Improved crayfish optimization algorithm based on multimodal collaborative search and dynamic distribution perturbation for estimating State-of-Health of lithium-ion batteries

Abstract

The crayfish optimization algorithm (COA) is a novel metaheuristic algorithm. In response to issues such as poor search capability, as well as the tendency to fall into premature convergence when COA solves complex optimization problems, an improved crayfish optimization algorithm based on multimodal collaborative search and dynamic distribution perturbation (MDCOA) is proposed. In MDCOA, a multimodal collaborative search strategy is proposed, which consists of two sub-strategies: dimension learning-based hunting (DLH) search and equilibrium hybrid search (EHS). Firstly, the DLH strategy is utilized to expand the neighborhood of crayfish population, enhancing the crayfish's utilization of neighborhood information. Secondly, the EHS is proposed to balance the intensity of global and local searches, and the global optimal solution is updated by comparing the fitness of DLH and EHS. To avoid premature convergence, dynamic distribution perturbation is proposed to nonlinearly disturb the algorithm. To verify the performance of the MDCOA, the parameter sensitivity of the algorithm and the impact of the two improvement mechanisms are analyzed using the CEC 2020 benchmark suite. Subsequently, MDCOA is compared with 18 other algorithms across multiple dimensions using the CEC 2022 and CEC 2017 benchmark suites. To verify the ability of MDCOA to deal with practical problems, it is used to optimize the hyperparameters of the Transformer-LSTM model for establishing a lithium-ion battery State-of-Health (SOH) estimation model. Simulation results based on actual data demonstrate that the Transformer-LSTM model optimized by MDCOA exhibits high estimation accuracy, with R² values above 97%, RMSE below 0.035, and MAE below 0.02 across four different lithium-ion battery datasets under various operating conditions. Therefore, MDCOA can be used to optimize the hyperparameters of Transformer-LSTM and apply it to lithium-ion batteries SOH estimation. The source code of MDCOA is publicly available on https://github.com/yylcsuft/MDCOA.