Research On Termination Detection Strategies Of Evolutionary Algorithms

The use of Evolutionary algorithms(EAs)to solve practical problems has become an important research area.In the EAs,the termination condition of the algorithm is one of the key factors affecting the effectiveness of the algorithm.If the algorithm stops too early,it is difficult for the algorithm to find an approximate optimal solution;if it stops too late,the algorithm will waste a lot of computing resources.But when there is no gradient information available,it is usually impossible to check whether it formally converges or whether the detection meets the optimal criteria.By setting the maximum number of iterations or the maximum number of evaluations to determine whether the algorithm stops or not,it is necessary to have a deep understanding of the problem and the algorithm used,so it is difficult to apply to the actual engineering optimization problem.The main contents of this thesis include:(1)This paper summarizes the existing termination conditions of EAs,provides a summary block diagram of the evolutionary algorithm research on termination conditions in the past few decades,and classifies the existing evolutionary algorithm termination conditions from the performance indicators and the termination criteria.(2)Six kinds of popular online stopping strategies on EAs are analyzed and summarized.Experiments are carried out on classical single-objective algorithms and multi-objective evolutionary algorithms.The quantitative indicator which can objectively evaluate the effect of termination condition is designed.The results are briefly analyzed and summarized the scope,stability and advantages and disadvantages of each termination decision strategy.(3)In order to solve the problem of computing resource consumption of online stopping strategy,the online stopping strategy is combined with curve fitting and regression prediction,and an online stop strategy based on prediction is proposed.This paperverifies that the termination strategy of EAs is combined with a variety of different knowledge,which can better balance the relationship between the quality of the solution and the computing resources.