11 June 2026 in Journal of the Electrochemical Society
This research tackles the challenges of elevated parameter sensitivity and substantial model error in estimating the State of Charge (SOC) of lithium-ion batteries by introducing an Unscented Kalman filter (UKF) method that combines NDO optimisation with neural network-based online compensation. Initially, the state-space equations of the battery are formulated using a second-order RC equivalent circuit model, and the UKF is employed for preliminary State of Charge estimation. Secondly, the NDO algorithm is implemented to achieve global optimisation of the process noise Q and observation noise R in the UKF. To overcome the constraints of conventional single-error-metric methodologies, a weighted objective function is formulated, integrating the mean absolute error (MAE) and root mean square error (RMSE) of SOC estimation in a weighted ratio of 0.8MAE + 0.2RMSE, thus achieving a balance between overall accuracy and error stability. A single-hidden-layer feedforward neural network is developed and trained online utilising the backpropagation algorithm to dynamically address voltage prediction residuals. In comparison to conventional UKF and PSO-UKF approaches, the estimation error is markedly diminished, substantiating the efficacy and superiority of the suggested method in complicated operational scenarios.