Universal Strategies For Enhancing Evolutionary Computation And Their Application For Landslides Research In Three Gorges Reservoir Region

Evolutionary computation is simple,easy to operate and versatile.Moreover,it is fit for parallel process in large scale.To obtain high quality solutions,the balance between exploration and exploitation should be maintained during the whole run of evolutionary computation.The development of landslide belongs to nonlinear dynamic system with the complexity and uncertainty.There are abundant data of landslides in the Three Gorges reservoir region.These data provide the possibility of data based landslide research.However,such research is remained to be further done.On the above background,mechanisms for maintaining the balance between exploration and exploitation which can be widely used in evolutionary computation,named universal strategies for enhancing evolutionary computation,are studied in this paper.Then,clustering of displacement time series of monitoring points in the Three Gorges reservoir region is carried out based on our study.To begin with,elitist-centric colony diversity of evolutionary computation is proposed in this paper.The exact value of colony diversity is the average of the distinct degree from each elitist to each other individual in pair.If elitism strategy is used in evolutionary computation,elitist-centric colony diversity with the same degree can express the value of colony diversity exactly.Besides,such a measure has a significant advantage on time complexity.Otherwise,whether elitism strategy is employed or not,elitist-centric colony diversity with one in degree can be used to decrease the time complexity of colony diversity computation at the cost of some error.Then,three universal strategies for enhancing evolutionary computation are proposed.They are different in application scenarios and can be used together with each other.Details are given as below.Firstly,global migration strategy with moving colony is proposed for hierarchical distributed evolutionary algorithms.In this strategy,the migration object is subpopulation.At the same time,moving emigration-replacement method is used.Experimental results show that,with this strategy,the solving ability of hierarchical distributed evolutionary algorithms can be enhanced if nonrandom topology is used.Secondly,subpopulation diversity based setting migration occasion is proposed as a strategy for distributed evolutionary algorithms.In this strategy,individuals still be migrated at intervals,but the probability of immigrant entering into local subpopulation is decided by a proposed formula.Experimental results show that this strategy can improve the solving ability of distributed evolutionary algorithms.Thirdly,distance based secondary selection is proposed for evolutionary algorithms.In each generation of evolution,secondary selection has a rate to replace main selection.In secondary selection,elitism strategy with one in degree is executed at first.Then,the distance from an individual to the elite is regarded as its temporary fitness(larger is better).After that,the selecting mode in main selection is still used to select individuals.Experimental results show that this strategy is conducive to resist stagnation or premature.As a result,better solutions are obtained in evolutionary algorithms.The remainder of this paper is an application of universal strategies for enhancing evolutionary computation in landslide research.A hierarchical distributed differential evolution clustering algorithm is realized to cluster displacement time series of monitoring points of landslides in the Three Gorges reservoir region.Clustering results provide us beneficial enlightenment for landslide prediction.