Deep Learning for Electric Vehicle Charging Load Forecasting: A Survey
TL;DR
This survey presents a comprehensive review of deep learning-based approaches for EV charging load forecasting, categorizing them into point and probabilistic forecasting paradigms, and differentiates single-architecture models, hybrid architectures, and ensemble learning models within each paradigm.
Abstract
The widespread adoption of electric vehicles (EVs) is reshaping modern energy systems, necessitating accurate and efficient forecasting of EV charging demand to ensure grid stability and infrastructure optimization. Traditional statistical and machine learning models offer foundational insights but often struggle with capturing the complex temporal, nonlinear, and spatial patterns inherent in EV charging behavior. Deep learning has emerged as a powerful alternative, demonstrating superior capabilities in modeling long-range dependencies and integrating diverse contextual factors. This survey presents a comprehensive review of deep learning-based approaches for EV charging load forecasting, categorizing them into point and probabilistic forecasting paradigms. Point forecasting methods provide precise single-value predictions suitable for real-time applications, while probabilistic forecasting models quantify uncertainty, enabling risk-aware energy management. The paper further differentiates single-architecture models, hybrid architectures, and ensemble learning models within each paradigm, highlighting different architectures that integrate convolutional, recurrent, and attention-based components. This work also reviews relevant datasets and identifies current challenges, research gaps, and future directions, emphasizing the critical role of deep learning in advancing intelligent, sustainable, and resilient EV charging systems.