Scholarlink

Autonomous electric cars (AEVs) rely on advanced driver assistance technology and electric propulsion, demanding sophisticated energy management systems (EMS). The state of charge (SOC) of batteries is critical for efficiency, range, power delivery, and overall driving experience. Traditional methods, such as voltage- and current-based approaches, struggle to give precision under fluctuating conditions, and present strategies are primarily model-based, making them inefficient for learning unpredictable battery states in complex real-world scenarios. Thus, advanced techniques like physics-guided deep learning are being utilized to forecast SoC in battery systems. Initially, data is gathered from the car battery dataset. The gathered data is pre-processed using Test-time batch normalization (TTN) and missing value imputation with the Collaborative matrix factorization (CMF-Impute) technique. The dataset’s dimensionality is subsequently decreased through Pairwise Controlled Manifold Approximation Projection (PaCMAP). The Parrot optimizer (PO) is used to select the optimal number of neighbor pairs in PaCMAP. Physics-guided residual neural networks (PhyResNet) are used to forecast SoC in autonomous electric vehicle (AEV) batteries. PhyResNet outperforms SVM, BPNN, ANN, RNN, and DT in prediction accuracy with 96.80% accuracy. The results revealed that the PhyResNet model made the most accurate predictions during the testing stage. It had the highest determination coefficient (R2) of 94% and the smallest root-mean-squared error of 1.26%. Thus, the proposed method enables accurate SoC forecasting and improves energy management in AEVs to optimize battery consumption.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient State of Charge Forecasting in Autonomous Electric Vehicles Using Physics-Gded Residual Neural Networks and Optimized Dimensionality Reduction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • K. P. Senthilkumar,
  • E. Anbalagan

摘要

Autonomous electric cars (AEVs) rely on advanced driver assistance technology and electric propulsion, demanding sophisticated energy management systems (EMS). The state of charge (SOC) of batteries is critical for efficiency, range, power delivery, and overall driving experience. Traditional methods, such as voltage- and current-based approaches, struggle to give precision under fluctuating conditions, and present strategies are primarily model-based, making them inefficient for learning unpredictable battery states in complex real-world scenarios. Thus, advanced techniques like physics-guided deep learning are being utilized to forecast SoC in battery systems. Initially, data is gathered from the car battery dataset. The gathered data is pre-processed using Test-time batch normalization (TTN) and missing value imputation with the Collaborative matrix factorization (CMF-Impute) technique. The dataset’s dimensionality is subsequently decreased through Pairwise Controlled Manifold Approximation Projection (PaCMAP). The Parrot optimizer (PO) is used to select the optimal number of neighbor pairs in PaCMAP. Physics-guided residual neural networks (PhyResNet) are used to forecast SoC in autonomous electric vehicle (AEV) batteries. PhyResNet outperforms SVM, BPNN, ANN, RNN, and DT in prediction accuracy with 96.80% accuracy. The results revealed that the PhyResNet model made the most accurate predictions during the testing stage. It had the highest determination coefficient (R2) of 94% and the smallest root-mean-squared error of 1.26%. Thus, the proposed method enables accurate SoC forecasting and improves energy management in AEVs to optimize battery consumption.