Lightweight machine learning framework using temporal features for electric vehicle demand response forecasting on...

Area of Science:

• Energy Management
• Machine Learning
• Electric Vehicles

Background:

• Rising electric vehicle (EV) adoption presents significant challenges for grid energy management, particularly in resource-constrained settings.
• Effective demand response (DR) strategies are crucial for balancing energy supply and demand with increasing EV integration.

Purpose of the Study:

• To develop and evaluate lightweight machine learning (ML) models for accurate EV charging load forecasting.
• To optimize various demand response (DR) strategies using predicted EV load profiles.
• To assess model performance in terms of prediction accuracy and computational efficiency for low-resource environments.

Main Methods:

• Utilized a Kaggle dataset of time-series EV charging data, performing preprocessing, down-sampling, and feature engineering.
• Implemented and compared five ML models: Linear Regression (LR), Support Vector Regression (SVR), k-Nearest Neighbours (kNN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost).
• Evaluated seven DR strategies (Peak Clipping, Valley Filling, Load Shifting, Load Levelling, Strategic Load Growth, Strategic Conservation, Flexible Load Shape) using MAE, RMSE, and R² metrics.

Main Results:

• XGBoost demonstrated the highest accuracy, achieving an R² score of 0.975 for Strategic Conservation and 0.943 for Valley Filling.
• Random Forest also performed well, with an R² score of 0.91, indicating strong predictive capabilities.
• Linear Regression and kNN models showed significantly lower performance, with R² values rarely exceeding 0.50 across most DR strategies.

Conclusions:

• Lightweight ML models are capable of delivering high performance for EV load prediction and DR modeling.
• These models offer scalable solutions for grid operators and policymakers in environments with limited computational resources.
• The findings highlight the potential of optimized DR strategies powered by efficient ML for managing EV energy demand.