Population Diversity Model Based On Dimension Entropy And Its Application

Swarm intelligence optimization algorithm is an optimization algorithm born from the observation and simulation of the behavior pattern of biological population in nature.Since its birth in the 1980 s,it has been attracting great attention.Many algorithms in its field,such as particle swarm algorithm,genetic algorithm,etc.,have been widely and successfully applied,and become a new way to solve optimization problems.However,the research on population diversity of Swarm Intelligence Algorithm is still relatively immature.On the one hand,there are few population diversity measurement models specifically designed for swarm intelligence algorithms.On the other hand,the research of swarm intelligence algorithm is rarely discussed from the perspective of population diversity.From the perspective of population diversity,the variation of population diversity in the iteration process of different swarm intelligence algorithms and its effect on algorithm performance are studied.Main work includes: first,after studying the existing several kinds of species diversity,based on the measurement model,the dimension entropy is proposed as a new method to measure species diversity,and compared with other models,the experimental results show that the dimension entropy can reflect the diversity of the population.and the dimension and the changes in population size has stronger robustness.Secondly,the effect of population diversity on the performance of swarm intelligence algorithm is studied.The results show that the algorithm with better performance has higher average population diversity.In addition,premature loss of population diversity is a possible reason why swarm intelligence algorithms fail to achieve good results.Third,an entropy based on dimension control population diversity strategy,this strategy in each iteration,when the population diversity from a predetermined curve,by deleting or copying group repeatability of the largest "redundant" particles,directly affects the population diversity,so as to control algorithm according to the predetermined curve for convergence.On this basis,the influence of different diversity convergence curves on different test functions is studied.Fourthly,this strategy is applied to three classical swarm intelligence algorithms,namely particle swarm optimization algorithm,backbone particle swarm optimization algorithm and differential evolution algorithm,and the effectiveness of this strategy is verified by performance comparison with the CEC2017 test function.In this paper,a new diversity model is proposed,and a strategy of updating according to the diversity guidance algorithm is further proposed.The results show that the population diversity is closely related to the performance of the algorithm.By controlling the population diversity,the algorithm will not fall into prematurity in the early stage,and the smooth convergence in the late stage can greatly improve the performance of the algorithm.