Improved Particle Swarm Optimization Algorithm And Its Application

Particle Swarm Optimization(PSO)is a kind of swarm based-intelligent optimization algorithm inspired by foraging behaviors of birds.Because of its self and global optimal learning mechanism,it has the advantages of fast convergence speed,good accuracy and easy implementation.But there are also some shortcomings,such as: premature convergence,easy to lose population diversity and fall into local optimum,poor performance on complex optimization problems and so on.In order to overcome these shortcomings,this dissertation has done the following work:(1)In order to reduce the premature convergence and the shortcoming of easily falling into local optimum,as well as improve the adaptability of PSO,an improved PSO algorithm based on the inertia weight of population diversity,time-varying learning factors and a new velocity updating formula is proposed,which is abbreviated as PSODM.In the PSODM,the population diversity is integrated into the inertia weight,which can adaptively balance the global exploration and local exploitation abilities.Time-varying cognitive and social learning factors can dynamically adjust the amplitude of each learning item during the whole search process.The new learning mechanism based on the personal bests enriches the learning strategy of the algorithm.The reinitialization strategy based on the number of stagnation effectively improves the population diversity and reduces the risk of falling into local optimum.Experimental results and statistical analysis verify the effectiveness and competitiveness of the PSODM algorithm.(2)In order to strengthen PSO performance and balance exploration and exploitation better,an exponential distribution-based mainstream learning PSO with chaotic inertia weight and disturbance called EMPSOCD is proposed.In EMPSOCD,a mainstream learning component is incorporated into the velocity-updating equation to increase population diversity by changing particles' flying direction.Secondly,in order to reduce this sensitivity,the acceleration coefficients and uniformly distributed random number are replaced by exponential distribution random number,i.e.,the acceleration coefficients have been discarded from the velocity-updating equation.In addition,the general linearly decreasing strategy is replaced by a chaotic map strategy,which can further balance the exploration and exploitation process.Finally,a disturbance technique based on chaotic map and sine function is adopted during search process,and a greedy selection mechanism is employed to keep elites.To verify the feasibility and validity of the proposed EMPSOCD algorithm,35 well-known benchmark functions are employed to test EMPSOCD convergence performance.EMPSOCD is compared with the other 17 excellent PSO variants and evolutionary algorithms developed recently.Experimental results and corresponding statistical analysis indicate that EMPSOCD outperforms other 17 algorithms for majority benchmark functions.In summary,the proposed EMPSOCD is successful,it has a competitive convergence performance.