Multi-purpose Planning for Practical Web Service Composition Problems

Many real-world applications are too complex to be specified by single goals. In this paper, we investigate the issues on how to compose such systems from existing services. First, we develop a multi-purpose model to specify the multiple goals of the system. Second, we develop planning techniques to achieve automated service composition and derive the workflow for the system which can deal with multiple goals. Most existing AI planers can only solve problems with single fixed goals. They cannot directly handle multi-purpose problems. Our approach is to prioritize the goals and perform planning to derive the partial plan (workflow) for each goal one at a time. For each new goal, the planner should start from any of the states in the current plan derived so far and try to reach the new goal. This allows the new sub plan to be naturally merged into the existing one to facilitate multi-purpose workflow construction. It also lets the original plan be reused as much as possible without duplicating the planning effort. We develop the Forward Multi-Purpose Planning (FMPP) algorithm to efficiently obtain multi-purpose plans. Case study systems is developed to illustrate the multi-purpose problem and the corresponding planning process.