1 Abstract-- The configuration of a power distribution system can be altered by changing the status of switches on the feeders. When a fault occurred on a distribution system feeder, the faulted section is isolated by opening the switches adjacent to the fault. When the fault is isolated, the system is reconfigured so the outage areas due to the fault can be re-energized. In this paper, a BDI (Believe-Desire-Intension)-based multi-agent architecture is proposed. The agents can response to the changes of the environment and take necessary conversation with other agents to reach the final goal, supplying services to all de-energized customers. These agents act independently but co-operatively. Decision tree is used in the proposed agent architecture for making restoration plan. Decision trees are built during the restoration process based on the fault location and load conditions through communications between agents. By using the decision tree, a more robust decision can be achieved by agents. Simulation results show that the proposed BDI-Based agent architecture is capable of proposing adequate restoration plan as expected. Index Terms-- Distribution Systems, Intelligent Agent, Service Restoration I. INTRODUCTION OST of the systems developed in the past decades that are used to solve power distribution system problems are centralized computation system. Many research results have been proposed [1-5]. In this paper, distributed intelligent agent architecture is proposed in order to develop the power distribution service restoration planner. Based on the decision trees developed on the fly in an agent, the proposed agent architecture is able to proposed adequate service restoration plan following the occurrence of a fault. Also, the agents will act autonomously and interact to each other after the fault in order to start the service restoration task. Each agent stores only local topological information. When it is necessary, agents will communicate with adjacent agents in order to acquire necessary information for decision on service restoration. By storing only local information, the This work was supported by the National Science Council, Taiwan, under NSC 97-2221-E-027 -110 Y.T Pan is with the Graduate Institute of Automation Technology, National Taipei University of Technology, Taipei, 10608, TAIWAN (e-mail: pytsung@tpts7.seed.net.tw). M.S. Tsai is with the Graduate Institute of Automation Technology, National Taipei University of Technology, Taipei, 10608, TAIWAN (e-mail: mstsai@ntut.edu.tw). communication bandwidth can be reduced. Furthermore, no information synchronization is needed since the information stored in each agent are local information. The proposed intelligent agent architecture not only can perform group restoration, a better zone restoration plan can also be identified by utilizing the restoration decision tree. Since agents are running in parallel, the zone restoration process can be performed simultaneously to improve the performance. In the proposed architecture, agents are installed in switches and feeders. The tasks of switch agents include monitoring and recording the operation conditions, responding and requesting information during an occurrence of a fault, etc. Feeder agents monitor the total connected loads and response to the requests from switch agents. Information exchange is important for the agent-based feeder service restoration since the operation condition of feeders may change. In order to make a correct restoration plan, acquiring the operation condition is very critical. In the proposed intelligent agent architecture, every agent is capable of initiating restoration process autonomously when the agent detects an occurrence of a fault. A restoration leader agent will be identified first by exchanging the information after a fault has occurred. This agent will initiate the necessary communication and try to find an adequate restoration plan based on the acquired topology and loading conditions. The leader agent is also responsible of making restoration plan based on received information. II. BDI AGENTS Multi agent system (MAS) is a new research field in artificial intelligent field. The main reason of highly interest by researchers in this field is that MAS is capable of acting autonomously. If the decision cannot be made due to insufficient information, communication between agents may be performed in order to accomplish a common goal. In this paper, BDI agent architecture is used in order to implement a distribution feeder service restoration planner. The internal state of a BDI agent can be represented by a specific structure. Each BDI agent contains four main components: Belief and Knowledge, Desires and Goals, Intension, and Plans or Obligations [3-7]. The structure of such a system is shown in Fig. 1. Development a BDI-based Intelligent Agent Architecture for Distribution Systems Restoration Planning Yen-Tsung Pan and Men-Shen Tsai, Member, IEEE M