State of Charge Estimation for Lithium-Ion Batteries Based on TCN-LSTM Neural Networks
Chunsheng Hu, Fangjuan Cheng, Liang Ma, Bohao Li
Abstract
Accurately estimating the state of charge (SOC) of lithium-ion batteries is critical for developing more reliable and efficient operation of electric vehicles. However, the commonly used models cannot simultaneously extract effective spatial and temporal features from the original data, leading to an inefficient SOC estimation. This paper proposes a novel neural network method for accurate and robust battery SOC estimation, which incorporates the temporal convolutional network (TCN) and the long short-term memory (LSTM), namely TCN-LSTM model. Specifically, the TCN is employed to extract more advanced spatial features among multivariate variables, and the LSTM captures long-term dependencies from time-series data and maps battery temporal information into current SOC and historical inputs. The proposed model performs well in various estimation conditions. The average value of mean absolute error, root mean square error, and maximum error of SOC estimation achieve 0.48%, 0.60%, and 2.3% at multiple temperature conditions, respectively, and reach 0.70%, 0.81%, and 2.7% for a different battery, respectively. In addition, the proposed method has better accuracy than the LSTM or TCN used independently and the CNN-LSTM network. The computational burden with varying length of input is also investigated. In summary, experiment results show that the proposed method has excellent generalization and robustness.