The Influence Of Noise Environment On Multi-objective Evolutionary Optimization

The evolutionary multi-objective algorithm has many advantages such as parallel operation,iterative optimization and good convergence.It does not need a particularly complicated prior knowledge background,and it has the characteristics to solve today's large number of multi-objective problems,making evolutionary algorithm a powerful research tool in the random search algorithm.First of all,evolutionary algorithm can converge to the optimal solution or the most optimal solution of the class in the current feasible domain.The essence of evolutionary algorithm is the screening and retention mechanism of individual population.The judgment of this mechanism comes from evaluating the fitness function of individual population.The individuals of the whole surviving population are obtained by the corresponding individual screening and deleting operation according to the fitness value function set by the algorithm.However,when the random noise of this uncertainty into the added internal evolutionary algorithm,evolutionary algorithm for population screening retention mechanism will be seriously affected,this phenomenon will happen to change is very bad performance of evolutionary algorithm: Pareto dominance relationship between individuals is disordered,filled with a large number of populations non dominated solution enables decision makers to haunt the feasible solution,and even make the evolutionary algorithm will not be able to search the optimal solution in the feasible domain of existence.Through the analysis of the nature of the phenomenon,is not difficult to find the root of the problem is that adding random noise makes the mechanism of evolutionary algorithms in the evaluation of individual fitness value numerical function on a certain amount of noise added to the random function,which makes the algorithm in the screening of fitness and prone to false positives when judging the individual,because this will cause misjudgment good individual loss thus making the algorithm cannot search to the optimal feasible solution exists.In view of the existence of this phenomenon,the work of this paper is divided into the following three aspects:1.The performance of evolutionary multi-objective algorithm under noisy environment is sure to be affected.Does the performance of the evolutionary multi-objective algorithm in the noisy environment still can be evaluated by the representative evaluation index? In this paper,It has evaluated the performance of evolutionary algorithm based on the three representative indicators of generational distance,spacing metric and hyper-volume in noisy environment.The results showed that the generation distance,spacing metric and hyper-volume index of non dominant population in the reality situation were not consistent with the population size and distribution of the non dominant population in the noise environment.So these indicators cannot effectively reflect the performance of multi-objective evolutionary algorithm in noisy environment.2.In this paper,a new evaluation index is put forward for the phenomenon,which the failure of evolutionary multi-objective performance evaluation index in noisy environment.The corresponding operation process of the new index is given,and the accuracy of the new index is tested and verified for the new index by testing function.Through the analysis of experimental data and results,it can see that the new index proposed in this paper is effective enough to evaluate the performance of the evolutionary multi-objective algorithm in the noisy environment,and We can see that the influence of different noise levels on the performance of evolutionary multi-objective algorithm is strong or weak by the new index.3.Multi-objective evolutionary algorithm in noisy environment is unable to achieve the expected convergence effect,therefore,this paper proposes an improved domination relationship is applied to the classical multi-objective evolutionary algorithm.It is concluded by this paper that the improved domination relationship in a certain extent improves the situation between the disordered dominating relationships among the individuals in the noisy environment.