Whale Optimization Algorithm for Optimizing Extreme Learning Machine in IGBT Aging Fault Prediction

The Insulated Gate Bipolar Transistor (IGBT) aging faults seriously affect the stable operation of power systems; therefore, it is necessary to predict these aging failures to enhance system safety. This paper proposes a prediction model for IGBT aging faults based on a Whale Optimization Algorithm (WOA) optimized Extreme Learning Machine (ELM). Using the accelerated aging dataset provided by the NASA Prognostics Center, the peak value of the collector-emitter turn-off voltage spike is selected as the failure prediction feature parameter. Exponential smoothing is employed for data preprocessing to reduce data volatility while preserving the original developmental trend, thus improving the network model's learning of the data's intrinsic characteristics. The WOA is utilized to optimize the ELM's hidden layer weights and biases, enhancing its performance. The fault diagnosis performance of the WOA-ELM model is validated against a fault test set and compared with the predictive results of the ELM and Long Short-Term Memory (LSTM) models. The tests show that the suggested framework achieves a high level of fault prediction precision, rendering it appropriate for forecasting aging faults in IGBTs and various electronic parts.