Hybrid Approach and its Generalization for Solving Premature Convergence of a Class of Genetic Algorithms

In this paper, the author analyses the mechanism that VCGA(variants of canonical genetic algorithms) may sometimes produce premature convergence, suggests a hybrid approach called HVCSDA(hybrid VCGA combined with steepest descent approach), and generalizes HVCSDA in order to broad its application. The approach can make the time series of super individual best maintained leave the state of premature convergence near to the global optimal solution. Two simulation examples show the efficience of HVCSDA and its generalization. In the benchmark problem of the 30 cities TSP(traveling salesman problem), the lenghth of routing is 6.82 by the HVCSDA's generlization. It is better than one that is 6.99 by new, modern heuristic search method——TABU search.