Date of Award

2026-08-01

Degree Name

Doctor of Philosophy

Department

Data Science

Advisor(s)

Abhijit Mandal

Second Advisor

Tzu-Liang (Bill) Tseng

Abstract

Electric vehicle (EV) charging stations are becoming increasingly complex energy systems in which charging decisions must coordinate photovoltaic (PV) generation, battery energy storage, and grid interaction under significant uncertainty. Future EV arrivals, charging demand, parking duration, PV generation, and electricity prices are inherently stochastic, making real-time energy management a challenging sequential decision-making problem. Unlike most existing approaches, the proposed framework formulates the problem as a finite-horizon discrete-time stochastic optimal control problem that explicitly models uncertainty through data-driven stochastic processes calibrated from historical observations. Since the resulting problem cannot be solved practically using classical stochastic dynamic programming, the control policy is parameterized by a neural network and optimized directly through stochastic simulation. The learned policy is evaluated under representative seasonal operating conditions and compared with alternative control strategies. Experimental results show that the proposed policy achieves the lowest expected operating cost while generalizing across representative seasonal conditions without retraining. The ablation study quantifies the complementary contributions of the photovoltaic system, battery energy storage system (BESS), and grid connection, while the sensitivity analysis demonstrates that the learned operating strategy remains robust under different EV pricing policies. These results show that the proposed framework provides an effective and interpretable methodology for real-time energy management of EV charging stations operating under uncertainty.

Language

en

Provenance

Received from ProQuest

File Size

167 p.

File Format

application/pdf

Rights Holder

Denisse Urenda Castañeda

Download

Share

COinS