the power device and combining it with the evaluation and detection of the aging status using a prediction model, the service life and aging status of the power device can be accurately predicted, and real-time monitoring of the aging status of the power device can be achieved. The method is particularly suitable for monitoring the aging status and remaining life of power devices in CNC machine tool drives, and solves the problems of multi-dimensional stress, individual differences and real-time updating in power device life prediction under complex working conditions, thereby avoiding CNC machine tool failures caused by power device failure.

[0026] Specifically, Figure 1 FIG. 1 is a flow chart of a method for predicting the life of a power device provided by the present invention, such as Figure 1 As shown, the method includes the following steps: Step 100: collecting key parameters of the power device to be predicted during operation; the key parameters include at least operating parameters and environmental variables of the power device; Step 200: Input the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; Step 300: Based on a Kalman filter algorithm, the first prediction result and the second prediction result are integrated to obtain a life prediction result of the power device.

[0027] Collect key parameters of the power device to be predicted during operation, which key parameters include at least the operating parameters and environmental variables of the power device, wherein the operating parameters of the power device include current, voltage and switching frequency, and the environmental variables of the power device include the operating condition data of the power device, and the operating condition data includes temperature and load.

[0028] Optionally, based on a preset collection time interval, the parameter value of each parameter of the key parameters of the power device is collected separately to obtain time series data of each parameter.

[0029] The collected key parameters are input into a pre-trained classification prediction model and a pre-trained trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model. The first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device.

[0030] The time series data corresponding to each parameter in the collected key parameters are used as input data for the pre-trained classification prediction model and the pre-trained trend prediction model, respectively, to obtain the first prediction result output by the classification prediction model and the second prediction result output by the trend prediction model.

[0031] In one embodiment, the trend prediction model is an RTE-ELM (Real-Time of Environmental Variables Model) model built based on the Extreme Learning Machine (ELM) model. The ELM model is a fast and efficient machine learning model, commonly used for classification and regression problems. Unlike traditional neural network models, the ELM model does not require iterative training and can randomly generate hidden layer weights and biases, which has the advantage of high computational efficiency. The RTE-ELM introduces real-time environmental variables (such as temperature, humidity, load, etc.) into the model, allowing the model to dynamically adapt to environmental changes when faced with complex and dynamically changing environments, thereby improving the stability and accuracy of the prediction.

[0032] In another embodiment, the trend prediction model is constructed based on a time series prediction model such as LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit), which can process the time series of key parameters and accurately predict the aging trend of power devices, thereby determining the remaining service life of the power devices.

[0033] Similarly, the classification prediction model can be built based on an SVM (Support Vector Machine) model or a random forest. There is no specific limitation on the classification prediction model and trend prediction model.

[0034] The classification prediction model is used to predict the aging status of power devices, based on which the remaining service life of the power devices can be determined. The trend prediction model is used to predict the changing trend of the aging status of power devices, based on which the changing trend of the remaining service life of the power devices can be determined.

[0035] Based on the Kalman filter algorithm, the first prediction result and the second prediction result are integrated to obtain a life prediction result for the power device, which represents the life value of the remaining useful life of the power device. In the case where the first prediction result represents the aging state of the power device and the second prediction result represents the changing trend of the aging state of the power device, based on the aging state of the power device and its changing trend, the remaining useful life of the power device can be more accurately predicted, thereby obtaining a life prediction result for the power device.

[0036] Among them, the Kalman filter algorithm is an algorithm that uses the linear system state equation to perform optimal state estimation by inputting and outputting observation data. Specifically, based on the observation data and prior knowledge, the predicted state and its covariance are recursively calculated, and then the predicted state is corrected according to the actual observation value to obtain the estimated value, so as to minimize the mean square error. In one embodiment, the first prediction result and the second prediction result are fused based on the Kalman filter algorithm, and the first prediction result is used as the prior state estimate and the second prediction result is used as the measurement value at the current moment. Based on the prior state estimate, the state value at the next moment is predicted based on the measurement value at the current moment, and the remaining service life of the power device is determined based on the predicted state estimate value at the next moment.

[0037] In one embodiment, referring to Figure 2 The power device lifespan prediction process shown in the figure inputs the key parameters of the power device to be predicted into a pre-trained classification prediction model and a trend prediction model, respectively. The classification prediction model outputs a first prediction result, and the trend prediction model outputs a second prediction result. The first and second prediction results are then fused using a Kalman filter algorithm to obtain a lifespan prediction result for the power device. The key parameters of the power device are collected during operation and include both operating parameters and environmental variables.

