A traditional distribution network carries electricity from a central power resource to consumers, and the power dispatch is controlled centrally. Distributed generators (DGs) emerge as an alternative power resource to distribution networks at a smaller and distributed scale, which will bring benefits such as reduced voltage drop and loss. However, because most of high penetration DGs are not utility owned and characterized by high degree of uncertainty such as solar and wind, the distribution network may perform differently from the conventionally expected behaviors. How to dynamically and efficiently manage the power dispatch in a distribution network to balance the supply and demand by considering the variability of DGs and loads becomes a significant research issue. In this paper, a multi-agent system (MAS) was proposed to solve this problem through introducing five types of autonomous agents, the electricity management mechanisms, the agent communication ontology, and the agent cooperation strategy. The simulation of the MAS by using InterPSS, JADE and JUNE well demonstrates the performance of the system on dynamic supply and demand balance by considering both efficiency and economy.