Research And Application Of Improved Interactive Genetic Algorithm

Optimization objectives of the optimization problems with implicit objectives cannot be specified by explicit mathematical functions.This kind of problem only regards the subjective requirements of users as the optimization goal,but the user subjective requirements are easily affected by factors such as the users' circumstances and the external environment,which makes it difficult to quantify the optimization objective function.There are many such problems in practical applications,such as product modeling design,image classification,recommendation systems,etc.The Interactive Genetic Algorithm(IGA)is one of the effective methods to solve optimization problems with implicit objectives.And it has been widespread focused by scholars at home and abroad.The running process of IGA algorithm involves frequent human-computer interaction,which leads to user fatigue and low efficiency of the algorithm.Therefore,the above two problems are considered in this dissertation.The main researching content is as follows.(1)An interactive genetic algorithm with evaluation deviation correction mechanism(EDCM-IGA)is proposed.Firstly,the user evaluation process is divided into three phases according to the users' cognitive law.In the first stage,because the user's cognition of the target individual is not clear enough and there is a big deviation in the evaluation of the evolutionary individual's fitness value,an evaluation deviation correction model is proposed to reduce the evaluation deviation.In the second stage,the users are constantly familiar with the target individuals who need to be evaluated,and the algorithm selects representative evolutionary individuals for the users,which reduces the number of user evaluations.In the third stage,the users are in a state of fatigue,so the algorithm takes the place of the users to estimate the fitness value of the evolutionary individual,which reduces the burden of user evaluation.Finally,the dissertation does a contrast experiment between the proposed algorithm and the traditional interactive genetic algorithm.The results show that the proposed algorithm can effectively improve the accuracy of user evaluation,reduce user evaluation time,and ease user fatigue.(2)An interactive genetic algorithm with the surrogate model of a sample-augmentation extreme learning machine(SAELM-IGA)is proposed.Firstly,this model is based on some individuals that have been evaluated and increases the number of training set samples by means of sample augmentation.Then,the model which is based on the augmented training set samples,is trained.And it completes the individual evaluation task in place of the users.Finally,in order to verify the validity of the model,surrogate model,such as sample-augmentation extreme learning machine,traditional extreme learning machine,support vector machine and weighted extreme learning machine are combined with the interactive genetic algorithm,which is applied to the carpet design system,and the contrast experiment is carried out.The experimental results show that the surrogate model based on sample-augmentation extreme learning machine can effectively reduce the number of user evaluations,shorten evaluation time,improve the search performance of the algorithm,and ease user fatigue.