Research On Natural Computation Model And Its Application In Optimal Computing

Natural computation which includes a group of nature-inspired computing models and algorithms is a general terminology; its main research content covers artificial neural networks, genetic algorithms, immune algorithms, particle swarm optimization, and ant colony algorithms and some other models. Owning advantages of ubiquity, global search ability,parallelism, low requirements in such characteristics of the problem to be solved as continuity, differentiability and convexity of solution set and so on, natural computing models are superior to traditional approaches in solving some complex optimization problems. As a result, they have been widely used in many fields such as optimization problems, intelligent controlling, pattern recognition, network computing and hardware designing and so on.Aiming at genetic algorithms, particle swarm optimization, artificial immune algorithms and some relative optimization problems, the main contributions of this thesis can be described as follows.1. This theis summarizes the design methods of crossover operator in good point set genetic algorithm under the conditions of real-value coding and binary coding. Then we pointed out that if we got little prior knowledge to the problem, the uniformly distributed initial population was considered easier to find the optimal solution than random initial population. Based on the good point theory, we proposed an approach to generate uniform population, and experimental result on cylindricity error evaluation shows the effectiveness of the proposed method.2. A fact is that when the length of chromosome was fixed, the discretization error derived from mapping between a binary and a real number is inevitable. Aiming at the domino phenomenon of convergence from the highest position to lowest position of binary coding in good point set genetic algorithm, a zooming factor is proposed to lengthen the length of chromosome indirectly to minimize the discretization error, so the search efficiency and solution accuracy are improved. The simulation results based on Benchmark test functions of different dimensions verify that the proposed good point set algorithm with zooming factor has the advantage of global convergence, high precision and search efficiency.3. This theis introduces the niche technique into particle swarm optimization, and the niche entropy is constructed to measure the diversity of the population on the basis of an adaptive niche identification approach and the evolutionary parameters of the PSO algorithm can be adjusted adaptively according to the niche entropy of population. Moreover, an effective good point approach to explore new schemas in search space is designed, which makes the diversity level of population always higher than the threshold set beforehand during the evolutionary process, so the proposed algorithm can obtain strong ability to search out the global optimal solutions. And a novel sequential multi-population niche PSO algorithm which aims at find out all the global and local optimal solutions is proposed. Experiments for the multimodal function optimization show that the proposed algorithm has strong adaptability and convergence.4. The aim of multi-objective optimization algorithm is quick to find out the Pareto optimal solutions which converge to the ideal Pareto front with a good performance in diversity. Based on the immune clonal theory, we introduce the fitness sharing strategy, and then a new multi-objective optimization evolutionary algorithm with good performance in diversity is proposed for maintaining the diversity of solutions. The proposed algorithm employs an external archive to preserve the nondominated solutions. The principle which includes sharing fitness and Pareto domination is used to update the external archive mentioned above and select the active antibodies for generating offspring. Moreover, for enhancing the search ability in decision space, we design the good point searching approach which can generate the good point set with uniform distribution. The proposed algorithm is tested on several multi-objective optimization problems and compared with many classical methods, much better performance in both the convergence and diversity of obtained solutions is observed.