For multiple virtual power plants (MVPP) in regional distribution network, the coordinated optimization based on the mutual power support between virtual power plants (VPPs) will impove the the flexibility and economical efficiency of scheduling. However, most of the research on MVPP coordinated optimal scheduling adopts the idea of centralized modeling and unified solution. With the increase of the number of aggregated resources, this method becomes more difficult to solve, and is not conducive to the privacy protection of each VPP. To solve these issues, this paper proposes a distributed coordinated optimal scheduling method or MVPP under decentralized control structure based on the idea of "information separation and decision coordination". Firstly, a distributed coordination and optimal control mechanism for MVPP is constructed. Secondly, the multi-time coordinated optimal scheduling model of MVPP is constructed, and the transaction price function between VPPs is constructed based on the relationship between supply and demand of power. Then, the Lagrange dual relaxation theory is used to relax the optimization model. Finally, the day-ahead multi-time coordinated optimal scheduling problem is transformed into a real-time optimal scheduling problem based on decentralized partially-observable Markov decision process (DEC-POMDP), and the optimization problem is solved based on the improved quantum genetic algorithm (QGA). Simulation results on the MVPP system composed of three VPPs show that, when MVPP adopts coordinated optimal scheduling, the total amount of electricity sold to the grid is 8 517.60 kW·h, the total amount of electricity purchased from the grid is 125.41 kW·h, and the total cost is 998.68 ¥. When each VPP adopts optimal scheduling independently, the total amount of electricity sold to the grid is 9 921.44 kW·h, the total amount of electricity purchased from the grid is 823.26 kW·h, and the total cost is 1 392.75 ¥. This is because when MVPP adopts coordinated optimal scheduling, the transaction price between VPPs is better than electricity market. Under the incentive of the electricity price, transactions between VPPs are prioritized, and the transaction volume with the grid is reduced accordingly, and the total cost is also reduced. Then, the MVPP coordinated optimal scheduling model is solved by the centralized and distributed optimization methods respectively, the calculation results of these two methods are basically consistent, and the calculation time is 42.67 s and 12.84 s respectively. Based on the proposed distributed optimization method, each VPP only needs to interact with Lagrangian multipliers, and each sub-problem can be calculated in parallel to improve the solution efficiency, which takes into account the information privacy and computing efficiency. Finally, comparing the day-ahead optimal scheduling and the real-time optimal scheduling based on DEC-POMDP, when the day-ahead optimal scheduling plan is executed, the power shortage or surplus caused by prediction error are made up or absorbed by the power grid. While the real-time optimal scheduling based on DEC-POMDP performs optimal calculation based on the measured values of wind power and load, VPP can cope with the fluctuation of wind power and load through internal resources coordination or mutual power support between VPPs, which will reduce the electricity transactions with the grid. The following conclusions can be drawn from the simulation analysis: (1) MVPP reduces the electricity transactions with electricity market based on coordinated optimal scheduling through the mutual power support among internal VPPs, and constructs the inter-VPP transaction price function to improve the enthusiasm of VPP to participate in direct trading, which has higher economical efficiency compared with the independent optimal scheduling of each VPP. (2) The distributed coordinated optimal scheduling model of MVPP based on the principle of Lagrange dual relaxation, can realize the global optimization only by exchanging a small amount of information between VPPs, and guarantee the information privacy of VPP better than the centralized optimization. (3) The MVPP multi-time coordinated optimal scheduling problem is transformed into a real-time optimal scheduling problem based on DEC-POMDP, which can effectively deal with the scheduling deviation caused by the prediction error and ensure the reasonable allocation of the resource output plan within one day. © 2023 Chinese Machine Press. All rights reserved.
