The Method And Application Of Service Agentcoalition Formation

The SOA (Service Oriented Architecture) is becoming more and more popular. Web service technology is becoming the first choice to settle the collaboration issue in inter-enterprise and inner-enterprise environment, which helps generating loosely-coupled application system and bringing more business opportunities. While Agent is autonomous entity, which has the ability to make decision, to collaborate, and has the some kind of intelligence. Combined service and agent together, we got Service Agent, which is a special agent. Service Agents manage a list of web services, optimize the choice, and offer a high-level service to the public. However, a single Service Agent has a limited range of services, and its capability is limited. Therefore, when multiple services are needed to complete a complex task, a number of Service Agents need to form coalition in order to deal with. So, the research of Service Agent Coalition Formation problem is of great significance.Agent Coalition Formation is a basic operation in a Multi-Agent System (MAS). Agents usually need to join an agent coalition in order to complete tasks in a collaborative way or to improve the capability of the coalition. Currently, the existing coalition formation algorithms do not take the internal task structure into consideration, but only simply count the capabilities this task required. In this paper, we consider the complex internal structure of the tasks. Such problems are of more practical significance, because in real life, there is a class of complex tasks. This kind of complex task consists of a set of sub-tasks. The logical dependency between the subtasks also needs to be considered. The logic dependency is described by BPMN, which allows the task to be a process or a workflow. When Agent coalition is formed and sub-tasks in the complex task are all completed by Agents, the payment of this complex task is gained.The existing Agent Coalition Formation algorithms, consider how to minimize the executing cost of Agents. However, due to the interdependence between the sub-tasks in a workflow, so it is of significance to also consider the transfer cost between the adjacent sub-tasks. Usually, it is easier if two adjacent sub-tasks are completed by the same Agent or the Agents near each other, so, in order to reduce the transfer cost, we also encourage each Agent to complete a continuous snippet of the sub-task workflow.In response to these issues, the main efforts of this paper are as follows:1. Firstly relevant researches on Service Computing, Agent and Agent Coalition Formation Problem are reviewed.2. We consider the Agent Coalition Formation Problem with logical dependency and transfer cost. And, we propose an algorithm with polynomial time complexity on the basis of Dynamic Planning, to solve this problem. This algorithm is a centralized algorithm. It can solve the problem very efficiently, when Service Agents agree to share cost information.3. We also consider the same problem when Service Agents do not agree to share cost information. We need to find a mechanism to solve the problem and also keep the total cost in a reletively lower. We propose a distributed solution to deal with this situation.4. We also researched and developed a Service Agent platform, to create and run Service Agents. And also we integrate our Service Agent Coalition algorithm in our CIT Workflow System.