Under the background of the current energy structure transformation, the proportion of distributed photovoltaic power generation in the power system is increasing, but its intermittency and uncertainty bring serious challenges to the stable operation of the power grid. In this paper, for the traditional virtual power plant in distributed PV monitoring and regulation of the lack of real-time and data processing delay problems, constructed a virtual power plant based on edge computing technology distributed PV real-time monitoring and regulation system. The study adopts centralized and decentralized control mode, taking edge computing nodes as the low-level control and the virtual power plant energy management center as the high-level control. Five main influencing factors, namely, solar irradiance, ambient temperature, air humidity, wind speed and barometric pressure, are screened out by the gray correlation method, and an improved LSTM-TCN prediction model is constructed for the ultra-short-term output prediction of distributed photovoltaic. Based on the experimental data validation at five sites in Australia, the LSTMTCN model has an MAE of 0.0388 and an RMSE of 0.0759 in typical summer scenarios, which improves the accuracy by 2.59% and 6.21%, respectively, compared with the traditional LSTM model. In the IEEE 33-node distribution network example, the total system load is 12MW and the total installed capacity of PV is 9.0MW, the proposed method realizes the ideal characteristics of virtual power plant with strong internal coupling and weak external coupling. The results show that the virtual power plant architecture based on edge computing can significantly improve the prediction accuracy of distributed PV and the system regulation effect, which provides an effective technical solution for building a smart grid.
