Loading...

Residential Energy Hub Management using Reinforcement Learning Methods

Ghader Tootoonchi, Alireza | 2023

50 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55848 (46)
  4. University: Sharif University of Technology
  5. Department: Energy Engineering
  6. Advisor(s): Moeini Aghtaie, Moein
  7. Abstract:
  8. In recent years, various studies have been conducted focusing on the optimization and scheduling of energy hubs (EH). EH optimization is important because it tries to create synergy in the energy supply system by considering the relationships between energy carriers. For this purpose, various algorithms such as analytical methods (linear programming) and evolutionary and heuristic algorithms have been used. Also, recently, various studies have addressed the applications of reinforcement learning for energy hub management. In this study, a hybrid model for energy hub management has been developed by combining classical linear models (Known as white-box models) and reinforcement learning (a black-box model). Algorithms used for reinforcement learning include SARSA, Expected SARSA (ESARSA), Q-learning (Q) and Double Q-learning (DQ). To check the performance of the hybrid model, its results have been validated with a MIP model. Unlike the MIP model, the hybrid model does not have complete information about the environment and only has information about the current time step at each time step. Nevertheless, the results indicate that the hybrid model’s accuracy is 98% in the considered deterministic environment, and when the environment uncertainty is taken into account, the accuracy is 91%. Among the evaluated algorithms, Q and ESARSA have a more appropriate performance. Because while the cost obtained with MIP was 1.17 USD per day, SARSA, ESARSA, Q and DQ obtained costs of 1.21, 1.20, 1.18 and 1.30, respectively
  9. Keywords:
  10. Reinforcement Learning ; Machine Learning ; Optimization Under Uncertainty ; Energy Hub ; Energy Storage ; Energy Management ; Optimal Scheduling ; Hybrid Optmization Framework

 Digital Object List

 Bookmark

No TOC