Multi-scenario operation optimization of electric-thermal coupling renewable energy system based on deep reinforcement learning

Abstract

Within the electric-heat coupling renewable energy system, the complex coupling relationship between electric energy and heat energy, integrated with large-scale of uncertainties in renewable energy and load demand, brings about significant challenges to the real-time economic dispatch of the power system. In view of this, this paper establishes an electric-heat coupling renewable energy optimization model including pumped storage power station and combined heat and power unit which is equipped with heat storage device, and a modified proximal policy optimization (MPPO) algorithms was proposed. First, a vast of electric-heat coupling renewable energy system optimal scheduling problems are expressed as Markov decision processes under the framework of deep reinforcement learning, and then the reward function mechanism is designed to guide the algorithm to generate the best dispatching plan. Next, the collection and sampling method of training samples, as well as the updating mechanism of network parameters and strategies are established. Finally, a case study is carried out with the typical power system, and the simulation results of the electric-heat coupling renewable energy system indicate that the proposed algorithm can effectively reduce the training cost, and the algorithm can handle with varieties of operating situations real-time scheduling of the electric-heat coupling renewable energy system.