I. Introduction
The extreme capacity growth of renewable power installed in Germany in recent years is expected to further increase. Especially in the distribution network, the feed power of renewable power exceeds the power consumption, which leads to the transition from passive distribution network to active distribution network[1] Therefore, new requirements are put forward for distribution network planning [1]–[3]. One of the main objectives of these requirements is to reproduce RES and innovate the uncertainty of grid components [2]. Generally, time series and probability are used to solve the problems [4]–[8]. In view of the fact that there are some worst cases in the current methods of planning progress, these deterministic conditions are pre-determined, which indicates that their selection is based solely on the load and power of the generator set, but not the network state, such as power flow or node voltage. Using these worst cases, the grid scale of high-voltage distribution network (11 kV)[4] is insufficient, and that of large-scale low-voltage network[5] is too large. These examples show that with the high popularization of renewable power, the complexity of distribution network planning increases. Traditional planning methods are no longer applicable.