Privacy-aware EV charging-station assignment: QoS–fairness optimization with traffic-analysis resistance

The assignment of charging stations (CS) within urban electric vehicle (EV) networks necessitated a careful equilibrium among user quality of service (QoS), equitable station load distribution, and the mitigation of traffic-analysis inference from publicly observable signals. We examined the online assignment challenge involving identical stations and established an objective aimed at maximizing the minimum station load as a pragmatic surrogate for alleviating persistent hotspots while ensuring low-latency performance. Seven event-driven heuristics (Randomized-Priority Vehicle Charging Algorithm with Coarse-Grained Probability ( RCP ), Randomized-Priority Vehicle Charging Algorithm with Fine-Grained Probability ( RFP ), Mixed-Random Vehicle Charging Algorithm with Dual Sorting ( MDS ), Randomized Reordered Assignment Algorithm with Dual Sorting ( RRD ), enhanced randomized-reordered assignment algorithm with dual sorting ( ERD ), enhanced randomized-reordered assignment with adaptive-threshold ( ERA ), reversed randomized priority vehicle charging algorithm with fine grained probability ( RRF )) were innovatively devised to assess the impact of randomized prioritization, dual sorting mechanisms, candidate reordering, enhanced scoring systems, and adaptive thresholding, all under realistic feasibility constraints (including connector compatibility, operational hours, and capacity limits). Privacy considerations were formalized within a passive observer framework and quantified through metrics such as station-selection entropy $H(\pi)$, station-time cohort k -anonymity, and the performance of potential attackers (including top-1 station prediction accuracy and calibration) based solely on publicly available covariates. Experiments combined a Riyadh metropolitan case study with a 1,800-instance benchmark suite. Statistical analysis over 1,800 paired instances (bootstrap 95% CIs; Wilcoxon with Holm–Bonferroni) showed that ERD and ERA improved Ts min over RCP by a median of +20 (95% CI [20–24] and [16–20], respectively; $p\ll{10}^{-{100}}$), at the cost of modest runtime increases (+3 and +5 ms per update). Upper-bound validation indicated that 27.17% of instances achieved A = UB , with median gap 0.006 and 90th-percentile gap 0.039. Dynamic evaluations with surges, outages, and telemetry delays reported tail QoS (p95 waiting/travel-to-charge) and max/min load ratios with confidence bands. The scope of privacy claims was limited to assignment-layer traffic-analysis resistance; protocol-level cryptographic security and active adversaries were treated as out of scope.