Energy management for electric vehicle charging stations : optimization and reinforcement learning approaches.

Truong, Van Binh (2023). Energy management for electric vehicle charging stations : optimization and reinforcement learning approaches. Thèse. Québec, Université du Québec, Institut national de la recherche scientifique, Maîtrise en télécommunications, 146 p.

PDF Document sous embargo Télécharger (1MB)

Résumé

Massive adoption of Electric vehicles (EVs) greatly contributes to address the pressing issues of greenhouse gas emissions, depletion of fossil fuels, and noise pollution in urban areas. In addition, proper integration of EVs and Renewable Energy Sources (RES) into the power grid can significantly enhance the efficiency and sustainability of the power infrastructure. By smartly managing the EV charging process and leveraging surplus RES, one can effectively balance energy demand and supply, eliminating the need for costly power infrastructure upgrades. While several EV charging control designs have been proposed recently in the literature, much more research is required to enable large-scale EV integration. The contributions of this thesis relate to the development of efficient EV charging designs as described in the following. Firstly, we propose a novel energy management and EVs charging framework for a smart building microgrid that integrates RES, considering battery degradation and random EVs’ arrivals/departures. We employ a generic non-linear degradation cost model for lithium batteries, considering the Depth of Discharge (DoD) as a determining factor. To effectively handle the uncertainties associated with EV arrivals/departures and inaccurate predictions of electricity load, RES energy, and energy price, we propose a two-layer framework by using the Model Predictive Control (MPC) technique. The upper layer, with an extended optimization horizon, focuses on minimizing total energy usage and battery degradation cost while meeting the required energy levels of EVs at their departures. The lower layer optimizes the charging/discharging operations of EVs and an Energy Storage System (ESS) over a shorter horizon based on the energy planned by the upper layer. In particular, the decision making process taken by the lower layer can greatly mitigate the inefficiency of the solution obtained by the upper layer. As a result, an efficient energy management and EV charging solution can be obtained by the proposed framework. Secondly, we propose a scalable EV charging strategy for a charging station supporting a large number of EVs considering random EVs’ arrivals and departures, battery degradation, and transformer Loss of Life (LoL) by using the reinforcement learning (RL) approach. To tackle the complexity of the traditional RL method, we employ the Factored Action based RL (FARL) technique, which efficiently transforms the action space of the underlying Markov Decision Process (MDP). We then introduce a hybrid learning architecture that combines the Convolutional Neural Network (CNN) with the Proximal Policy Optimization (PPO) algorithm to learn an efficient EV charging pattern. In fact, this hybrid approach enables to extract relevant features from the high-dimensional state space, thereby facilitating to learn an efficient EV charging strategy. For both proposed frameworks, we thoroughly evaluate their performance through extensive numerical studies where we compare them against state-of-the-art designs. These comprehensive evaluations provide compelling evidence of the effectiveness and superiority of our frameworks, highlighting their potential for practical implementation and adoption in real-world scenarios. The findings of our research contribute to the advancement of EV charging technologies and pave the way for more sustainable and efficient transportation and energy systems.

Type de document:	Thèse Thèse
Directeur de mémoire/thèse:	Le, Long
Mots-clés libres:	NA
Centre:	Centre Énergie Matériaux Télécommunications
Date de dépôt:	25 janv. 2024 19:26
Dernière modification:	25 janv. 2024 19:26
URI:	https://espace.inrs.ca/id/eprint/14143

Gestion Actions (Identification requise)

Modifier la notice