An improved transformer-GRU neural model optimized by polar light optimizer for SOC estimation of lithium-ion batteries under complex operating conditions

Abstract

Estimating the battery’s state-of-charge (SOC) is essential for determining how safe electric cars are and their remaining range. An SOC estimation technique for lithium-ion batteries based on the Transformer architecture is presented in this paper. In order to effectively interpret the original data information, the variational mode decomposition (VMD) algorithm is applied to decompose the Panasonic datasets, enabling effective interpretation of the original data information by isolating intrinsic mode functions (IMFs) with distinct frequency characteristics. The decomposition state is then evaluated using the center-frequency method. After that, the Transformer is altered by giving the decoder more positional encoding. The problem of manually setting the network hyper-parameters in SOC estimation is finally resolved by optimizing the tuned Transformer neural network’s learning rate parameters, regularization coefficients, and the number of self-attention mechanism heads using the polar lights optimization algorithm. This optimization technique guarantees that the model can more successfully adjust to the varied data characteristics of particular application scenarios while maintaining Transformer-GRU’s benefits in terms of long-range dependency modeling and low computational cost. The accuracy, stability, and applicability of the method were verified through experimental comparison of various estimation methods, working conditions, and temperature conditions.