New Filter-based Supervised Learning Approach for State of Charge Estimation of Li-Ion Battery for EV Applications

The estimation of the state of charge of Li-ion batteries is an important metric. The appropriate estimation plays an important role in assessing the performance of the batteries. The limited instruments are available for such purpose with certain limitations such as large time requirements etc. Numerous other immeasurable elements, such as battery chemistry, the surrounding environment, the aging factor, etc., may also have an impact on the SOC. A new approach based on a filter technique with supervised learning is proposed in this paper. A long-short-term memory and hybrid neural network combining the benefits of a convolutional neural network (CNN) and a bidirectional long-short-term memory (BiLSTM) neural network is proposed. CNN automatically decreases the dimensionality and extracts the deep features from the battery dataset. The obtained data is fed into the LSTM. As a result, minimizes overfitting and speeds up the learning process. Training the dataset model takes considerably less time, and the estimation results are more precise. The results demonstrate that the proposed hybrid model is computationally efficient and provides better estimation results with the single model. The estimation performances are below 1% and have superior estimation accuracy and adaptability in comparison to existing literature.