Mind Evolutionary Computation--Development Of Searching Strategies And Analysis Of Algorithm Performance

Evolutionary computation is a computational model that simulates evolutionary process in nature, and has been paid many attentions and studies by many researchers in the world. In the 1960s, genetic algorithm (GA) was proposed by professor J. Holland from Michigan University. In the passed 30 years, the theoretical study and application of GA are most active, and has rapidly become a highlight in the fields of international academy. However, there are many shortcomings in GA, which should not be neglected. In the early stage, researchers have noticed the problem of GA's prematurity. Another crucial problem is computing efficiency of GA, due to non-directionality of its evolutionary mechanism.To overcome the problems of GA, Mind Evolutionary Computation (MEC) was proposed by Chengyi Sun in 1998, which imitates two phenomena of human society - similartaxis and dissimilation. After several years' studies on its theory and experiment, MEC has been made great progress. So far, a preliminary system has already been established for MEC.In this thesis, based on the review of EC and MEC, the performance of MEC is discussed, including the development of two strategies, comparison and analysis of searching efficiency, consideration on the behavior of MEC in high-dimension space, and comparison of several algorithms. The originalities in this thesis are as follows, which enriches the framework of MEC.Firstly, two new strategies of MEC are designed. One is a dissimilation strategy by rejected regions for avoiding searching repeatedly in dissimilation process and improving MEC computing efficiency. And another is prediction-based similartaxis strategy, which can enhance the search efficiency in similartaxis process and has much stronger adaptive ability than basic MEC.Secondly, the search efficiency of MEC is discussed. In this part, a series of functions are built to test several different algorithms. Also, the computing cost and the search efficiency are defined. Then two measurements - search efficiency and convergence rate, are given to compare searching performance of algorithms. Experimental results show that MEC has better performance, especially for strongly deceptive problems, superiority of MEC is pretty obvious.Thirdly, the optimization performance of MEC in the high-dimension space is tested. The complexity of the high-dimension function is considered, and a two-level MEC algorithm under Windows is developed. The first-level MEC is adopted to optimize the parameters of the second-level MEC, while the second-level one is used to solve function optimization. Experimental study has been carried out with this two-level MEC algorithm in high-dimension space. The results show that computing cost is a power function of the dimension.Furthermore, operations similartaxis and dissimilation are examined to explain the reason that MEC has high search efficiency. As been analyzed, searching of MEC is not carried out in the whole solution space with equal intensity, but in some focal regions. Meanwhile, within each selected region, the search focus only on a narrow band. Those features of MEC explain clearly why it has excellent performance.Finally, several famous Evolutionary Algorithms (EAs) are introduced, and all of them have attracted much attention of the researchers recently. After the essences of these algorithms are studied intensively, their behaviors are discussed objectively, and their differences in the principles are compared. So we will rightly understand the status of MEC in the EAs, benefit from the advantage of these algorithms, and further complete the framework of MEC.