摘要
对结温进行精准预测和提取仍然是SiC MOSFET器件应用中需要攻克的技术难题,对此,结合粒子群优化-反向传播(PSO-BP)神经网络预测精度高以及径向基函数(RBF)神经网络函数逼近能力强等优点,提出了一种基于PSO-BP与RBF级联神经网络的SiC MOSFET结温预测模型。模型将SiC MOSFET的漏极电流和通态电压作为PSO-BP神经网络的输入,再将PSO-BP神经网络的结温预测值以及插值法得到的温度插值作为RBF神经网络的输入。基于LTspice仿真获取的数据集验证了模型对SiC MOSFET结温预测的有效性。结果表明,该模型对SiC MOSFET结温预测的绝对误差在0.005℃以内,均方根误差和平均绝对误差分别低至0.0082和0.0015,相比单一的BP、PSO-BP以及RBF神经网络模型,其预测精度得到了大大提高。该级联神经网络模型可实现对SiC MOSFET结温(> 75℃)的精准预测。
Accurate prediction and extraction of the junction temperature is still a challenge in the application of SiC MOSFET.A prediction model based on cascaded PSO-BP and RBF neural networks was proposed by combining the high prediction accuracy of particle swarm optimization-back propagation(PSO-BP)network and the powerful function approximation capacity of radial basis function(RBF)neural network.In the proposed model,the drain current and on-state voltage were introduced as the inputs of PSO-BP neural network.The predicted junction temperature value from PSO-BP neural network and the interpolated temperature value were taken as the inputs of the RBF neural network.Based on the LTspice simulation data set,the effectiveness of the model in MOSFET junction temperature prediction was validated.The results show that the absolute error of the model is lower than 0.005℃,and the root mean square error and mean absolute error are as low as 0.0082 and 0.0015,respectively.Compared with single BP,PSO-BP and RBF neural networks,the prediction accuracy was greatly improved.An accurate prediction of SiC MOSFET junction temperature(>75℃)can be achieved with the proposed model.
作者
朱靖
刘新超
谷晓钢
黄玲琴
ZHU Jing;LIU Xinchao;GU Xiaogang;HUANG Lingqin(College of Electrical Engineering and Automation,Jiangsu Normal University,Xuzhou 221000,Jiangsu Province,China)
出处
《电子元件与材料》
CAS
北大核心
2023年第4期458-466,共9页
Electronic Components And Materials
基金
国家自然科学基金(62074071,61801197)
2021江苏高校“青蓝工程”资助项目
江苏省研究生科研与实践创新计划项目(KYCX21_2624)
江苏省现代教育技术研究所规划课题(2021-R-91616)。
