An AI-Augmented Dataset of Multi-Prototype Electric Vehicle Charging Load Profiles in China.

Liu, R; Li, Y; Guo, N; Li, D; Qu, H; Zhou, Z; Guo, Y; Yan, Z; Liu, J · 2026 · PubMed Central

DOI: 10.1038/s41597-026-07273-5

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper introduces MP-EVData, a comprehensive dataset designed to address the challenges of integrating large-scale electric vehicles (EVs) into power grids by providing high-fidelity, diversified charging load profiles. The authors identify a critical gap in existing research: while numerous EV datasets exist, they often focus on private residential charging or lack parallel, comparable data from multiple station types within the same geographical and temporal context. This limitation hinders controlled comparative analysis of load characteristics across different vehicle prototypes. To resolve this, the study presents a dataset capturing station-level charging loads from a major Chinese metropolis (Shanghai) throughout 2024, covering five distinct prototypes: taxi demonstration stations, bus depots, residential charging stations, battery swapping stations, and heavy-duty truck stations. The dataset comprises data from 10 representative stations, selected to reflect typical operational scales and user bases. Raw session-level records, including energy consumption and timestamps, were collected via a public monitoring platform and rigorously anonymized to protect privacy. These discrete records were preprocessed to remove anomalies and aggregated into continuous station-level load profiles at 15-minute and hourly resolutions. To support data-intensive deep learning applications, the authors augmented the real-world data with a synthetic dataset generated using four generative models: Variational Autoencoders (VAE), Generative Adversarial Networks (GAN), Denoising Diffusion Probabilistic Models (DDPM), and Gaussian Mixture Models (GMM). Synthetic generation focused on the four conventional charging prototypes, excluding battery swapping due to its stochastic, discrete nature. Technical validation reveals significant heterogeneity in load patterns across prototypes. Bus depots exhibit massive, unimodal overnight peaks, while taxi stations show bimodal patterns corresponding to midday breaks and late-night off-peak charging. Residential stations display evening ramp-ups, and heavy-duty truck stations show irregular, high-power events tied to logistics schedules. Statistical analysis highlights vast differences in scale and volatility; for instance, bus depot mean loads are over 150 times greater than residential stations. Weekly analysis further distinguishes commercial fleets, which operate continuously regardless of the day of the week, from residential users, whose charging behavior shifts significantly between workdays and holidays. The dataset also demonstrates clear price-responsive behavior under time-of-use pricing schemes. MP-EVData serves as a crucial benchmark for advancing research in load forecasting, smart charging algorithms, and urban infrastructure planning. By eliminating external variables such as geography and policy, it enables direct, controlled comparisons of charging behaviors. The inclusion of AI-generated synthetic data provides researchers with scalable resources for training complex models. The dataset and associated code are publicly available, facilitating reproducibility and supporting the development of grid-interactive systems that manage EV charging as a flexible load to enhance grid stability and renewable energy integration.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• micromobility e scooter