V2G-MAAC: A Semi-Synthetic Benchmark Dataset for Multi-Authority Access Control in Vehicle-to-Grid Charging Networks
Description
V2G-MAAC is a semi-synthetic benchmark dataset for multi-authority access control in vehicle-to-grid (V2G) charging networks. It is built on real EV charging-session structure from ACN-Data (Lee, Li & Low, 2019) and extends it with three independently injected, ground-truth threat axes spanning the three authorities present in a V2G session — the Charge Point Operator (CPO), the e-Mobility Service Provider (eMSP), and the Electric Vehicle (EV) itself — plus a fourth, parallel identity-delegation axis (acting identity vs. home identity for each authority). The release contains seven variants sharing one generator configuration: a reference configuration (main), three independent reseeds (seed_101, seed_103, seed_104) for distinguishing genuine reproducibility from variant-induced change, and three stress conditions (adversarial: threat-layer parameters shift over time; noisy_benign: elevated behavioral noise among honest actors; sparse_feedback: reduced availability of running, history-dependent features). Each variant contains 125,110 charging sessions (108,060 train / 17,050 test), with a day-blocked temporal split reconstructable from the included split column. Each session carries billing/latency/energy-rate deviation features, charge-session timing, an actor’s own flagged-session history, EV session-reuse share, and certificate-validity/roaming/peak-hour context flags, alongside per-axis binary ground truth (gt_cpo_malicious, gt_emsp_malicious, gt_ev_sybil — the first two onset-gated) and pooled convenience labels (risk_factor_count, composite_risk_score). A fourth, parallel identity-delegation axis (gt_cpo_delegate_malicious, gt_emsp_delegate_malicious, gt_ev_delegate_abuse) is reported as an exploratory finding, not a fully evaluated baseline-ladder rung. The release includes a data dictionary, a splits specification, a “Datasheets for Datasets” questionnaire, machine-readable citation metadata, and a file-integrity manifest — see README.md for the complete file list and usage notes. The generator source code is not included in this release; full generation methodology, threat model, and all reported results are described in the companion paper. Baseline implementations (RBAC/ABAC and six learned baselines) and the evaluation harness that reproduces every reported result from these files are released separately on https://github.com/mridul99/V2G-MAAC
Files
Institutions
- National Institute of Technology PatnaBihar, Patna
Categories
Funders
- IHUB NTIHAC Foundation, IIT Kanpur