Game theoretic and multi agent reinforcement learning based adapti 2027 Appl

With the development of renewable energy and the increase of carbon emission limits, the operation optimization of power systems under the consideration of the coupling of the electric energy market, carbon emission market, and green certificate (GC) market has received wide attention. However, the high nonlinearity and dynamic characteristics of the power system after introducing renewable energy sources make it difficult to obtain the optimal operation decision under the coupled electricity-carbon-green certificate market. To solve this, we formulate a hierarchical game between the VPP operator and multiple aggregators. Specifically, a multi-agent twin delayed deep deterministic policy gradient (MATD3)-based MADRL algorithm employs a centralized training with decentralized execution (CTDE) framework to optimize the operator's multi-aggregator dispatch strategy. Meanwhile, multi-type resource aggregator models are developed to minimize operational costs, with a generalized regression neural network (GRNN) introduced to approximate the nonlinear mapping between the V.