Research And Application On Adaptive Evolutionary Algorithms

Evolutionary algorithm is good at solving hard combinational optimization prob-lem, but the full potentiality of a parameterized evolutionary algorithm won’t be achieved until the parameters are fine tuned. Parameter control is a focused issue in evolutionary algorithms. The key of parameter control is to assign right parameter values at the right time in a run. In this article, we propose the Fitness Level based Adaptive Operator Se-lection (FLAOS). In FLAOS, the discovered objective values are divided into intervals, the fitness levels. A parameter assigned table corresponding to a fitness level describes the selection probabilities of a set of operators. An evolutionary algorithm with FLAOS is suggested to solve one-dimensional cutting stock problems (CSPs) with contiguity and the structure learning of bayesian network.Compared with the previous evolutionary algorithms, FLAOS has advantages as follows:1. Parameters in algorithms needn’t be predefined, an this will reduce the cost a lot.2. Because the parameters is altering in the run, for different problem instances, we can also obtain the best parameters so that get the better performance.Experimental studies have been carried out to test the effectiveness of the FLAOS. The solutions found by FLAOS are better than or comparable to those solutions found by previous methods.