Remaining useful life prediction for lithium-ion batteries based on an improved GWO–SVR algorithm

Lithium-ion batteries have been applied in civil aircraft such as the B787 with excellent performance. As the service time of lithium-ion batteries increases, their performance continues to decline. Therefore, accurately predicting the remaining useful life of lithium-ion batteries is helpful for timely maintenance or replacement, which is important for flight safety. This study extracts features from charge and discharge data of lithium-ion batteries with incremental capacity analysis to predict the remaining useful life of lithium-ion batteries. To this end, this study calculates the degree of correlation between the features and battery capacity based on grey correlation analysis, and then accordingly filters the features. Finally, a prediction method for the remaining useful life of lithium-ion batteries is proposed based on improved grey wolf optimization (IGWO) and support vector regression (SVR). The IGWO algorithm is proposed to solve the issue wherein grey wolf optimization (GWO) is prone to stagnation at local optima. As a research hotspot in the field of optimization algorithms in recent years, GWO has excellent optimization performance. However, it faces the problem of falling into local optimization and premature convergence in practical applications. To solve this problem, this study proposes IGWO to optimize and rewrite the position update equation and add memory and flight functions to each individual in the wolf pack so as to enhance the global search ability of the algorithm and improve its convergence speed. Furthermore, IGWO uses skew tent mapping to generate chaotic sequences to optimize the initial distribution of the grey wolf pack in the optimization space. Thus, it achieves a more uniform initial distribution effect than the traditional random generation method. This paper conducts an optimization comparison experiment based on commonly used benchmark functions to compare the optimization ability of GWO before and after improvement. The results show that the IGWO algorithm effectively avoids the stagnation at a local optimal value that the GWO algorithm will fall into, with faster convergence speed and better optimization than GWO for almost all functions. In several of these test functions, the optimization accuracy of IGWO is dozens of times higher than that of GWO. The remaining useful life prediction abilities of IGWO-SVR, GWO-SVR, and SVR are compared based on the NASA lithium-ion battery dataset. The results show that the model trained with IGWO-SVR achieves higher prediction accuracy on the data among all four batteries, and the root mean square error of the prediction results is reduced by more than 10% compared with GWO-SVR.