Many scientific, engineering and economic problems need the optimization of a set of parameters with the aim of minimizing or maximizing the objective function. The traditional sequential optimization methods such as Newton method, Rosenbrock method, Powell method etc., are often using a localsearch algorithm that iteratively refines the solution of the problem. Unfortunately, with the extension of human activities, these traditional methods exhibited their weakness to deal with complex problems. They often fail in working upon many realworld problems that usually have a large search space, multi local optimum, and even are not welldefined. Therefore, effective optimization methods have become one of the main objectives for scientific researchers.Modern optimization methods such as artificial neural network, tabu search, genetic alorithm and ant colony algorithm etc., have shown capabilities of finding optimal solutions to many realword complex problems within a reasonable amount of time. These methods have forged close ties with neural science, altifical intelligence, statistical mechanics, and biology evolution etc., some of them are called intelligent optimization algorithms. Recently, particle swarm optimization (PSO) algorithm has been gradually attracted more attention over another intelligent algorithm. PSO is simple in concept, few in parameters, and easy in implementation. However, both theory and application of PSO are still far from marure. Several aspects of PSO such as basic structure, characters, improvement and realization are systematically discussed in our work. The main contributions given in this dissertation are as follows:1. The convergence of PSO was investigated in details for understanding clearly the internal mechanisms. The convergence conditions were derived based on the simplified version of PSO. Furthermore, the original and standard PSO (SPSO) were analyzed by adopting these conditions. The tracks of particles in PSO were simulated in the specific initial conditions.2. There are a few control parameters used in executing PSO algorithm, the effects of these parameters on PSO performance was systematically studied, and the guideline of better choosing these parameters was summarized. In additipn, a composite particle swarm optimization algorithm, which determines the controlparameter automatically, was proposed, and the new algorithm has been applied successfully to the parameter estimation of heavy oil thermal cracking model.3. To overcome the weakness of premature regarding the PSO used in multimodal problems, a new adaptive particle swarm optimization (APSO) algorithm was proposed by employing the negative feedback strategy. On the one hand, the exploration and exploitation ability of the algorithm were regulated through introducing two criteria in the evolutionary process, i.e. the populationdistributionentropy and the averagedistanceamongstpoints, to preserve population diversity. On the other hand, dynamic inertia weight varied with population diversity was employed to improve the convergence speed. Preliminary experiments upon multimodal function showed that, in comparing with standard PSO, APSO is more efficient to solve the premature problem with higher global successive rate and accuracy, and is also good in convergence speed. Finally, the new algorithm has been applied to the XOR classification problem, and satisfactory results were obtained.4. Due to the lower local search ability and the lack of higher diversity of particles in PSO, an hybrid particle swarm optimization (HPSO) was proposed, where the HookeJeeves pattern search is combined to standard PSO to speed up the local search, also mutation operation is embedded to avoid the common defect of premature convergence. The performance of HPSO was demonstrated through extensive benchmark functions and compared with those by the PSO. Finally, a detailed application of the new algorithm to optimize the operating conditions based on the model of artificial neural network and support vector machine w...