[0038] In this embodiment, key parameters such as operating parameters and environmental variables collected during the operation of power devices are used as inputs to classification prediction models and trend prediction models to predict the aging state of the power devices and their changing trends. Finally, the prediction results are integrated based on the Kalman filter algorithm to obtain a prediction result for the power device lifespan. By using key parameters such as operating parameters and environmental variables of power devices, the impact of multi-dimensional stress factors on the aging state of power devices can be comprehensively considered when predicting the aging state and remaining useful life of power devices. By combining the Kalman filter algorithm with the integration of prediction results from different models, the accuracy of power device lifespan prediction under complex operating conditions is improved.

[0039] Optionally, the key parameters collected during the operation of the power device to be predicted need to be preprocessed to ensure the validity of the data. Based on this, after step 100, the following steps may also be included: Step 110 , preprocessing the key parameters to clean abnormal data in the key parameters; wherein the preprocessing includes normalization processing and filtering processing, and the abnormal data includes noise data and abnormal value data.

[0040] The key parameters of the power device are preprocessed to clean abnormal data in the key parameters, wherein the preprocessing of the key parameters includes normalization processing and filtering processing, and the abnormal data to be cleaned includes noise data and abnormal value data.

[0041] Optionally, key parameters include operating parameters and environmental variables. The operating parameters may include one or more parameters, and the environmental variables may include one or more variables. For different operating parameters and environmental variables, a parallel approach can be used to preprocess the time series data corresponding to the different parameters. Abnormal data to be cleaned includes outliers identified from the time series data corresponding to each key parameter. By preprocessing the key parameters, noise and outliers are eliminated, ensuring that the prediction model can make predictions on a valid data set, thereby ensuring the accuracy of the prediction results.

[0042] Optionally, before inputting the collected key parameters of the power device into the classification prediction model and the trend prediction model for prediction, model pre-training is required. The pre-trained classification prediction model and the trend prediction model are obtained by joint training based on the Kalman filter algorithm. Based on this, before step 200, the following steps may also be included: Step 101: Acquire historical key parameters corresponding to the power device, and construct a sample data set based on the historical key parameters; the sample data set includes label data, and the label data is used to represent the remaining service life of the power device; Step 102 : Based on a preset number of iterations, the preset classification prediction model and the preset trend prediction model are jointly trained using the sample data set to obtain a pre-trained classification prediction model and a pre-trained trend prediction model.

[0043] Obtain historical key parameters corresponding to the power device to be predicted, and construct a sample data set based on the historical key parameters. Use the sample data set to jointly train a preset classification prediction model and a preset trend prediction model to obtain a pre-trained classification prediction model and a pre-trained trend prediction model.

[0044] Optionally, the historical key parameters are acquired during the historical operation of a power device of the same type as the power device to be predicted. The constructed sample dataset includes label data, which is used to characterize the aging state of the power device, that is, the remaining useful life of the power device.

[0045] Optionally, the joint training of the classification prediction model and the trend prediction model is achieved based on a preset number of iterations. Based on the preset number of iterations, the classification prediction model and the trend prediction model are iteratively trained respectively, and then based on the Kalman filter algorithm, the model parameters are adjusted according to the training results, and then the classification prediction model and the trend prediction model are iteratively trained again until the number of adjustments to the model parameters reaches the preset number of iteration cycles.

[0046] Therefore, step 102 may further include: Step 112, iteratively training a preset classification prediction model using the sample data set based on a preset number of iterations to obtain a first prediction accuracy of the classification prediction model, and iteratively training a preset trend prediction model using the sample data set to obtain a second prediction accuracy of the trend prediction model; Step 122: determining an initial measurement error according to the first prediction accuracy and the second prediction accuracy, and calculating a Kalman gain value of the Kalman filter algorithm based on the initial measurement error and a preset estimation error; Step 132: Using the Kalman gain value, the first prediction value and the second prediction value are fused to obtain a fused prediction result, and the estimated error is updated according to the Kalman gain value; the first prediction value is a prediction value corresponding to the prediction result of the classification prediction model, and the second prediction value is a prediction value corresponding to the prediction result of the trend prediction model; Step 142, adjust the model parameters based on the fusion prediction result, return and execute the step of iteratively training the preset classification prediction model based on the preset number of iterations using the sample data set, until the number of adjustments to the model parameters reaches the preset number of iteration cycles, and obtain a pre-trained classification prediction model and a pre-trained trend prediction model; the model parameters include the first model parameters of the classification prediction model, and the second model parameters of the trend prediction model.

[0047] Based on a preset number of iterations, the preset classification prediction model is iteratively trained using a sample data set to obtain a first prediction accuracy of the classification prediction model, and based on a preset number of iterations, the preset trend prediction model is iteratively trained using a sample data set to obtain a second prediction accuracy of the trend prediction model.

[0048] Optionally, the number of iterations of the classification prediction model and the number of iterations of the trend prediction model may be the same or different, and there is no specific limitation on this.

