Deep Learning based VPP Active Power Dispatching Equivalent Modelling for Global Dispatching Optimization

The active power dispatching equivalent model of the virtual power plant (VPP) and the global dispatching optimization issue of a regional power system integrated with the VPP equivalent model are studied in this paper. Considering the feasibility of a dispatched active power output curve for VPP, the active power dispatching equivalent model of the VPP is built as two equivalent models. One is the feasibility equivalent model trained by fine trees based on the data sets of the active power output curve of the VPP and its feasibility flag, and the other one is the cost equivalent model trained as a deep neural network based on the data sets of the active power output curve and its daily generation cost. Global dispatching issue is formulated as a multi-objective optimization model and solved by NSGAII algorithm. The active power dispatching equivalent model and the multi-objective optimization model is verified by case study, and results show that the proposed VPP equivalent model makes it possible for the power dispatching center to schedule VPP’s optimal power generation plan so as to maximize the generation revenue of the VPP and minimize the total generation cost of the system.