Abstract

As electric-vehicle (EV) adoption accelerates, driven by policy commitments and financial incentives, the public charging network must scale to meet rising demand while preserving a satisfactory quality of experience (QoE). Infrastructure growth is uneven across regions, producing local capacity shortfalls that erode user confidence and deter public charging use. Addressing these challenges requires systematic visibility into session dynamics, robust demand forecasts, and principled deployment rules that reduce uncertainty in per-site waiting-time estimation. This thesis develops analytical, theoretical, and data-driven methods to: (i) characterize per-site session dynamics and waiting-time statistics; (ii) define reliable, key QoE metrics (waiting time, blocking probability, utilization, queue length); (iii) deliver a client-server decision-support platform for site-level visualization and diagnostics; (iv) propose demand-management incentives and queueing models to improve QoE; and (v) quantify and forecast the QoE impact of new deployments. Using large, real-world datasets, the study shows that charging times are frequently better modeled by Erlang-$k$ distributions and that per-site request processes can be accurately approximated by single-server queueing systems under common scheduling policies. These empirical findings establish the probabilistic foundation required for reliable waiting-time estimation and capacity planning. The derived QoE metrics drive a tailored machine-learning forecasting pipeline trained on empirical data, and extensive simulation validates the forecasts and supports evidence-based expansion decisions. To limit overload and reduce congestion, the thesis introduces a Data-driven, Incentive-based Charging Truncation (DICT) policy that encourages drivers to stop charging near 80\% state of charge. A closed-form fit for the resulting service-time distribution is derived and analyzed within an M/G/C/K queueing framework. DICT is benchmarked against resizing and proximity-based expansion strategies to identify conditions where incentive policies outperform or complement physical expansion. Finally, a counterfactual machine-learning framework estimates how Level-3 fast-charger deployments affect congestion and QoE at nearby sites. The framework controls for spatial proximity, charger capacity and power ratings, and local amenities; it maps counterfactual demand trajectories into queueing inputs to produce site-level QoE forecasts that inform deployment choices. Together, these contributions integrate empirical discovery, queueing theory, forecasting, decision-support software, and incentive design to enable capacity-aware, QoE-preserving expansion of public EV charging infrastructure.