[0049] An initial measurement error is determined based on the first prediction accuracy and the second prediction accuracy, and a Kalman gain value of the Kalman filter algorithm is calculated based on the initial measurement error and a preset estimation error. In one embodiment, the Kalman gain value is calculated based on the initial measurement error and the estimation error, as shown in the following formula 1: ; (1) in, For the The Kalman gain value of the round iteration, represents the iteration round, and represents the estimation error, represents the initial measurement error.

[0050] Furthermore, the Kalman gain value is used to characterize the confidence weight of the prediction results of the classification prediction model and the trend prediction model in the life prediction results. For prediction results with high prediction accuracy, the corresponding confidence weight is greater. Therefore, based on the first prediction accuracy and the second prediction accuracy, the initial measurement error is determined, and the value of the initial measurement error is adjusted, and then the Kalman gain value is adjusted to realize the allocation of confidence weights for the prediction results of the classification prediction model and the trend prediction model.

[0051] The Kalman gain value is calculated based on the estimation error and the initial measurement error as shown in Formula 1, and the estimation error needs to be iteratively updated based on the Kalman gain value. Therefore, after each iterative training of the classification prediction model and the trend prediction model, the Kalman gain value is adjusted by adjusting the initial value of the initial measurement error for the next iterative training.

[0052] After calculating the Kalman gain value based on the initial measurement error and the estimated error, the first predicted value of the classification prediction model and the second predicted value of the trend prediction model are fused using the Kalman gain value to obtain a fused prediction result, and the estimated error is updated according to the Kalman gain value. Further, based on the fused prediction result, the model parameters are adjusted. After adjusting the model parameters, the classification prediction model and the trend prediction model are iteratively trained again until the number of adjustments to the model parameters reaches a preset number of iteration cycles, thereby obtaining a pre-trained classification prediction model and a pre-trained trend prediction model. The model parameters that need to be adjusted include the first model parameters of the classification prediction model and the second model parameters of the trend prediction model.

[0053] Optionally, based on a preset number of iterations, the classification prediction model is iteratively trained using a sample data set to obtain a first prediction accuracy of the classification prediction model and a first prediction value during each iteration. The first prediction value is the prediction value corresponding to the prediction result of the classification prediction model during each iteration. Accordingly, based on a preset number of iterations, the trend prediction model is iteratively trained using a sample data set to obtain a second prediction accuracy of the trend prediction model and a second prediction value during each iteration. The second prediction value is the prediction value corresponding to the prediction result of the trend prediction model during each iteration. Both the first prediction value and the second prediction value are used to characterize the remaining useful life of the power device.

[0054] In one embodiment, after iterative training of the classification prediction model and the trend prediction model, the initial measurement error is determined based on the first prediction accuracy of the classification prediction model and the second prediction accuracy of the trend prediction model, and then the initial value of the Kalman gain is calculated based on the initialization value of the estimated error and the initial measurement error, and the first prediction value of the first iteration and the second prediction value of the first iteration are fused to obtain the fused prediction value of the first iteration, and then the estimated error is updated based on the fused prediction value, and a new Kalman gain value is recalculated based on the updated estimated error, and the first prediction value and the second prediction value of the next iteration are fused. In this way, the prediction values ​​of the classification prediction model and the trend prediction model in each round of iterative training are fused to obtain a fused prediction result, and the first model parameters of the classification prediction model and the second model parameters of the trend prediction model are adjusted according to the fusion result. Then, the classification prediction model and the trend prediction model after the model parameters are adjusted are re-trained to realize the iterative cycle of joint training of the classification prediction model and the trend prediction model until the preset number of iterative cycles is reached.

[0055] Optionally, the Kalman filter algorithm may be used to fuse the prediction results of the classification prediction model and the trend prediction model, as shown in the following formula 2: ; (2) In formula 2, It is The predicted value of the remaining service life corresponding to the fusion prediction result of the iteration, The trend forecast model The predicted value of the remaining service life corresponding to the prediction result of the iteration, Time classification prediction model The predicted value of the remaining service life corresponding to the prediction result of the iteration, It is The Kalman gain value corresponding to the iteration.

[0056] Based on the fusion method shown in Formula 2, the Kalman filter algorithm uses the predicted value of the classification prediction model as the prior estimate and the predicted value of the trend prediction model as the observation value at the current moment to predict the estimated value of the remaining useful life at the next moment.

[0057] Optionally, after the prediction results of the classification prediction model and the trend prediction model are integrated, the estimation error is iteratively updated based on the Kalman gain value. A feasible iterative update method is as follows: ; (3) in, It is The estimated error of the round iteration is used to determine the corresponding initial measurement error according to the fusion prediction result of this round iteration, and then based on the initial measurement error, and According to the formula 1, calculate , used for classification prediction model and trend prediction model The prediction results of the iterative rounds are fused.

