Applying deep reinforcement learning to solve the low-carbon economic dispatch problem in smart microgrids

Abstract

Abstract Smart microgrid (SMG) technology can effectively integrate the advantages of new and renewable energy power generation - distributed generation, and, at the same time, provide a new technical approach for the large-scale application of new and renewable energy grid-connected power generation. Smart microgrids can improve the comprehensive utilization efficiency of energy and serve as an effective complementary power grid of the primary power grid to enhance the reliability of power supply and power quality, which is the latest frontier topic in power engineering research. This paper, aiming at the characteristics of distributed generation such as solar power generation, wind power generation, fuel cell power generation, and micro gas turbine in the smart microgrid system, considering the fuel, efficiency, operation, and maintenance costs of different types and capacities, the different greenhouse gas emissions, and the particularity of solar power generation and wind power generation, a mathematical model of low-carbon economic dispatch of smart microgrid that comprehensively considers the cost of power generation and emission is proposed. This paper presents a scheduling decision model based on an improved dual-delay depth deterministic strategy gradient algorithm. The improved double-delay deep deterministic strategy gradient algorithm is used to construct and train the neural network to avoid overestimation and improve the stability of the network output. At the same time, to improve the efficiency of network training, the historical experience data in the training process were stored, and the experience was replayed and sampled. Examples verify the superiority and practicability of the method proposed in this paper.