Comparative Study of Different Deep Learning Approaches for State of Charge Estimation in Lithium Iron Phosphate Battery

State of charge (SOC) is one of the main tasks of battery management systems. Since the SOC is an unmeasurable parameter, several methods have been developed to estimate its value. With the rapid rise of artificial intelligence, subsequently, deep learning methods have gained a lot of attention in this field due to their robustness and high prediction accuracy. This paper investigates three sequence models, a subunit of deep learning, for SOC estimation. The algorithms are long short-term memory (LSTM), bidirectional long short-term memory with attention mechanism (BLSTM-AM), and finally transformer network. For training, a dataset of lithium iron phosphate (LFP) was used, where the battery underwent a US06 test at 25°C. Afterward, the robustness of the methods was evaluated using a testing dataset of the same battery at an ambient temperature but with a dynamic stress test (DST). Finally, the performance of the three methods was compared using root mean squared error (RMSE) and r-squared as metrics. The results showed the transformer network as the most robust one while the LSTM network gave the least pleasant results.