Residential Demand Response using Q-Learning with Pursuit Algorithm for Action Selection

Demand response (DR) has been extensively explored in various works utilizing reinforcement learning (RL) techniques. However, the prevailing objective functions in these studies primarily focus on minimizing energy costs for consumers or reducing maximum utility demand, potentially overlooking consumer discomfort. It is crucial to acknowledge that different load scenarios can lead to varying levels of discomfort for consumers. This paper proposes a novel RL framework that incorporates the notion of discomfort into the decision-making process. To achieve this, we employ the pursuit algorithm, a highly efficient method for action selection during the learning phase. Furthermore, we conduct a comprehensive analysis to assess the impact of the convergence rate on the algorithm's performance.