Bayesian Optimisation in Deep Learning for Electric Vehicle SOC Prediction
DOI:
https://doi.org/10.63001/tbs.2025.v20.i01.S1.pp17-21Keywords:
SOC, Lithium-Ion Battery, DL, Bayesian Optimization, CNN2D, BI-LSTM, RMSE, Max Error, Electric Vehicles, Hyperparameter Tuning, Feature Selection, Flask ApplicationAbstract
SOC estimate is essential for electric car lithium-ion battery efficiency. Traditional deep learning models optimized using Grid Search often fail to achieve optimal hyperparameters, leading to reduced prediction accuracy. To address this, we employed Bayesian Optimization, which efficiently selects the best hyperparameters by leveraging past evaluations. Our study evaluated multiple DL models, including GRU, LSTM and BI-LSTM, optimized with 70 neurons, where BI-LSTM achieved the lowest RMSE and Max Error. As an extension, we implemented the CNN2D algorithm, known for its superior feature selection and optimization capabilities using convolutional layers and MaxPooling2D. CNN2D outperformed previous models with reduced RMSE and Max Error. Additionally, we developed a Flask-based application enabling users to predict SOC by uploading CSV data, enhancing accessibility and usability.