[0058] In this embodiment, the model training process includes two iterations: one for iterative training of the classification prediction model and the trend prediction model, and another for adjusting model parameters based on the Kalman filter algorithm. Furthermore, in other embodiments, the model training process may also include three iterations. Specifically, during the iterative training of the classification prediction model, a population optimization algorithm is used in each iteration to optimize the model parameters of the classification prediction model.

[0059] Therefore, in step 112, iteratively training the classification prediction model using the sample data set may also include: Step 401: input the sample data set into a preset classification prediction model and perform iterative training on the classification prediction model; Step 402: In each iteration, the first model parameter of the classification prediction model is iteratively optimized based on a preset population optimization algorithm with the function value of the objective function minimized as the optimization goal to obtain the optimal parameter combination; In which, the objective function is constructed based on a regularization parameter and a slack variable, and the first model parameters include the regularization parameter and the slack variable; the slack variable is used to characterize that some predicted values ​​in the prediction results of the classification prediction model are allowed to exceed a preset error range.

[0060] The sample data set is input into the preset classification prediction model, and the classification prediction model is iteratively trained. In each round of iteration, the function value of the objective function is minimized as the optimization goal. Based on the preset population optimization algorithm, the first model parameter of the classification prediction model is iteratively optimized to obtain the optimal parameter combination.

[0061] The objective function is constructed based on a regularization parameter and a slack variable, and the first model parameter of the classification prediction model includes the regularization parameter and the slack variable. The slack variable is used to represent the tolerance of some predicted values ​​in the prediction results of the classification prediction model to exceed a preset error range, while the regularization parameter is used to regularize the cumulative value of the slack variable of each swarm particle in the swarm. The swarm optimization algorithm can be any one of a particle swarm optimization algorithm, an ant colony optimization algorithm, and a genetic algorithm.

[0062] Furthermore, in step 402, the preset population optimization algorithm may further include: Step 501: Based on a preset population size, randomly initialize population particles in the parameter space of the first model parameter of the classification prediction model; the particle position of any population particle corresponds to a parameter combination of the first model parameter of the classification prediction model; Step 502: determining the optimal position of each particle in the population and the optimal position of the population according to the function value of the objective function corresponding to each particle in the population, with minimizing the function value of the objective function as the optimization goal; Step 503: Iteratively update the particle positions and particle velocities of the swarm particles based on the individual optimal positions and the swarm optimal positions, return to and execute the step of determining the individual optimal positions and the swarm optimal positions of the swarm particles based on the function value of the objective function corresponding to each swarm particle, until the number of iterative updates of the swarm particles reaches a preset maximum number of iterations, and then obtain the optimal parameter combination of the first model parameters based on the individual optimal positions and the swarm optimal positions.

[0063] First, based on a preset population size, which represents the number of particles in the population, the population particles are randomly initialized within the parameter space of the first model parameter of the classification prediction model, including initializing the initial position and initial velocity of each population particle. The position of any population particle corresponds to a parameter combination of the first model parameter of the classification prediction model.

[0064] Furthermore, according to the function value of the objective function corresponding to each swarm particle, the optimal position of the individual and the optimal position of the swarm particles are determined with minimizing the function value of the objective function as the optimization goal. Based on the individual and the optimal positions of the swarm particles, the particle positions and the particle velocities of the swarm particles are iteratively updated. Then, the function value of the objective function corresponding to each swarm particle is re-determined, and with minimizing the function value of the objective function as the optimization goal, new optimal positions of the individual and the optimal positions of the swarm are determined. In this manner, the particle positions and the particle velocities of the swarm particles are iteratively updated to seek the optimal parameter combination. Until the number of iterative updates of the swarm particles reaches a preset maximum number of iterations, the optimal parameter combination of the first model parameters is obtained according to the individual and the optimal positions of the swarm.

[0065] Optionally, the classification prediction model is constructed based on a support vector machine (SVM) model, and the objective function is constructed based on the prediction error of the classification prediction model, a preset regularization parameter, and a slack variable. In one embodiment, the constructed objective function is shown in the following formula 3: ; (4) in, is the base of natural logarithm, which is a constant. is the regularization parameter, and is a slack variable used to represent the error range that the predicted value is allowed to exceed. is the population size.

[0066] In each round of iterative optimization, the swarm particles update the particle velocity and particle position as follows: ; (5) In formula 5, represents the first particle in the population The particle in The particle velocity during the iteration, represents the first particle in the population The particle in The particle position during the iteration, is the inertia weight, and They are individual learning factor and group learning factor, and is a random number, It is The optimal position of the individual in the iterative process, It is The optimal position of the group during the iteration.

[0067] The individual optimal position is the local optimal solution of the parameter combination, and the population optimal position is the global optimal solution of the parameter combination. Optionally, the parameter combination of the first model parameter of the classification prediction model includes a regularization parameter and a slack variable, and may also include the kernel function parameter λ of the SVM model.

