Abstract
<jats:p>Accurate prediction of the remaining driving range of an electric vehicle (EV) is one of the key challenges in the development of modern energy management systems and intelligent transportation systems. Reliable range estimation directly affects safety, operational convenience, and user confidence in electric vehicles. Traditional physics-based prediction models rely on energy balance equations and enable interpretation of the influence of major energy consumption factors; however, their accuracy under real-world driving conditions is limited. This limitation arises from insufficient consideration of complex nonlinear effects, including battery degradation, individual driving style, traffic variability, road topology, and weather conditions. On the other hand, machine learning methods, particularly recurrent neural networks, demonstrate high accuracy in time-series prediction tasks but often lack physical interpretability and may suffer from reduced robustness under limited or nonstationary data conditions. This paper proposes a hybrid approach for predicting the remaining driving range of an electric vehicle that combines the advantages of physical modeling and machine learning. A physics-based model is employed to generate a baseline energy consumption forecast, while a Long Short-Term Memory (LSTM) neural network is trained to compensate for the residual errors of the physical model using multidimensional telemetry data. Experimental validation of the proposed approach was conducted using open datasets, including the Vehicle Energy Dataset, EVBattery, and EVIoT–PredictiveMaint, which cover various driving modes, temperature conditions, and battery states. The obtained results demonstrate that the hybrid model provides a significant improvement in prediction accuracy, reducing the mean absolute error by 20–25% compared to conventional physics-based models and standalone machine learning approaches. The proposed method combines high accuracy, interpretability, and practical applicability, making it promising for implementation in onboard electric vehicle systems.</jats:p>