Several Intelligent Algorithms For Optimization Problems

Intelligent optimization algorithm is gradually developed by the use of certain similarities between the complex systems (e.g., natural or social) and optimization problems. It obtains the next feasible solutions by the operation on a set of initial solutions in the search space according to the certain rules of probability. Therefore, the search mechanism of the algorithm determines its optimization performance.This thesis is devoted to three intelligent optimization algorithms: Particle Swarm Optimization (PSO) algorithm, Shuffled Frog Leaping Algorithm (SFLA) and Harmony Search (HS) algorithm. On the basis of these, the algorithms are further improved and generalized to obtain satisfactory results. The main works are summarized as follows:1. Standard PSO algorithm traps into local optima easily and has low convergence accuracy when it is used to address complex optimization functions. In order to overcome the shortcomings, the updating strategy of PSO algorithm is modified by the use of the attraction-repulsion mechanism in the field of biology, and then an attraction-repulsion mechanism-based PSO algorithm is proposed. Theoretical analysis and numerical experiments demonstrate that the proposed algorithm has a better optimization performance.2. The further research of SFLA is made. Specially speaking, the search mechanism of the algorithm is modified to maintain its subpopulation diversity. Simulation results demonstrate the improved SFLA enhances convergence velocity and avoids premature convergence effectively, thus improving the efficiency of search for complex functions and validating the feasibility and effectiveness of the SFLA.3. Firstly, the establishment of parameters of HS algorithm is discussed by the use of uniform design; experimental results show the method is feasible. Secondly, for single individual evolutionary HS algorithm, two improved HS algorithms are proposed: dynamic multi-memory HS algorithm and HS algorithm for solving complex functions, experimental results indicate the improved algorithms for complex functions show the strong adaptability, stability, robustness and global search capability.