From the predecessor [6,7] , it has been found that RPBIL has very good convergence rate compared to some established EAs . Nevertheless , one weak point of the method is that , as the probability matrix relies on a current best solution , it could produce a population whose members are only the neighbours of the best solution . That means a premature convergence can possibly take place . In order to alleviate this problem , mutation and crossover of DE [9] will be in corporated into the procedure of the multiobjective RPBIL and this hybrid algorithmis defined as RPBIL-DE . The hybridisation is carried out in such away that , for every generation , a newly generated population from the probability trays will be recombined with members inthe Pareto archive by means of DE operators before performing function evaluation . More details of the procedure are given in Algorithm 3 where F is a scaling factor, pc is a DE crossover probability , and CR is the probability of choosing an element of an offspring c . The concept of RPBIL is a simple type of estimation of distribution algorithms (EDA) . The rationale for integrating DE operators into the procedure of RPBIL is because it has been reported that a hybrid of EDA and DE successfully improved evolutionary search performance for single objective optimisation . For example , the mixed distribution based univariate estimation of distribution algorithm (MUEDA) was employed in combination with DE in [43] . On the other hand,it has been shown that the use of a Gaussian model with diagonal covariance matrix (GM/DCM) for EDA combined with DE [38] . It has been shown that the hybrid versions were superior to their original EAs
đang được dịch, vui lòng đợi..
