Research On Three New Particle Swarm Optimization Algorithms And Their Applications

Complex optimization problems usually have the characteristics of high dimension,nonlinear,multi-objective or discrete,so it is difficult to solve them by traditional optimization methods.Particle swarm optimization(PSO)has been widely used in solving complex optimization problems due to its robust robustness and global optimization ability,but it is prone to premature convergence and low convergence accuracy.In this paper,three kinds of new particle swarm optimization algorithms are constructed,and they are applied to path planning problems,travel salesman problems and feedforward neural network image compression problems.Specific strategies are as follows:1.Simplified particle swarm optimization algorithm(HSPSOBOA)based on hybrid butterfly optimization algorithm is proposed for robot path planning.The butterfly optimization algorithm is added to the simplified particle swarm optimization algorithm to improve the global optimization ability.The local and global optimization capabilities of nonlinear learning factor adjustment strategy balancing algorithm based on similarity are proposed.The mean updating rate is introduced to expand the particle search interval.2.An adaptive particle swarm optimization algorithm(SAPSO)based on S-type function is proposed to solve the traveling salesman problem.It uses reverse learning strategy to deal with initial population to maintain population diversity.Using the distance from particle to elite,the S-type function adaptively adjusts the local and global optimization ability of the inertial weight balancing algorithm.Perturbation of elite particles helps particles jump out of local extreme values.3.A dual adaptive particle swarm optimization algorithm(SDAPSO)based on selection strategy is proposed to solve the image compression problem of feedforward neural networks.The optimization and acceleration of the algorithm based on probabilistic adaptive variation are carried out for the selected population.HSPSOBOA,SAPSO and SDAPSO are used to numerically test 10 test functions given by CEC2005,and the three algorithms are applied to solve the robot path planning problem,the traveling salesman problem and the feedforward neural network image compression problem respectively.The experimental results show that HSPSOBOA,SAPSO and SDAPSO algorithms have obvious advantages over other algorithms in terms of convergence accuracy,stability and search performance.