Abstract:

Fiber Bragg grating(FBG) sensor is gradually emerging as an innovative solution to enhance the accuracy of state of charge(SOC) estimation for lithium-ion batteries.However, both the positions and the number of strain sensors on the surface of lithium-ion batteries affect the accuracy of SOC estimation.Therefore, a multi-strain feature selection method based on the sequential forward selection(SFS) algorithm is proposed.To address the issue of fluctuations in model generalization ability existing in the SFS algorithm during the feature selection process, the elastic net and leave-one-out cross-validation(LOOCV) mechanisms are introduced to improve the stability of feature selection and the generalization performance of the model.Subsequently, the genetic algorithm-optimized backpropagation neural network(GA-BP) is utilized to estimate the SOC of lithium-ion batteries.Experimental results show that the root mean square error(RMSE) of lithium-ion battery SOC estimation is reduced to 0.635%, the mean absolute error(MAE) is 0.419%, and the coefficient of determination(R²) reaches 0.999 after feature selection.The proposed method provides a solution to the feature selection problem of multi-position FBG strain signals and offers technical reference for high-precision SOC estimation of lithium-ion batteries based on optical fiber sensor.

Key words: fiber Bragg grating(FBG), strain, lithium-ion battery, state of charge(SOC) estimation, feature screening