Residual Life Prediction of Rolling Bearings Based on Transformer-BiGRU-Attention Model with Improved Sparrow Optimization Algorithm

Remaining useful life (RUL) prediction is one of the core problems of PHM, and the traditional deep learning method has low accuracy in predicting the remaining useful life of rolling bearings. A hybrid deep learning model of Transformer-BiGRU-Attention based on improved sparrow optimization algorithm is proposed for rolling bearing remaining useful life prediction. The current sparrow optimization algorithm has the problems of slow convergence, lack of robustness and falling into local optimum. Therefore, an enhanced version of the sparrow search algorithm, termed the Chaotic Mapping-Sparrow Search Algorithm (CSSA), is introduced. The utilization of chaotic cube mapping for initializing the population enhances the quality and diversity of distribution within the population. This approach effectively prevents the algorithm from getting trapped in local optima and optimizes both the speed of convergence and the accuracy of the solutions. Tests on RUL prediction of rolling bearings show that the proposed method can more accurately achieve the RUL prediction of bearings under variable operating conditions, and the accuracy of RUL prediction results is higher than that of current methods.