Research On Drift Effect And Utilization Of Genetic Algorithm

Genetic algorithm is a widely used optimization algorithm with many advantages,but it is easy to fall into local optimum and low precision in the coupling function.At present,the commonly used genetic algorithm had joined the elitist strategy to speed up the convergence of the algorithm and improved the accuracy,but the algorithm had a problem of premature convergence.In this paper,we used a standard genetic algorithm without elitism strategy to find out that population of genetic algorithm did not converge on a certain mountain,but it gathered on different peaks in different time periods.The phenomenon that population drifts from one peak to another is called the drift effect of the genetic algorithm.In this paper,we experimented with multiple benchmark functions,using visualization methods to track and recorded the specific location of each individual when the genetic algorithm had a drift effect.By analyzing the experimental results data,this paper further summarizes the following series of laws that occur: 1.Remove the elitism strategy,2.Fitness function is a coupling function,3.Need to adjust the fitness function to make the function flat.In addition,the commonly used genetic algorithm used mutation that mainly used floating-point coding.Few variables were changed at the same time by floating-point mutation.This mutation limited the search orientation of the genetic algorithm.Combining the advantages of floating-point coding and binary bit mutation,this paper proposes a floating-point number mutation method.The mutation method has used the binary coding form stored in the floating-point hardware to perform bit mutation,which has enhanced the global search capability of the algorithm.Combining the drift effect of genetic algorithm and the improvement scheme such as removing elite strategy,this paper proposes a drift genetic algorithm.In the previous iteration,the genetic algorithm has removed the elitism strategy and only recorded the best individuals which did not join the population.At this point,the algorithm converged to a certain peak.After a number of iterations,if the population remains stable at this peak,the genetic algorithm appropriately adjusts the fitness function to cause the population to converge quickly to another peak.Repeat the above operation until the genetic algorithm did not drift,and the previous iteration was completed.In the later iterations,the genetic algorithm joined the elitism strategy and added the best individuals of the previous record to the population for iteration until the maximum number of iterations set by the algorithm.Finally,it is proved by experiments that the drift genetic algorithm has stronger global search ability,higher stability and higher precision than traditional genetic algorithm.