Electric Vehicle Range Prediction Models: A Review (www.mdpi.com)

This paper presents a systematic review of electric vehicle (EV) driving-range prediction methods, cataloguing machine learning, mathematical, simulation and hybrid approaches and evaluating study quality, data sources, feature sets and performance metrics. Using a structured search and multi-point quality assessment, the authors extract model types (classical ML, deep learning such as LSTM/autoencoders, ensembles, PSO‑LSSVM, and physics-based/drivetrain simulations), data origins (experimental, simulated, public datasets), common features (battery SOC/age, temperature, driving cycles, vehicle dynamics, route and traffic), and evaluation measures (MAE, RMSE and accuracy/error ranges). They find a clear trend toward hybrid physics+ML models, deep-learning on large routing/telemetry datasets, and growing use of explainability tools (e.g., SHAP) and ensembles. For the AI/ML community the review highlights practical gaps and research directions: many studies report promising MAE/RMSE but suffer from limited or non-public datasets, inconsistent validation strategies, and poor transfer to real-world conditions (thermal effects, battery ageing, variable traffic). Key technical implications are that combining domain knowledge (battery/vehicle physics) with data-driven models improves robustness, while standardized benchmarks, uncertainty quantification, and stronger external validation are needed to move range predictors from lab prototypes to reliable in‑vehicle systems.