Residential Demand Response Considered Strategic Bidding for Load Aggregators with Soft Actor-Critic Algorithm

Abstract

With more flexible loads and prosumers actively integrated into modern power system, competitive electricity market serves a more important role on demand side management. To facilitate residential customers flexibly participate in Demand Response (DR) programs, load aggregators (LA), as a representative of regional users, gather regional DR resources to transact with grid operators. On the basis of LAs’ perspective, this paper systemically designed an uncertainty considered strategic bidding model of residential DR, aiming at improve DR reliability, meanwhile maximal the financial benefits of LAs in day-ahead market, which able to provide a guidance on reasonably estimating the DR amount of declaration, offering a wise incentive price to encourage users to accomplish the declaration, meanwhile avoiding the DR rebound peak. Moreover, to quantify the DR uncertainties, an electricity usage habits and reward sensitivities considered probabilistic model is also proposed. In addition, to optimize the continuous-action-space problem, soft actor-critic (SAC) featured deep reinforcement learning (DRL) algorithm is implemented to the decision-making process. With comparison to the traditional method and another DRL approach, SAC is proved its effectiveness in solving DR optimization problems.