Intelligence Algorithm-Based Convergence Speed Controller And Its Application

Swarm intelligence optimization algorithm, a kind of new optimization method, has been widely applied to solve function optimization, neural network design, signal processing and pattern recognition optimization problem and so on. The algorithm simulates the behavior of the social community through the exchange of information between individuals to achieve the goal of optimization. As its simple implementation, it can parallel computing and has high efficiency. Last twenty years, it has attracted much attention. Although the intelligent optimization algorithm has a large number of research results, there are still many problems to be solved, such as how to solve premature convergence, or how to make better use of the limited computing resources.In this paper, we studied the improved the Differential Evolution algorithm(DE) and Particle Swarm Optimization algorithm(PSO). In addition, an application using the improved algorithms is also presented in this paper. On this basis, we proposed an algorithm framework: convergence speed controller(CSC). Our researches include the following two parts, which are consist of theory field and application field.(1) The CSC framework has been adopted to improve the performance of classic DE and PSO algorithm. The feature of CSC is that it will detect and adjust the convergence speed. And this feature improves the searching ability of the algorithms. While the CSC framework detects the convergence speed of the objective algorithm is abnormal fast, CSC framework will slow down the speed by mutating the individual or even the whole population. On contrast, if the convergence speed is not considerable, CSC will speed up the convergence speed. While the algorithm shows its disability in searching global optimization or the optimization problems are tough, the performance of CSC framework would be significant. Especially, CSC-PSO shows its outstanding performance while it is used to deal with CEC 2010 highdimension optimization problems. Moreover, CSC-PSO outperforms some state-of-art PSO variants. CSC framework helps the PSO, which will easily prematurely converge, to maintain its convergence speed on a stable and safe level until the CSC-PSO finds out the good enough optimal solution. Similarly, CSC-DE also has great performance in solving CEC’2005 low dimension optimization problems.(2) We designed software to predict the atomic clusters’ steady structure, and using the CSC-DE and CPSO-CSC as the core of the software’s algorithm. The cluster of atoms is a collection of hundreds of atoms. The stability of the atomic clusters is related to the number of atoms in the cluster and their position in space. The software makes full use of the advantage of CSC-DE in solving low dimensional optimization problems and the advantages of CPSO-CSC in solving high dimensional optimization problems. And the clusters will be divided into two classed: low and high dimension. Then the software deals with the prediction by using the two algorithms respectively. We use the Lennard-Jones, potential energy model as the objective function, to optimize the cluster structure, and returned the calculation results in the chart.