New Algorithms Based On Decomposing And Local Search For Large Scale Global Optimization

With the rapid development of technologies in computing,the large-scale global optimization(LSGO)problems have appeared in many fields,and with the increase of the scale of the problems,these optimization problems become more complex.Thus,new efficient algorithms for large-scale global optimization problems are extremely important.There are several characteristics for large-scale global optimization problems,which can be summarized as follows: very high dimensions,lots of local optimal solutions,very high complexity of the search space and the rapid change of function values.Therefore,getting the optimal solution is very difficult.Currently,many nature inspired algorithms based on the theory of biological evolution were proposed,including evolutionary algorithms,PSO algorithms and DE algorithms,etc.These algorithms have demonstrated the good performance on tackling the LSGO non differentiable problems.However,with the dimension of the problem increasing,the performance of the nature inspired algorithms will decrease.There are two main reasons: 1)The number of local minima increases rapidly as the dimension increases.The basic evolutionary algorithms are easier to trap into local minima and would not be able to escape from them.2)The search domain becomes very large with the dimension grows.The exploration ability of the traditional evolutionary algorithms are not enough.Therefore,it is necessary to put forward new algorithms which can both search the entire area effciently and jump out local minima.A potential way is using cooperative co-evolutionary algorithms(CCEAs for short).They are a kind of global optimization algorithms which use divide and conquer strategy to divide LSGO problems into smaller scale sub-problems,and introduce the cooperation of these sub-problems and local search.Currently,researchers have invented a series of methods including DECC,DECC-G,DECC-DG,MLCC,CCVIL,CPSO-SK and CPSOHK,CCPSO,CCPSO2 for LSGO problems.However,the variable grouping strategies of these algorithms are not effective enough,or even oftenly unsuccessful.Also,the efficiency of the local search schemes are not satisfactory.To overcome these shortcomings,this paper proposes two improved CCEA algorithms.The nmain works are as folllows:1.An improved DECC-DG algorithm called DECC-TSDG is proposed.The variable grouping scheme DG in DECC-DG is improved and a new variable grouping scheme with two steps called TSDG is proposed.TSDG spend less time during grouping process than DG without sacrificing the accuracy of grouping results.Thus the new algorithm saves computing resources which is very valuable and makes the grouping process more efficient.2.A new variable dependence criterion is proposed first.It considers the situation in which overlap variables appear in more than one groups and thus can be applicable to more problems.Then by introducing Alpha stable distribution into DE,an impaoved De is proposed and it is used as a local search algorithm.In this way,computing resource is saved and the local search ability is enhanced.As a result,the better solution can be obtained.Finally,based on these,a new algorithm callled Alpha-DECC-IOG is proposed.3.The experiments are conducted for the proposed algorithms and the results show the proposed algorithms are effective and efficient.