[0068] Optionally, the trend prediction model is constructed based on the ELM model. The constructed trend prediction model includes at least a hidden layer and an output layer. Based on this, in step 112, iterative training of the preset trend prediction model using the sample data set may also include: Step 601: input the sample data set into a preset trend prediction model, and perform iterative training on the trend prediction model; In each round of iteration, the trend prediction model performs nonlinear mapping on the sample data in the sample data set by randomly generating the weight matrix, bias vector and preset activation function of the hidden layer to form the output of the hidden layer; Step 602, obtaining the output weight matrix by minimizing the target error between the output weight matrix of the trend prediction model and the output matrix; The output matrix is ​​calculated based on the weight matrix, the bias vector and the preset real-time environmental variables. The prediction result of the trend prediction model is the product of the output weight matrix and the output matrix; the real-time environmental variables are variables among the environmental variables, which are used to characterize the dynamic change factors of the environment in which the power device is located.

[0069] When iteratively training the trend prediction model, the sample data set is input into the preset trend prediction model, and the trend prediction model is iteratively trained. In each round of iteration, the trend prediction model randomly generates a weight matrix and bias vector of the hidden layer, and uses a preset activation function to perform nonlinear mapping on the sample data in the sample data set to form the output of the hidden layer.

[0070] Furthermore, the output weight matrix of the trend prediction model is obtained by minimizing the target error between the output weight matrix and the output matrix. The output matrix of the trend prediction model is calculated based on the hidden layer weight matrix, the bias vector, and preset real-time environmental variables. The prediction result of the trend prediction model is the product of the output weight matrix and the output matrix. Real-time environmental variables are variables among the collected environmental variables that are used to characterize the dynamic changes in the environment of the power device, such as temperature and load.

[0071] In the ELM model, the input data is nonlinearly mapped through the randomly generated hidden layer weight matrix W and bias vector b, as well as the preset activation function g(.) to form the output of the hidden layer: H=g(XW+b). W represents the randomly generated weight matrix from the input layer to the hidden layer, and b is the randomly generated bias vector. By minimizing the error between the output weight matrix β and the output matrix T, the output weight matrix β is obtained. The output weight matrix is ​​the weight matrix of the output layer. Specifically, β=H + T, where H + is the Moore-Penrose pseudo-inverse matrix of the output matrix of the hidden layer, T is the output matrix, and the output matrix is ​​the output of the output layer.

[0072] The traditional ELM model is susceptible to noise interference in complex environments, which affects the prediction effect. In this embodiment, by introducing real-time environmental variables , which enhances the adaptability and robustness of the ELM model. Its output matrix is: .

[0073] Among them, real-time environment variables Generated by standard normal distribution, its probability density function is: ; (6) In formula 6, is a random variable that represents the change in real-time environmental variables. By introducing real-time environmental variables and building an RTE-ELM model based on the ELM model, we can maintain the stability and accuracy of predictions under environmental changes while effectively avoiding model overfitting. Finally, the weight matrix of the output layer of the trend prediction model is obtained through the formula: Calculate and get, is the output matrix of the hidden layer The Moore-Penrose pseudo-inverse matrix, It is the predicted value corresponding to the final prediction result of the trend prediction model, which is used to characterize the remaining service life of the power device.

[0074] In one embodiment, referring to Figure 3 The training process for a prediction model used for power device lifetime prediction shows that the RET-ELM model, constructed using an extreme learning machine (ELM) model that incorporates the real-time environmental variable RET, serves as the trend prediction model. A PSO-SVM model, constructed using a particle swarm optimization algorithm (PSO) to optimize the parameters of a support vector machine (SVM) model, is used as the classification prediction model.

[0075] Furthermore, based on the sample data set constructed based on the historical key parameters of the power device, the RET-ELM model and the PSO-SVM model were iteratively trained to obtain the first prediction value of the RET-ELM model in each round of iteration. , and the second predicted value of the PSO-SVM model in each round of iteration Then, based on the Kalman filter algorithm, the first prediction value and the second prediction value are dynamically fused to obtain the fusion prediction result. The model parameters of the RET-ELM model and the PSO-SVM model are further adjusted according to the fusion prediction results, and then the RET-ELM model and the PSO-SVM model after the model parameters are adjusted are iteratively trained again, thereby realizing a cycle of iterative training of the RET-ELM model and the PSO-SVM model.

[0076] Among them, when the PSO-SVM model is iteratively trained, in each round of iterative training, the particle swarm optimization algorithm PSO is used to iteratively optimize the parameter combination of the SVM model to find the optimal parameter combination of the SVM model in this iteration, and the first prediction value of this iteration is generated based on the optimal parameter combination.

