PROJECT TITLE :

An Estimation of Distribution Algorithm With Cheap and Expensive Local Search Methods

ABSTRACT:

In an estimation of distribution algorithm (EDA), world population distribution is modeled by a probabilistic model, from that new trial solutions are sampled, whereas individual location information is not directly and absolutely exploited. During this paper, we tend to counsel to combine an EDA with low cost and expensive native search (LS) strategies for making use of each world statistical information and individual location info. In our approach, part of a brand new resolution is sampled from a changed univariate histogram probabilistic model and the remainder is generated by refining a parent answer through a low cost LS methodology that does not need any function analysis. When the population has converged, an upscale LS methodology is applied to enhance a promising solution found so way. Controlled experiments are carried out to investigate the consequences of the algorithm parts and the management parameters, the scalability on the number of variables, and therefore the running time. The proposed algorithm has been compared with 2 state-of-the-art algorithms on two take a look at suites of 27 check instances. Experimental results have shown that, for straightforward check instances, our algorithm will produce higher or similar solutions but with faster convergence speed than the compared methods and for a few sophisticated check instances it will realize higher solutions.