Battery aging-aware vehicle-to-grid control in home charging systems via deep reinforcement learning

The increasing integration of renewable energy sources and the electrification of the transportation sector are fundamentally reshaping modern energy systems. In this context, electric vehicles have emerged not only as low-emission transport modes but also as distributed, mobile energy storage units capable of supporting grid stability through Vehicle-to-Grid operation. Among various deployment environments, residential home energy systems provide an especially promising setting for Vehicle-to-Grid implementation due to long vehicle connection durations, rising adoption of rooftop photovoltaic systems, and growing availability of smart home infrastructure. This dissertation presents a comprehensive framework for modeling, controlling, and evaluating Vehicle-to-Grid operation in residential energy systems, with a particular focus on battery degradation, economic performance, and user-centric control. A modular Vehicle-to-Grid simulation model is developed, incorporating real-world driving behavior, household electricity demand, photovoltaic generation, charging infrastructure characteristics, and electricity market signals. A semi-empirical battery aging model is experimentally validated and integrated to quantify capacity degradation under different charging strategies, capturing both calendar and cycle aging effects.A key contribution of this work is the development of a control framework based on deep reinforcement learning to enable aging-aware, real-time Vehicle-to-Grid operation. The charging problem is formulated as a Markov Decision Process and solved using state-of-the-art algorithms tailored for energy applications. The proposed controller jointly optimizes electricity cost, battery longevity, and user mobility satisfaction under realistic uncertainties such as dynamic pricing and variable availability. Benchmark comparisons with rule-based and optimization-based Model Predictive Control strategies confirm the superior performance, robustness, and flexibility of the deep reinforcement learning approach, establishing it as a scalable foundation for intelligent Vehicle-to-Grid deployment. Building upon this control foundation, the techno-economic potential of Vehicle-to-Grid operation is systematically assessed across a broad set of residential scenarios. Two Vehicle-to-Grid paradigms are defined and compared: a Session-Based approach constrained by user-defined departure time and state-of-charge targets, and a Full-Automatic approach that leverages the ability of reinforcement learning to learn adaptive charging behavior from mobility profiles. These strategies are evaluated under varying conditions of user mobility, battery capacity, and photovoltaic system size, and benchmarked against conventional home storage systems. The findings highlight the strategic value of Vehicle-to-Grid as a complement or alternative to stationary storage, especially when supported by adaptive, battery-conscious control. They also underline the importance of integrating battery degradation into both system modeling and decision-making. Beyond methodological contributions, the results point to future opportunities for expanding Vehicle-to-Grid beyond single-EV households, enabling online learning for real-world adaptation, and establishing market and regulatory structures that support its widespread adoption. This work contributes to the advancement of Vehicle-to-Grid technology by providing a scientifically grounded, technically validated, and practically scalable control and evaluation framework tailored to residential energy systems. It supports the broader transition toward decentralized, flexible, and user-centric energy systems where electric vehicles play an active and intelligent role.