Research on clustering and co-regulation of virtual power plant resources based on machine learning

In order to realize the strategic goal of environmental protection and low carbon, designing a set of resource clustering and regulation strategies that take into account energy saving and operating costs has become a research challenge for virtual power plants. In this study, the ICEEMDAN-CNN-SSDAE hybrid model is used to realize high-precision prediction of electricity price and load data in virtual power plants. The objective function and constraints of resource clustering and cooperative regulation of virtual power plants are established under the condition of demand response, and solved by Markov process. Finally, the virtual power plant resource clustering and co-regulation model is constructed on the basis of the deep reinforcement learning model framework by combining the prediction model and objective function. The results show that the ICEEMDAN-CNN-SSDAE model proposed in this paper can guarantee high prediction speed (0.062s and 0.059s) while having high prediction accuracy. It is also found that the average capacity of the output power of each component in the virtual power plant system after the model clusters and optimizes the regulation of a virtual power plant resources increases by 0.535-0.686 MW/h compared with the pre-optimization period, and the economic efficiency and energy utilization are also improved to different degrees. The research in this paper verifies the rationality and effectiveness of the proposed model, and provides certain theoretical basis and guidance method for virtual power plant resource clustering and cooperative scheduling.