基于遗传算法优化卷积神经网络的IGBT剩余寿命预测

Prediction Remaining Life of IGBT Using Convolutional Neural Network Optimized by Genetic Algorithm

摘要: 绝缘栅双极性晶体管(Insulated gate bipolar transistor, IGBT)是一种常用于功率电子设备的半导体器件，在电力变换和驱动系统中有着广泛的应用，其稳定性和可靠性对电力系统和工业应用具有重要意义。传统的基于数据驱动的方法针对不同场景无法做到统一，且其预测过程不够直观。本文提出一种更具普适性且直接预测出剩余寿命的方法，首先对IGBT加速老化时的稳态数据和瞬态数据进行分析，提取出和老化相关的9个特征因素，建立卷积神经网络的IGBT剩余寿命预测模型，并用遗传算法进行优化，以提高预测模型的性能和收敛速度。结果表明，基于遗传算法改进的卷积神经网络预测模型准确率高，误差小，相比于SVR、MLP和CNN网络，更能准确地预测出IGBT的剩余寿命。

Abstract: Insulated gate bipolar transistor (IGBT) is a semiconductor device commonly used in power electronic devices. It has a wide range of applications in power conversion and drive systems, and its stability and reliability are of great significance for power systems and industrial applications. Traditional data-driven methods cannot achieve uniformity for different scenarios, and their prediction process is not intuitive enough. This article proposes a more universal and direct method for predicting the remaining lifespan of IGBT. Firstly, the steady-state and transient data during accelerated aging of IGBT are analyzed, and 9 characteristic factors related to aging are extracted. A convolutional neural network model for predicting the remaining lifespan of IGBT is established, and optimized using genetic algorithm to improve the performance and convergence speed of the prediction model. The results show that the improved convolutional neural network prediction model based on genetic algorithm has high accuracy and small error, and can more accurately predict the remaining life of IGBT compared to SVR, MLP, and CNN networks.