Abstract:

State estimation and active equalization of lithium batteries are key technologies to improve battery performance and extend service life. Aiming at the problem that state of charge(SOC) estimation method of parameter model ignores the actual working conditions of electric vehicles and causes large estimation deviation, an extreme learning machine(GA-ELM) neural network algorithm is proposed based on genetic algorithm to estimate SOC of battery. The parameters of ELM are optimized by genetic algorithm to improve the estimation accuracy and generalization ability, and the training and testing are carried out under UDDS working condition data. At the same time, the bidirectional Buck-Boost balanced topology can quickly transfer energy between cells, while reducing the complexity of the transfer path. The SOC estimated by the extreme learning machine of genetic algorithm is used as the equilibrium variable, and the experiment is carried out by Matlab/Simulink simulation platform. The results show that the average error of the GA-ELM neural network proposed in this paper is 0.15%, while that of the traditional ELM neural network is 0.56%. Therefore, the neural network proposed can estimate SOC more accurately. At the same time, the energy balance between the battery pack can be quickly completed, which proves the feasibility of the scheme.

Key words: Genetic algorithm, Buck-Boost circuit, SOC estimation, extreme learning machine, battery equalization