Electric Vehicle Charging Load Modeling: A Survey, Trends, Challenges and Opportunities

TL;DR

The paper addresses the challenge of accurately modeling EV charging load under uncertainty by synthesizing information fusion across data, models, and decision strategies. It categorizes state-of-the-art methods into statistical, simulated, and data-driven families, and surveys advances including ML, graph-based approaches, transformers, PINNs, generative models, and LLMs. It analyzes three bottom-up information fusion levels and discusses critical challenges such as transferability, data quality, privacy, and the need for gray-box approaches that combine physics with data-driven insights. The work provides a roadmap of future research directions and practical guidance to enhance forecasting, planning, and grid integration of EV charging systems.

Abstract

The evolution of electric vehicles (EVs) is reshaping the automotive industry, advocating for more sustainable transportation practices. Accurately predicting EV charging behavior is essential for effective infrastructure planning and optimization. However, the charging load of EVs is significantly influenced by uncertainties and randomness, posing challenges for accurate estimation. Furthermore, existing literature reviews lack a systematic analysis of modeling approaches focused on information fusion. This paper comprehensively reviews EV charging load models from the past five years. We categorize state-of-the-art modeling methods into statistical, simulated, and data-driven approaches, examining the advantages and drawbacks of each. Additionally, we analyze the three bottom-up level operations of information fusion in existing models. We conclude by discussing the challenges and opportunities in the field, offering guidance for future research endeavors to advance our understanding and explore practical research directions.

Figures (7)

• Figure 1: EVs charging station market size from 2023 to 2032$^1$
• Figure 2: The model tree about adopted EV charging load modeling methods. TODO
• Figure 3: Logic Flow of Monte Carlo Simulation for Micro-Grid EV Charging Load. This flowchart illustrates the logical dependencies within the simulation process, with each arrow representing the flow of information or dependency between components. This flow is summarized based on literature helmus2020dataroadnet_powergridroadnet_whetherhan2020orderedfuzzy4spatio_temporal_roadnetspatio_temporal_roadnet1spatio_temporal_trafficmoghanlou2023probabilisticbib1bib3long-shenzhenzhang2021evzhang2020dailyroadnet_roadcondliu2020electricni2020methodologytayyab2021infrastructurexing2022multibian2022multitian2022electricgou2021chargingbib0williams2024drivingzhuang2022loadzhang2023batterychen2023spatioliu2024electric.
• Figure 4: Sample architecture of graph-based spatio-temporal EV charging modeling. This architecture featuremining_GCN_GRU, typically employs a sequential combination of spatial and temporal learning layers to extract spatio-temporal features from graph-structured input data.
• Figure 5: A general architecture for GAN in profile imputation. Extracted from Shen et al.'s work shen2022short. The solid lines with the arrows represent the data flow directions, while the dashed lines with the arrows represent the backpropagation process.