Abstract:

In response to the issue of incomplete fiber bragg grating(FBG) sensor data in lithium battery monitoring, a method for FBG strain data completion based on non-linear independent components estimation(NICE) is proposed. To optimize the annealing parameters in the NICE model, the particle swarm optimization(PSO) algorithm is employed for adaptive parameter optimization, thereby improving the quality of data generation. On this basis to further enhance the effectiveness of fiber Bragg grating sensing data in state of charge(SOC) estimation for lithium-ion batteries, an SOC estimation model based on an attention mechanism and bidirectional gated recurrent unit(Bi-GRU-Att) is constructed in this work. Experimental results show that the proposed PSO-NICE algorithm significantly reduces the earth mover’s distance compared to the generative adversarial network(GAN) data generation algorithm at data missing rates of 10%, 30%, 50%, and 70%. Notably, at a missing rate of 70%, the EM distance is reduced by 73.41%. Compared with traditional zero-value imputation, the proposed data completion method reduces the root mean square error(RMSE) and mean absolute error(MAE) in SOC estimation by 42.384% and 37.256%, respectively. The proposed approach provides an effective solution and technical reference for addressing fiber-optic sensing data loss in practical applications.

Key words: fiber bragg grating, lithium batteries, missing data, completion methods, state of charge estimation