Accuracy of IGBT Junction Temperature Prediction: An Improved Sailfish Algorithm to Optimize Support Vector Machine

This study improves the accuracy of junction temperature prediction, as the insulated gate bipolar transistor (IGBT) reliability is important for the safe operation of its working system due to junction temperature is limited in its actual performance and reliability. A model based on an improved sailfish optimization algorithm to optimize support vector machine (ISFO-SVM) is proposed to solve the problem that the junction temperature prediction accuracy is not high enough. The proposed algorithm is improved by adaptive nonlinear iterative factor, Le'vy flight and differential mutation strategy to optimize the support vector machine (SVM) internal parameters to predict junction temperature. The results indicate that ISFO-SVM performs better under the same evaluation indexes. The root mean squared error average value decreased by 67.189%, and the mean absolute percentage error average value decreased by 63.189%, compared with the sailfish optimization algorithm to optimize the SVM. The prediction error of ISFO-SVM is smaller and the error value is in the [−5 °C, 5 °C] range accounting for 98.270% of the total test samples. ISFO-SVM has a higher fitting degree than the actual junction temperature and the R2 has reached 99.660%. The model predicts the junction temperature of IGBT modules and provides scientific guidance for system reliability evaluation to maintain safe and stable operation effectively.