Improving the Performance of Heuristic Searches with Judicious Initial Point Selection

Abstract

In this paper we claim that local optimization can produce proper start point for genetic search. We completely test this claim on partitioning problem and on the performance of genetic search in a real problem that is finding aggregation tree in the sensor networks. The presented method (named Tendency algorithm) increases the performance of heuristic searches, and can be used in parallel with other tuning methods. The paper justifies the logic behind tendency algorithm by measuring the "entropy" of solution (in regard to optimal solution), and by numerous empirical tests.