Probabilistic prediction-based multi-objective optimization approach for multi-energy virtual power plant

Abstract

Virtual power plants (VPPs) are encountering multiple challenges due to market uncertainties and power network instability. In this paper, a novel probabilistic prediction-based multi-objective optimization framework for VPP is proposed to maximize operating profit while minimizing pollutant emissions and voltage deviations in the distribution network, which considers the uncertainties of wind power and electricity price. In this framework, the VPP that participates in the energy and ancillary service markets firstly aggregates the wind farms, the electric vehicle charging stations (EVCS), and the combined cooling, heating, and power subsystems to improve the utilization efficiency and operational flexibility of multiple energy sources. Then, a new Pareto optimizer, called multi-objective hybrid sand cat swarm optimization and strength firefly algorithm, is proposed to tackle the multi-objective optimization model of VPP. The proposed hybrid algorithm utilizes the advantages of sand cat swarm optimization and strength firefly algorithm mechanisms to facilitate local exploitation and global exploration. Finally, a new deep reinforcement learning probabilistic prediction approach based on quantile regression deep deterministic policy gradient is modeled to evaluate the uncertainties. The proposed models and methods have been thoroughly discussed on a modified distributed network. It is calculated that compared with the VPP without EVCS, the operating profit of the proposed VPP increases by 18.69%, and the emissions and voltage deviation of the proposed VPP are reduced by 3.42% and 10.44%, respectively. Experimental results also prove that the performance of the proposed Pareto optimizer and probabilistic prediction approach is superior to other benchmark techniques.