反向传播
人工神经网络
碳化硅
MOSFET
晶体管
均方误差
电子工程
功率(物理)
均方根
计算机科学
控制理论(社会学)
线性回归
半导体器件建模
相关系数
可靠性(半导体)
还原(数学)
阈值电压
算法
能量(信号处理)
电压
回归
近似误差
功率半导体器件
逻辑门
材料科学
节点(物理)
最大功率原理
能源消耗
CMOS芯片
平均绝对百分比误差
回归分析
切换时间
作者
Youyang Wang,Wenxiao Wang,Shilong Zhu,Hui Zhu,Yu Zhang,Daohui Li,Xiaobo Zhu,Wenhua Gu
标识
DOI:10.1109/tpel.2025.3642243
摘要
Accurate and rapid determination of switching loss is crucial for energy consumption assessment and system cooling design of silicon carbide (SiC) metal-oxide-semiconductor field-effect transistor (MOSFET) power modules. However, traditional switching loss prediction methods struggle to balance low modeling complexity with high prediction accuracy. This article proposes an effective switching loss prediction method for SiC MOSFET power modules based on multi-layer backpropagation artificial neural network (ANN). The proposed method accurately models the regression relationship between the static parameters and switching loss of SiC MOSFETs, allowing for the direct prediction of switching loss using static parameters obtained from tests or datasheets. By leveraging static parameters that inherently encode structural and parasitic effects, the model adapts to both intra-batch and inter-batch manufacturing fluctuations without explicit physical parameter tuning. Experimental results conducted on real test-acquired datasets demonstrate that the proposed method can achieve a minimum mean absolute percentage error (MAPE) and root mean squared error (RMSE) of 1.13% and 7.26×10-4, a maximum coefficient of determination (R2) of 0.881, with a maximum error of no more than 7.43% and average prediction time of 4.95 ms per module.
科研通智能强力驱动
Strongly Powered by AbleSci AI