Enhanced quantile regression long short-term memory hybrid neural network for the state of charge point and interval estimation of lithium-ion batteries

The state of charge (SOC) estimation accuracy of lithium-ion batteries directly affects the reliability and management efficiency of clean energy storage systems. However, due to the nonlinear characteristics of batteries and complex working conditions, there are still significant challenges in high-precision SOC estimation. Therefore, this paper proposes a hybrid neural network model based on long short-term memory (LSTM). Specifically, the model extracts multidimensional features through two-dimensional convolution and LSTM neural network with attention mechanism is performed for estimation. In addition, the quantile regression loss function is used in the training of the hybrid neural network to give it confidence interval estimation capability. Finally, the experimental data of different working conditions at multiple temperatures were utilized to validate and analyze the proposed method. The results show that the proposed estimation method has an MAE less than 0.58 %, an MSE less than 0.008 %, an RMSE less than 0.81 %, an R2 greater than 99.91 %, and a stable confidence interval estimation capability. In summary, this paper innovatively proposes an effective SOC estimation solution, which provides new ideas for future SOC estimation of energy storage battery management systems, and has important theoretical and practical application significance.