DATA-DRIVEN AND LEARNING-BASED MULTI-OBJECTIVE EV ROUTING: A SURVEY OF PREDICTIVE MODELS AND OPTIMIZATION FRAMEWORKS

Abstract

The growing adoption of electric vehicles (EVs) has resulted in the surge in the use of intelligent routing systems to optimally balance the time of travel, energy consumption, and the need for charging. EV routing is complex by nature, because of limited battery capacity, non-linear energy consumption, dynamic traffic situation and the uncertain availability of charging stations. Traditional deterministic approach has difficulty in accommodating these multiple constraint sides, which is the reason for the appearance of data driven methodologies. Leveraging open datasets - including Open Street Map (OSM), Open Mobility, NREL and TomTom - researchers are able to capture spatial, temporal, and energy-related information that is very detailed to help inform predictive models. Machine learning (ML), including graph neural network (GNN) for representing the road net, LSTMs / GNNs for time-series forecasting of traffic, and regression or K-Neighbours (KN) for estimating energy consumption help to predict some important routing parameters accurately. Complementing these predictions, multi-objective optimization frameworks ranging from evolutionary algorithms, to swarm intelligence, to hybrid metaheuristics, allow to analyse systematically the trade-off between travel time, energy cost, state-of-charge and charging delays. This review compiles existing research into a well-established taxonomy, compares datasets, ML models and optimization strategies, identifies existing research gaps in issues of scalability, handling uncertainty and generalization, as well as propose future research directions.