Research On Large Scale Global Optimization Algorithms Based On Variable Grouping

With the rapid development of computer technology and information science,optimization problem has been applied in engineering more and more widely.How to design a simple and efficient algorithm,so that it can obtain more accurate optimal solution with limited computing resources,has become a hotspot for solving engineering optimization problems.There are three main challenges in solving large scale optimization problems: firstly,the search space is too large,it increases exponentially with the dimension increases,so that the algorithm would get a local optimal solution possibility.Secondly,objective functions are too complex,the functions are nonlinear,non-convex,non-differentiable etc.many classic algorithms face enormous challenges to solve these problems.Third,there are many local minima,and the traditional algorithm is easy to fall into the local optimal solution.Evolutionary algorithm has many advantages in solving these problems,such as high robustness,self-adaptability,self-learning and parallel search,global search capability,does not require a differentiable objective function,so it is widely used in engineering optimization to solve some complex optimization problem of modeling,and it gets big success.However,with the increase of the dimension,the evolutionary algorithm is easy to fall into the local optimal solution,which makes the solution of the problem more difficult.Study on the evolutionary algorithm mainly has the following two challenges: 1)It is difficult to solve large-scale problems due to sharp increase of the search space.2)Premature convergence.In view of these two challenges,this article mainly makes the following several aspects:1.Variable grouping: evolutionary algorithm is facing great challenges in solving high dimensional problems,so there is a co-evolutionary algorithm based on group of variables,it divides the large-scale problem into several related sub problems,and uses evolutionary algorithms to solve every sub problem until the termination condition is reached.The difficulty of co-evolutionary algorithm is choose of the grouping strategy.In this paper,the formula based grouping is adopted,besides we make a deep grouping for the non-separate problems and fully separate problems.2.The allocation of iterative number: when the evolutionary algorithm is used to optimize the sub groups,the contribution degree of each sub problem to the original problem is different.In this thesis,we get a method to calculate the sub-problem contribution,and give more chance to the sub problem with high contribution,so it forms a competition mechanism in the sub problems.Algorithm will get more accurate solutions under the limited computational resources.3.A new algorithm for completely separable problem: completely separable problem is a collection of several one-dimensional problems.The existing evolutionary algorithms is not suitable to solve this problem.This paper presents a new algorithm based on interval sampling,uniform sampling in the feasible region,has reached the purpose of detecting the optimal solution,combined with the idea of clustering to maintain the diversity of samples and avoid falling into local optimal solution.On this basis,we construct a new separable problem solving algorithm.4.A modified local search operator: to improve the efficiency of algorithm to solve the problem,we add the local search algorithm in the solving process.We proposed an improved algorithm to simulate the function in a descent direction the point to improve the search efficiency of the algorithm.In order to test the performance of the above algorithms,we tested them on the commonly used standard test set and compared it with several well-known algorithms.The results show that our algorithm is effective.