[0077] The trained RET-ELM model and PSO-SVM model can predict the aging status and remaining service life of power devices based on the key parameters collected in real time during the operation of power devices, and fuse the prediction results based on the Kalman filter algorithm as the life prediction results of power devices, realizing multi-model fusion life prediction, ensuring the prediction accuracy and real-time performance under complex working conditions, and improving the accuracy of the assessment of the aging status and remaining service life of power devices.

[0078] In this embodiment, facing the complex working conditions of power devices in CNC machine tools, by introducing real-time environmental variables, multi-dimensional stress factors such as temperature fluctuations and load changes can be comprehensively considered in the prediction model, thereby enhancing the robustness of the prediction model under different working conditions and ensuring the stability and reliability of the prediction results. In this way, the aging condition and remaining service life of power devices can be more accurately evaluated under complex and changeable working conditions, avoiding the prediction inaccuracy problem existing in single-dimensional models.

[0079] Furthermore, the aging rate of power devices is affected by a combination of factors, including individual differences and operating conditions. In view of the fact that the prediction model is difficult to adapt to the different aging characteristics of different power devices, resulting in large prediction errors, the embodiment of the present invention uses a population optimization algorithm to iteratively optimize the model parameters during the model training stage, so that the prediction model can adaptively adjust the model parameters according to different types of power devices and operating conditions, so that the prediction model can better reflect the aging status of different types of power devices, thereby improving the accuracy of individualized life prediction.

[0080] During the actual operation of CNC machine tools, the life status of power devices is constantly changing. Especially when the operating conditions change, it is difficult to make timely and accurate predictions of the aging status and remaining service life of power devices under different working conditions at different times based on the prediction model trained only on historical data. The embodiment of the present invention introduces the Kalman filter algorithm to fuse the prediction results of different prediction models, and can make real-time adjustments to the life prediction results of power devices. In addition, the Kalman filter algorithm can realize recursive updates of the prediction model, so that the model can be adjusted according to real-time data and output the optimal prediction results. Through the recursive estimation of the Kalman filter algorithm, the current measurement value and the predicted value can be dynamically fused, and the prediction error can be gradually reduced, which not only improves the prediction accuracy, but also ensures the real-time and accuracy of the prediction results, and can timely reflect the current aging status and remaining service life of the power device.

[0081] By accurately predicting the aging status and remaining life of power devices in advance, CNC machine tool downtime caused by sudden power device failures can be avoided, reducing the frequency of manual maintenance and the risk of emergency shutdowns. Furthermore, by predicting the aging status of power devices, CNC machine tool maintenance personnel can schedule maintenance and replacement of power devices at appropriate times, optimizing maintenance plans and reducing unnecessary maintenance costs, thereby improving the operating efficiency and overall reliability of CNC machine tools. Accurately predicting the life of power devices not only improves equipment safety but also reduces unnecessary equipment maintenance, significantly reducing production costs.

[0082] The power device life prediction device provided by the present invention is described below. The power device life prediction device described below and the power device life prediction method described above can be referenced to each other.

[0083] Reference Figure 4 The power device life prediction device provided by the embodiment of the present invention includes: The data acquisition module 10 is used to acquire key parameters of the power device to be predicted during operation; the key parameters include at least the operating parameters and environmental variables of the power device; A prediction module 20 is configured to input the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; The fusion module 30 is configured to fuse the first prediction result and the second prediction result based on a Kalman filter algorithm to obtain a life prediction result for the power device.

[0084] In one embodiment, the power device lifetime prediction apparatus further includes a model training module for: Acquiring historical key parameters corresponding to the power device and constructing a sample data set based on the historical key parameters; the sample data set includes label data, and the label data is used to characterize the remaining service life of the power device; Based on a preset number of iterations, the preset classification prediction model and the preset trend prediction model are jointly trained using the sample data set to obtain a pre-trained classification prediction model and a pre-trained trend prediction model.

[0085] In one embodiment, the model training module is further used to: Iteratively training a preset classification prediction model using the sample data set based on a preset number of iterations to obtain a first prediction accuracy of the classification prediction model, and iteratively training a preset trend prediction model using the sample data set to obtain a second prediction accuracy of the trend prediction model; determining an initial measurement error according to the first prediction accuracy and the second prediction accuracy, and calculating a Kalman gain value of the Kalman filter algorithm based on the initial measurement error and a preset estimation error; The first prediction value and the second prediction value are fused using the Kalman gain value to obtain a fused prediction result, and the estimated error is updated according to the Kalman gain value; the first prediction value is a prediction value corresponding to the prediction result of the classification prediction model, and the second prediction value is a prediction value corresponding to the prediction result of the trend prediction model; The model parameters are adjusted based on the fusion prediction result, and the step of iteratively training the preset classification prediction model based on the preset number of iterations using the sample data set is returned and executed until the number of adjustments to the model parameters reaches the preset number of iteration cycles, thereby obtaining a pre-trained classification prediction model and a pre-trained trend prediction model; the model parameters include the first model parameters of the classification prediction model and the second model parameters of the trend prediction model.

