Research On Search Strategy Of Evolutionary Computation

The optimal theory is a mathematical branch, the research focuses on some mathematical problems' optimal solutions, namely to the actual problem which produces, finding the optimal solution from the multitudinous candidate solutions. It has highly application and technical characteristics.Nowadays, research in linear programming (LP), non-linear programming (NLP), random programming (RP), multi-objective programming, integer programming (IP) has grown tremendously. New methods continuously emerged, and the practical application widely used. Evolutionary computation which is used to solve complex , difficult global optimal problems has rapid progress. As evolutionary computation developed, algorithm design is its key as well as the emphases people focus on and research in. The research revolves two subjects: First, to expanse the application domain; Next, to cause it more effective. The former is for the goal of designing and discovering effective EC search strategies, solving the problem the past unable to solve or cannot to be solved effectively; The latter is to revise and improve emphatically the algorithm, causes it to be more effective.The dissertation revolves the two subjects of EC algorithm designing, unifies the characteristics of EC, uses essential characteristics of traditional algorithms, has conducted the thorough research to the EC search strategies, and has yielded a series of satisfactory results. The dissertation consists of three major parts.The first part is the optimal theory, basic EC theory analysis summary and the research on EC search strategies. It deals with traditional algorithm search strategy, summarizes their essential characteristics. As to the conflict of EC generating new individuals randomly, we proposed a neighborhood search strategy based on similarity. Neighborhood search brings the ability to self-adaptively generate new individuals easily as well as deal with all kinds of optimization problem in the uniform frame. Aiming at balancing search results and search speed, we proposed a search strategy to classify the individuals by their fitness. By individuals classification to differentiate respective function in search process, that's the excellent individual to explore the local optimal solution and others to explore the search domain to find new local optimal solution. Aiming at EC bad efficiency, uniting traditional algorithm and numerical analysis's speed up thought, we proposed a speed up search strategy by combining parent population' information. Aiming at different conditions, we design 3 speed-up operators and apply them in the algorithm. The search efficiency is highly improved. To improve the convergence speed, with simulate annealing, we design a neighborhood contractive technical. The search speed is enhanced. According to search strategies mentioned above, united EC characteristics, we proposed a new algorithm framework SFEC (SimilarityFrame of Evolutionary Computation).The second part is to check the ability of the algorithm frame. As to functions optimization, integer programming, multi-objective optimization, TSP with their special characteristics, the respective algorithm designed under the algorithm frame. In the case of functions optimization, ten normal test functions are adopted. The results show that algorithm can not only find 9 test functions, bump function under 30 dimensions global optimal solutions also has quickly speed and strong numerical reliability. As to the bump function from 30 to 100 dimensions, the algorithm can find the global optimal solution in short time but the numerical reliability is not so ideal. As to bump function above 150 dimensions, the algorithm is hard to find its global optimal solution. In the case of integer programming, using 2 test functions the algorithm find the global optimal result in very short time. In the case of multi-object optimization, the algorithm solves the 5 classical test problems successfully. In the case of combination problem, the author test TSP, which turns out in low dimension satisfactory results obtained but in high dimension deep further research should be carried.The third part is the application. SFEC has a potential for a wide range of application. We apply the algorithm into typical problems. We use network topology design and self-adaptive management of distributed information systems as the examples in network optimization, to build and solve the mathematical models. As for designing network topology, we take common network reliability and its cost as object functions to build the mathematical model, and as for ATM network, the burden and its cost are used to build the model. As for self-adaptive management of distributed information systems, we introduce 3-dimensional variable to build the mathematical model on file content. These are all multi-objective optimization problems. We use the SFEC to solve them by weight allocation. The numerical experimental results indicated the new algorithm can solve the network optimal problems effectively and the mathematical model can reflect the features of the system, also satisfy optimal characteristics.At last, the research results in the dissertation are summarized. We suggest that more research work in this area should be done, and predict bright prospect of future research.