I. Introduction
With the vigorous development of rooftop photovoltaic and the construction of new power system, distributed energy has ushered in explosive growth. However, due to the characteristics of intermittence, randomness, and volatility of distributed energy, its high proportion of grid connection will conduct a great impact on power flow distribution, power quality, and network loss of the power system [1]. Therefore, it is necessary to dynamically dispatch and control distributed energy according to load to improve the stability of new power system and the absorption of distributed energy [2]. However, the traditional dispatching and control methods based on optimization theory face the problem of dimensionality disaster, and are difficult to obtain the optimal decision-making model by quantifying the mapping relationship between large-dimensional state information and dispatching and control decision [3].