An accurate state of charge (SOC) estimation is important for ensuring the safe operation of electric vehicles (EVs). Recurrent Neural Networks (RNNs), known for their time-sequence capabilities, offer advantages in battery SOC estimation. Bidirectional RNNs (BiRNN), in particular, have recently been introduced in this domain. Conventional testing methods for BiRNN often utilize the entire dataset, allowing the model to access future data for present SOC estimation, which is impractical for real-world applications. To address this issue, this paper proposes a real-time testing method that restricts the BiRNN to historical data by employing a sliding window. For comparison, the BiRNN is also tested with a same-sized sliding window containing both historical and future data. Furthermore, this paper provides a comparative evaluation of unidirectional RNNs and BiRNNs for SOC estimation. The results reveal that the performance of BiRNNs under the historical data scenario is similar to unidirectional RNNs, with a minor improvement of an average 0.08% mean absolute error (MAE) on the testing dataset. When exposed to future data, BiRNNs demonstrate an average MAE improvement of 0.25%.
A Comparison Study of Unidirectional and Bidirectional Recurrent Neural Network for Battery State of Charge Estimation
2024-06-19
2798373 byte
Conference paper
Electronic Resource
English
Unidirectional and Bidirectional Battery State-of-Charge Indicators and Charge Controllers
Automotive engineering | 1978
|