基于PSO-BP神经网络模型的IGBT老化预测

Ageing Prediction of IGBT Based on PSO-BP Neural Network Model

IGBT(绝缘栅双极型晶体管)在使用过程中会产生热量,在其内部产生疲劳累积,最终老化失效。而在IGBT老化过程中,其输出电气参数会发生变化,可据此判断IGBT的老化状态。为此,基于NASA PCoE研究中心提供的IGBT加速老化数据,选取IGBT关断过程中集射极电压尖峰值为失效特征参数;采用卡尔曼滤波对该参数进行预处理后,应用BP神经网络建立IGBT老化预测模型;分别使用遗传算法和粒子群优化算法对BP神经网络初始权值及阈值进行寻优,解决其易陷入局部最优的缺陷;应用均方误差、平均绝对误差、相关系数作为评价指标,对网络性能进行评估。结果表明,经优化后的网络效果均优于BP神经网络,而其中以PSO-BP算法所构建的IGBT老化预测模型为最优模型,可以更准确地实现IGBT的老化预测。

IGBT(insulated gate bipolar transistor) generates heat during use, causing inside fatigue buildup and ultimately ageing failure. In the ageing process of the IGBT, the output electrical parameters will change,and thus the ageing state of the IGBT is judged. Therefore, based on the IGBT accelerated aging data provid ed by the NASA PCoE research center, the peak value of the collector voltage(Vce) in the IGBT turn-off process is selected as the failure characteristic parameter;after preprocessing with Kalman filter, BP neural network is used to establish IGBT ageing prediction model;the GA and PSO algorithms are used to optimize the initial weights and thresholds of BP network respectively to overcome its proneness to local optimal solution;the mean square error(MSE), the mean absolute error(MAE), and the correlation coefficient are used as evaluation indicators to evaluate the network performance. The results show that the optimized network effect is better than the BP network, and the IGBT ageing prediction model constructed by PSO-BP algorithm is the optimal model that can accurately predict IGBT ageing.