An improved random drift particle swarm optimization-feed forward backpropagation neural network for high-precision state-of-charge estimation of lithium-ion batteries

Abstract

A predictive model with high accuracy and stability of the state of charge (SOC) estimation for lithium-ion batteries plays a significant role in electric vehicles. An improved random drift particle swarm optimization-feed forward backpropagation neural network (IRDPSO-FFBPNN) is established in this paper. Basically, a three-layer FFBPNN is established, and its learning process is analyzed in detail. Then, to avoid the particle out-of-control, inducting weight parameter σ to achieve dynamic control weight convergence. What's more, the cross-reorganization of data is proposed to enhance the utilization. Finally, a further performance comparison with other networks is made under different working conditions to prove the effectiveness of the IRDPSO-FFBPNN. The experimental results showed that the maximum SOC error of the IRDPSO-FFBPNN is 0.1021 % in 45 s, 0.1237 % in 116 s under BBDST and DST with different temperatures, respectively, which performed better both in terms of time-consumption and accuracy.