[0086] In one embodiment, the model training module is further used to: Inputting the sample data set into a preset classification prediction model, and iteratively training the classification prediction model; In each round of iteration, the function value of the objective function is minimized as the optimization goal, and the first model parameter of the classification prediction model is iteratively optimized based on the preset population optimization algorithm to obtain the optimal parameter combination; In which, the objective function is constructed based on a regularization parameter and a slack variable, and the first model parameters include the regularization parameter and the slack variable; the slack variable is used to characterize that some predicted values ​​in the prediction results of the classification prediction model are allowed to exceed a preset error range.

[0087] In one embodiment, the model training module is further used to: Based on a preset population size, randomly initializing population particles within a parameter space of a first model parameter of the classification prediction model; a particle position of any population particle corresponds to a parameter combination of the first model parameter of the classification prediction model; According to the function value of the objective function corresponding to each of the swarm particles, minimizing the function value of the objective function is used as an optimization goal to determine the individual optimal position and the swarm optimal position of the swarm particles; Based on the individual optimal position and the population optimal position, the particle position and particle velocity of the population particles are iteratively updated, and the step of determining the individual optimal position and the population optimal position of the population particles according to the function value of the objective function corresponding to each population particle is returned and executed until the number of iterative updates of the population particles reaches a preset maximum number of iterations, and the optimal parameter combination of the first model parameters is obtained according to the individual optimal position and the population optimal position.

[0088] In one embodiment, the trend prediction model includes a hidden layer and an output layer; the model training module is further used to: Inputting the sample data set into a preset trend prediction model, and performing iterative training on the trend prediction model; In each round of iteration, the trend prediction model performs nonlinear mapping on the sample data in the sample data set by randomly generating the weight matrix, bias vector and preset activation function of the hidden layer to form the output of the hidden layer; The output weight matrix is ​​obtained by minimizing the target error between the output weight matrix and the output matrix of the trend prediction model; The output matrix is ​​calculated based on the weight matrix, the bias vector and the preset real-time environmental variables. The prediction result of the trend prediction model is the product of the output weight matrix and the output matrix; the real-time environmental variables are variables among the environmental variables, which are used to characterize the dynamic change factors of the environment in which the power device is located.

[0089] In one embodiment, the data acquisition module 10 is further configured to: The key parameters are preprocessed to clean abnormal data in the key parameters; wherein the preprocessing includes normalization processing and filtering processing, and the abnormal data includes noise data and abnormal value data.

[0090] Figure 5 An example of a physical structure diagram of an electronic device is shown below. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the steps of the power device life prediction method, for example, including: Collecting key parameters during the operation of the power device to be predicted; the key parameters include at least operating parameters and environmental variables of the power device; Inputting the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; Based on the Kalman filter algorithm, the first prediction result and the second prediction result are fused to obtain a life prediction result of the power device.

[0091] Furthermore, the logic instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0092] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the power device life prediction method provided by the above methods, for example, including: Collecting key parameters during the operation of the power device to be predicted; the key parameters include at least operating parameters and environmental variables of the power device; Inputting the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; Based on the Kalman filter algorithm, the first prediction result and the second prediction result are fused to obtain a life prediction result of the power device.

[0093] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the power device lifetime prediction method provided by the above methods are implemented, for example, including: Collecting key parameters during the operation of the power device to be predicted; the key parameters include at least operating parameters and environmental variables of the power device; Inputting the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; Based on the Kalman filter algorithm, the first prediction result and the second prediction result are fused to obtain a life prediction result of the power device.

[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0095] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for predicting the life of a power device, characterized in that: include: Collecting key parameters during the operation of the power device to be predicted; the key parameters include at least operating parameters and environmental variables of the power device; Inputting the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; Based on the Kalman filter algorithm, the first prediction result and the second prediction result are fused to obtain a life prediction result of the power device.

2. The power device life prediction method according to claim 1, characterized in that: Before inputting the keyword parameters into the pre-trained classification prediction model and trend prediction model respectively, the method further includes: Acquiring historical key parameters corresponding to the power device and constructing a sample data set based on the historical key parameters; the sample data set includes label data, and the label data is used to characterize the remaining service life of the power device; Based on a preset number of iterations, the preset classification prediction model and the preset trend prediction model are jointly trained using the sample data set to obtain a pre-trained classification prediction model and a pre-trained trend prediction model.

