Evolutionary Algorithms With Cooperative Coevolution

Optimal problem is an important issue and wildly existed in our daily life and work. As an efficient heuristic optimization algorithm, evolutionary algorithm has been get great progress and developed a series of new algorithms and techniques, such as GP, DE, PSO, ACO and so on. Those classic algorithms can reach better result in low-dimension problems and they can find the last optimal solution correctly and quickly if there are litter variables. But with the variables number growth, the problem get complex, many algorithms which have excellent performance in low-dimension problems lost their performance. In actual project, many optimal problems have large scale, and may also existing variable-correlation problem. For example, density, consumption and heat dissipation problems exist in the designing of Integrated Circuit(IC), which also need powerful optimization algorithms.Recent years, many researches in high dimensional numerical optimization problems mainly concentrated in the variable correlation, solution space reduce, evolution state judgment, algorithm selecting and using of adaptive methods. This thesis mainly focuses on the design for large scale global optimization problems. In this paper, based on particle swarm optimization (PSO) algorithm and cooperative co-evolution, we studied some techniques in large scale global optimization. The main work and the innovation are summarized as follows:1. Studied cooperative co-evolution based particle swarm optimization (CCPSO) algorithm, analyzed the performance of CCPSO in large scale global optimization..2. When using CCPSO algorithm, we tracked the particle velocity and found a phenomenon that the velocities of all the particles of a dimension become divergent when the real optimal solution of such dimension is near the bound of the search space. The divergence o f velocity will decrease the convergence speed of algorithm. Based on this observation, proposed velocity divergence detection and re-initialization technique dealing the divergence velocity and achieved good results.3. Studied the technique of combination multi evolutionary algorithms. Based on "no free launch", using competitive learning method, tracked the performance of each offspring algorithm. Good performance algorithms have more probability to participate optimization though roulette wheel selection so that different optimization problems can choice a better algorithm suitable for themselves. 4. During the implementation process of evolutionary algorithms, the search area will also change over time. Different fitness landscape in different search area need for different algorithms. In this paper, through analysis the distribution of the particles to know the current state of evolution, then using different algorithm to optimize and achieved better results.