In the active distribution network (ADN), the photovoltaic (PV) output has fluctuations and the load power has real degeneration. When multiple faults occur in the distribution network, the repair time is longer, and PV may have insufficient output during certain periods of time. The power supply is restored to as many important loads as possible in the non-faulty power loss zone. This paper combines the short-time scale optimization scheduling method to establish the dynamic repair model for the ADN sub-area, sub-scenario, and time-segment based on a multi-agent system (MAS). In the model, the maximum load value of the network and the minimum number of switching times are the upper targets; the maximum recovery power of the different recovery periods and different switching times in the non-faulty power loss zone is the premise, and the maximum total power supply is the lower target; and PV and battery storage and distributed generation participate in joint optimization. An island path optimization method is proposed, which can quickly find the optimal recovery path in each period of non-faulty power loss zone. The multi-objective ant colony algorithm is designed to obtain the optimal repair scheme for the whole network. The IEEE 69-point distribution system is taken as an example to prove the feasibility and effectiveness of the proposed strategy.