High Precision IGBT Health Evaluation Method: Extreme Learning Machine Optimized by Improved Krill Herd Algorithm

Abstract

The insulated gate bipolar transformer (IGBT) is widely used in industry, aerospace, renewable energy and other fields, which requires high reliability. However, in the research field of IGBT health evaluation, there are still problems that the accuracy is not high enough and the reliability manager of IGBT cannot accurately grasp. In this study, an IGBT health status evaluation model based on improved krill herd optimized extreme learning machine (IKH-ELM) is proposed to accurately evaluate the health status of IGBT, provide guidance for timely replacement of components, thus reducing the failure probability and maintenance cost, and improving the system reliability. Firstly, the power cycle accelerated aging test, short pulse-high amplitude current test and thermal resistance measurement test are carried out to obtain the failure predictor parameters data of IGBT. Then, an IGBT health status classification method based on cloud model is proposed to make the health status assessment results more scientific. In addition, since the random selection of weights and thresholds of ELM affects the performance of the model, the krill herd algorithm is improved to optimize ELM for obtaining better prediction ability. Based on this, a prediction model based on IKH-ELM is established. Finally, the proposed model is proved to be more accurate and effective under the same experimental environment and the same index.