State of charge estimation under complex conditions using Latin hypercube sampling and levy flight enhanced adaptive weighted particle swarm optimization

Accurate state-of-charge (SOC) estimation of lithium-ion batteries under fluctuating ambient temperatures and complex operating conditions remains a critical challenge for battery management systems. To address this issue, an improved particle swarm optimization algorithm, termed L2AWPSO, is proposed in this paper. By integrating Latin hypercube sampling (LHS), Lévy flight, and adaptive weight strategies, the proposed algorithm enhances global search capability and improves convergence stability. Based on the proposed optimization strategy, two novel SOC estimation methods are further developed: L2AWPSO-AEKF and L2AWPSO-LSTM-UKF. The L2AWPSO-AEKF method employs the optimizer to dynamically match the noise covariance window length M[jls-end-space/], thereby improving robustness under modeling uncertainty and time-varying noise conditions. In contrast, the L2AWPSO-LSTM-UKF method combines the strong feature extraction capability of long short-term memory (LSTM) networks with the robust noise suppression characteristics of the unscented Kalman filter, enabling high-accuracy SOC estimation under time-varying temperature conditions. Extensive experiments conducted under multiple operating conditions demonstrate that both hybrid methods consistently achieve lower prediction errors than comparison approaches. Moreover, L2AWPSO-AEKF significantly reduces training time and improves online response efficiency, while L2AWPSO-LSTM-UKF exhibits superior adaptability under severe temperature variations. These results verify the effectiveness and adaptability of the proposed methods for SOC estimation under complex and temperature-fluctuating operating conditions.