Optimizing electric vehicle energy consumption prediction through machine learning and ensemble approaches

Area of Science:

• Energy Systems
• Machine Learning
• Transportation Engineering

Background:

• Accurate electric vehicle (EV) energy consumption prediction is vital for efficiency and infrastructure planning.
• Challenges exist due to complex interactions between driving conditions, vehicle specs, and environment.
• Real-world data analysis is key to understanding and optimizing EV energy usage.

Purpose of the Study:

• To develop and validate a data-driven machine learning approach for predicting EV energy consumption.
• To systematically compare hyperparameter optimization methods for K-Nearest Neighbors (KNN) regression.
• To create and evaluate a stacking hybrid ensemble model for enhanced prediction accuracy.

Main Methods:

• Utilized an extensive real-world dataset from Colorado for analysis.
• Employed K-Nearest Neighbors (KNN) as a base model with hyperparameter optimization via GridSearchCV, RandomizedSearchCV, Optuna, and Particle Swarm Optimization (PSO).
• Developed a stacking hybrid ensemble model combining KNN with tree-based models and incorporated novel temporal feature engineering.

Main Results:

• The stacking hybrid ensemble model demonstrated superior performance, achieving the lowest prediction errors (MAE = 0.645880, RMSE = 1.788540) and highest accuracy (R² = 0.960078).
• Optuna was identified as the most effective hyperparameter optimization technique for the KNN model.
• Temporal feature extraction and optimized ensemble modeling significantly boosted prediction accuracy.

Conclusions:

• Ensemble learning and advanced optimization methods are highly effective for improving EV energy consumption prediction.
• The developed model offers a deployable tool for EV manufacturers and policymakers in sustainable energy management.
• Optimized ensemble modeling with temporal features provides a robust solution for accurate EV energy usage forecasting.