3. The power device life state method according to claim 2, characterized in that: The method of jointly training a preset classification prediction model and a preset trend prediction model based on a preset number of iterations using the sample data set to obtain a pre-trained classification prediction model and a pre-trained trend prediction model includes: Iteratively training a preset classification prediction model using the sample data set based on a preset number of iterations to obtain a first prediction accuracy of the classification prediction model, and iteratively training a preset trend prediction model using the sample data set to obtain a second prediction accuracy of the trend prediction model; determining an initial measurement error according to the first prediction accuracy and the second prediction accuracy, and calculating a Kalman gain value of the Kalman filter algorithm based on the initial measurement error and a preset estimation error; The first prediction value and the second prediction value are fused using the Kalman gain value to obtain a fused prediction result, and the estimated error is updated according to the Kalman gain value; the first prediction value is a prediction value corresponding to the prediction result of the classification prediction model, and the second prediction value is a prediction value corresponding to the prediction result of the trend prediction model; The model parameters are adjusted based on the fusion prediction result, and the step of iteratively training the preset classification prediction model based on the preset number of iterations using the sample data set is returned and executed until the number of adjustments to the model parameters reaches the preset number of iteration cycles, thereby obtaining a pre-trained classification prediction model and a pre-trained trend prediction model; the model parameters include the first model parameters of the classification prediction model and the second model parameters of the trend prediction model.

4. The method for predicting the life of a power device according to claim 3, wherein: The iterative training of the preset classification prediction model using the sample data set includes: Inputting the sample data set into a preset classification prediction model, and iteratively training the classification prediction model; In each round of iteration, the function value of the objective function is minimized as the optimization goal, and the first model parameter of the classification prediction model is iteratively optimized based on the preset population optimization algorithm to obtain the optimal parameter combination; In which, the objective function is constructed based on the regularization parameter and the slack variable, and the first model parameter includes the regularization parameter and the slack variable; the slack variable is used to characterize that some predicted values ​​in the prediction results of the classification prediction model are allowed to exceed the preset error range.

5. The method for predicting the life of a power device according to claim 4, wherein: The preset population optimization algorithm is used to iteratively optimize the first model parameters of the classification prediction model to obtain the optimal parameter combination, including: Based on a preset population size, randomly initializing population particles within a parameter space of a first model parameter of the classification prediction model; a particle position of any population particle corresponds to a parameter combination of the first model parameter of the classification prediction model; According to the function value of the objective function corresponding to each of the swarm particles, minimizing the function value of the objective function is used as an optimization goal to determine the individual optimal position and the swarm optimal position of the particles in the swarm; Based on the individual optimal position and the population optimal position, the particle position and particle velocity of the population particles are iteratively updated, and the step of determining the individual optimal position and the population optimal position of the population particles according to the function value of the objective function corresponding to each population particle is returned and executed until the number of iterative updates of the population particles reaches a preset maximum number of iterations, and the optimal parameter combination of the first model parameters is obtained according to the individual optimal position and the population optimal position.

6. The method for predicting the life of a power device according to claim 3, wherein: The trend prediction model includes a hidden layer and an output layer; the iterative training of the preset trend prediction model using the sample data set includes: Inputting the sample data set into a preset trend prediction model, and performing iterative training on the trend prediction model; In each round of iteration, the trend prediction model performs nonlinear mapping on the sample data in the sample data set by randomly generating the weight matrix, bias vector and preset activation function of the hidden layer to form the output of the hidden layer; The output weight matrix is ​​obtained by minimizing the target error between the output weight matrix and the output matrix of the trend prediction model; The output matrix is ​​calculated based on the weight matrix, the bias vector and the preset real-time environmental variables. The prediction result of the trend prediction model is the product of the output weight matrix and the output matrix; the real-time environmental variables are variables among the environmental variables, which are used to characterize the dynamic change factors of the environment in which the power device is located.

7. The method for predicting the life of a power device according to claim 1, wherein: After collecting the key parameters of the power device to be predicted during operation, the method further includes: The key parameters are preprocessed to clean abnormal data in the key parameters; wherein the preprocessing includes normalization processing and filtering processing, and the abnormal data includes noise data and abnormal value data.

8. A power device life prediction device, characterized in that: include: A data acquisition module, configured to acquire key parameters during the operation of the power device to be predicted; the key parameters at least include operating parameters and environmental variables of the power device; a prediction module, configured to input the keyword parameters into a pre-trained classification prediction model and a trend prediction model, respectively, to obtain a first prediction result output by the classification prediction model and a second prediction result output by the trend prediction model; the first prediction result is used to characterize the aging state of the power device, and the second prediction result is used to characterize the changing trend of the aging state of the power device; A fusion module is used to fuse the first prediction result and the second prediction result based on a Kalman filter algorithm to obtain a life prediction result of the power device.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the power device lifetime prediction method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the power device lifetime prediction method according to any one of claims 1 to 7 is implemented.

Images