Several Classes Of Improved Particle Swarm Optimization

Particle swarm optimization (PSO) is a population-based, self-adaptive search optimization technique first introduced by J.kennedy and R.C.Eberhart in 1995. PSO is simple in concept, few in parameters and easy in implementation, so it has attracted a lot of attention from researchers around the world and found applications in many areas since its introduction.Several classes of improved PSO are proposed in this paper. The main works of the dissertation can be summarized as follows:In Chapter Three, firstly, two new position update equations are proposed by using the strategy of extrapolation in Mathematics. Thus, a new class of PSO with induction-enhanced is given. Secondly, a modified PSO algorithm with flying time adaptive adjusted is proposed. The flying time of every particle in this algorithm is adaptive adjusted in pace with addition of the evolutionary generations; Thus, the algorithm overcomes the difficulty of the traditional PSO that the particle's ability of searching is decreasing during the last time of iteration, which is caused by the flying time of every particle is fixed on one. Thirdly, a class of dynamic-population PSO for searching peaks of some multi-peak functions is proposed. This algorithm transforms all peaks of multi-peak problems into these whose peaks are equally high by functional transformation, which in order to find all peaks in the same probability. The size of particle swarm can be altered during the searching, so the problem that have to determine the size of swarm because we can not obtain the numbers of peaks of the given multi-peak function in standard PSO is resolved. The experiments manifest that these algorithms are very effective.In Chapter Four, firstly, a new class of no monotone trust region algorithms based on conic model for unconstrained optimization is proposed. Secondly, a hybrid searching method consists of trust region algorithm with global convergence and PSO with random searching in local field is established. A class of trust region algorithms based on PSO is proposed. Global convergence is proved under certain conditions.In Chapter Five, PSO with random searching is used to solve nonlinear equation and system problems. A large number of numerical experiments indicate PSO is a very effective algorithm to these problems.