Electric vehicle energy consumption prediction using stacked generalization: an ensemble learning approach

In this paper, we present an ensemble stacked generalization (ESG) approach for better prediction of electric vehicles (EVs) energy consumption. ESG is a weighted combination of multiple base regression models to one meta-regressor, which enhances the model prediction and decreases the model variance over a single regressor model. For the current study, we develop ESG by combining three individual base machine learning algorithms, i.e., Decision Tree (DT), Random Forest (RF), and K-Nearest Neighbor (KNN), to predict the EVs' energy consumption. Tackling the challenge of predicting EVs' energy consumption, the data were collected from Aichi Prefecture, Japan, combining the digital elevation map with long-term GPS tracking data. EVs energy consumption in terms of energy efficiency (kWh/km) was estimated using several important variables such as average trip speed (km/h), trip distance, nighttime lighting, air conditioner (A/C), heater usage ratio, and road gradient. Several statistical evaluation metrics were used to evaluate the performance of the proposed methods. The prediction results show that ESG is more robust in predicting EVs' energy consumption and outperformed other models by yielding more acceptable values for proposed evaluation metrics. The results further demonstrate that the accuracy of predictive models for EVs energy consumption can be reasonably accomplished by adopting stacking techniques. The finding of this study could provide essential guidance to decision-makers and practitioners for planned development and optimal placing of EV charging infrastructures in urban areas.

Automated summary

This paper introduces an ensemble stacked generalization (ESG) approach for better prediction of electric vehicles (EVs) energy consumption. By combining multiple base regression models into one meta-regressor, ESG enhances model prediction and reduces model variance. The results demonstrate that ESG is more robust in predicting EVs' energy consumption and outperforms other models, providing essential guidance for decision-makers and practitioners in urban areas.