Solving JSSP by Introducing Hamilton Similarity and Time Dependent Fitness Scaling

Abstract

In this paper we proposed and tested a niching genetic algorithm (GA), which for comparison of individuals in the population uses so-called Hamilton similarity. The advantage of the Hamilton similarity lies in the fact that there is no need for context sensitive information in order to successfully compare two population members. Furthermore, the algorithm was tested on the famous Job Shop Scheduling Problem (JSSP) - benchmark mt10, and statistical results of the test were given. Significantly smaller standard deviation of the proposed GA compared to Simple GA clearly demonstrates its superiority. In addition to the Hamilton similarity, time dependent fitness scaling was proposed which in conjunction with niching significantly reduces the probability of the algorithm to get stuck in one of the less desirable local optimum. Finally, suggestions for future research are given.