An Effective Near-Optimal State-Space Search Method: An Application to a Scheduling Problem

Abstract

This paper presents aneffective near-optimal search method forstate-space problems. The method, LTA^ast(Learning Threshold A^ast), accepts athreshold parameter, p, as an input and finds asolution within that range of the optimum. Thelarger the parameter, the faster the methodfinds a solution. LTA^ast is based on acombination of recursion and dynamic memoryand, like A^ast, keeps information about allstates in memory. In contrast to A^asthowever, which represents each node as acomplete state, LTA^ast represents each nodeusing an operator. This representation of thenodes makes LTA^ast dramatically efficientwith respect to memory usage. Another advantageof LTA^ast is that it eliminates any need forcomputational effort to maintain a priorityqueue, and this elimination significantlyincreases speed. To test the effectiveness andefficiency of the method we have applied it toNP-hard problems in scheduling. The testresults indicate that the method is effectivein trading speed with the quality of solutionsand that it is efficient in producing solutionsfor p = 0.