A Study On The Evolutionary Algorithm Based On The Behavior-Division And Their Application

Conventional evolutionary algorithms are usually to achieve optimization with a population, there will be a global search (slow convergence) or premature convergence and parameter setting complex and affecting the application performance. However, The behavior-division in multi-population can more effectively to solve the contradictions between global search and fast convergence, the idea has been confirmed in team progress algorithm(TPA) .In this thesis, two Dual-Population differential evolution algorithms are presented, namely Dual-Population differential evolution (DPDE) algorithm and Dual-Population Hybrid difference evolution (DPHDE) algorithms. The former is DE algorithm combines the principle of Dual-Population Behavior-division search to implement the performance of global search and fast convergence, the latter is further to combine with the memory characteristics of particles in particle swarm (PSO) algorithm to implement a fast optimization and global convergence performance. The numerical experiments show that, the DPDE and DPHDE have these performance advantages of global search capability and fast convergence and the less computation, When comparing the two algorithms ,there are shown that the DPDE algorithm with higher stability and simpler parameter, the DPHDE with faster convergence speed and litter computational.These proposed algorithms are applied to amplitude and phase optimization examples of the directional pattern synthesis of linear array, getting these results of a piece side-lobe level will be reduced to the same value of the two new algorithm is better than DE algorithm requires less computational.