Particle swarm optimization ( PSO ) in 1995 by Dr. Eberhart and Dr. Kennedy raised together, it comes from the study of birds of prey behavior. The basic core is the use of a population of individuals sharing of information so that the movement of the entire population to produce evolution from disorder to order in problem solving space to obtain the optimal solution. In PSO in all particles having a position vector (a position in the solution space of the particle) and velocity vector (the direction and velocity determines the next flight), and the fitness value may be calculated current location where the target function. At each iteration, the population of particles according to their flying experience each particle and other particles flying experience in the search space, dynamically change the speed of each particle, to adjust the trajectory of each individual in the search space , thereby how to adjust and change the direction and speed of flight to determine when the next iteration. So gradually iteration, the final particle entire population will gradually become the optimal solution.
For PSO algorithm, two inertia weight and acceleration constants factors directly affect the value of the superiority of the algorithm. Asanga Ratnaweera et al proposed extension of PSO algorithm, by experiments with variable inertia method when weighting factor ( PSO TVIW- random method inertia weight factor () PSO-RANDIW) The advantages and disadvantages of different methods.
The first is the introduction of time-varying acceleration coefficient ( TVAC) , by changing the acceleration factor and over time reduces the cognitive component, increasing social component : at the beginning, the cognitive component of the larger and smaller social composition, thus allowing the particles to move in the search space, but not the best move towards the population , while the smaller and larger social cognitive component composition allows optimization of the particle in the late converge to global optimum. This method is called PSO-TVAC method , can be optimized to enhance the early stages of a global search, and to encourage convergence to the global optimum particle at the end of the search.
The second is having a "mutation" and time-varying acceleration coefficients particle swarm optimization device ( MPSO-TVAC) , the "mutation" into PSO 's strategy ( MPSO), by providing additional populations to enhance the global search ability of particle diversity . Under this new strategy, when the global optimal solution is not improved with increasing generation, will randomly select a particle, then a random perturbation (variation step) was added to a random selection mode velocity vector of the particles of predefined probability (mutation probability) by.
The third is having a self-organized layered PSO acceleration time-varying coefficient filter ( HPSO-TVAC) , which introduces a novel concept of "self-organized layered particle swarm optimization device (HPSO)", to provide particles needed momentum to find the global optimal solution, but there is no PSO algorithm previously speed entry . In this method, the speed of the previous item remains zero, and the random velocity (velocity reinitialization) reinitialized stagnation speed vector particles during molding in a search space , in which method, according to the behavior of the particles in the search space automatically generating a series of PSO in the internal main particle swarm optimization, until a convergence criteria is satisfied so far.
By experiment, five reference to evaluate the performance results show that the compared PSO-TVIW, PSO-TVAC significantly improved convergence rate and the optimum value, especially for the unimodal function , however, the performance of the method of PSO-TVAC in multimodal poor function ; the MPSO TVAC on most-selected reference was observed a significant improvement in performance ; with the compared PSO-RANDIW method, PSO-TVAC performance multi-mode functionality has been improved , however, for unimodal function, which converges slowly; using MPSO-TVAC policy Rastrigrin function weak, other similar functions, but slow convergence for most reference, has a fixed acceleration factor (MPSO-FAC) 2 the method of MPSO the performance is very poor . And for most of the reference, HPSO-TVAC compared with PSO-TVIW and PSO-RANDIW method , performance has significantly improved , however, HPSO Schaffer-TVAC method on the performance and function f6 Rosenbrock function in small dimensionsThe relatively poor performance. A fixed acceleration coefficient ( at c1 = c2 = 2) case, the performance of the method HPSO very poor. Further, all methods of , PSO-RANDIW method shows a significantly faster convergence rate in the early stage of the optimization process.