基于BP神经网络的IGBT模块开关损耗求解

如何准确求解绝缘栅双极晶体管(IGBT)模块的开关损耗值,是电力变换器性能和寿命研究中的关键问题之一。针对现有IGBT开关损耗模型难以准确求解开关损耗值的缺陷,引入了基于粒子群算法优化的误差反向传播(BP)前馈神经网络模型。将影响开关损耗的5个主要因素(集射工作电压、集电极电流、驱动电压、驱动电阻、结温)作为BP神经网络的输入向量,并采用粒子群算法优化网络的初始权值与阀值,通过共轭梯度法的学习规则加速收敛,从而获得开关损耗的精确求解值。该模型实现了在额定值范围内对各种工况下的IGBT模块开关损耗值的可靠预测,其在100组测试验证样本下所出现的最大误差比率为3.85%。

Solution of Switching Loss of IGBT Module Based on Back Propagation Neural Network

How to solve the switching loss value of the IGBT module accurately is a key problem in the performance and lifetime studies of power converter.Aiming at the disadvantages that the existing IGBT switching loss solving models are difficult to accurately solve the switching loss value,the error back propagation(BP)neural network model optimized by the particle swarm algorithm is proposed.With taking the 5 main factors(Collector-emitter voltage,collector current,gate drive voltage,gate drive resistance,junction temperature)that affect the press-contact IGBT module switching loss as the input vector,and optimizing the initial weight and threshold of BP neural network by the particle swarm algorithm,the exact solution of the switching loss is obtained accordingly.The model realizes the reliable prediction of the switching loss of the IGBT module on various working conditions in the rated range,and the maximum error rate in 100 sets of validation samples is 3.85%.