Research On Particle Swarm Optimization Algorithms And Applications For Some Optimization Problems

There are various types of complex optimization problems in almost every field,such as economy and finance,logistics management,network security,and machine learning.They are characterized by high dimensionality,nonlinear,multi-objective or discrete.Traditional optimization methods cannot solve them at all.Therefore,it is a key task for researchers and engineers to study the efficient algorithms of these problems.Particle swarm optimization is a random search algorithm based on probability,which has strong robustness and global convergence.Because the particle swarm optimization algorithm has few parameters and is easy to implement,it attracted the attention of scholars as soon as it was proposed.At present,it has been successfully applied in many complex optimization problems in practice.This dissertation is focused on different improved particle swarm optimization algorithms for several complex optimization problems,including constrained optimization,mixed-integer programming,and multiobjective optimization.At the same time,two improved particle swarm optimization algorithms are proposed for stock price forecasting problem and multi-objective shop floor scheduling problem.The main research work and achievements are summarized as follows:(1)For solving the nonlinear constrained optimization problem,a particle swarm optimization algorithm based on the improved Deb criterion(FIPSO)is proposed.Based on the Deb criterion,the algorithm retains the information of "excellent" infeasible solutions.It uses this information to guide the algorithm to escape from the local best solution and quickly converge to the global best solution.Additionally,to further improve the global search ability of the algorithm,the DE strategy is used to optimize the personal best position of the particle,which speeds up the convergence speed of the algorithm.The performance of our method was tested on 24 benchmark problems at IEEE CEC2006,and simulation results show that the FIPSO algorithm is effective.(2)For solving the nonlinear mixed-integer programming problem,two improved particle swarm optimization algorithms,the EMPSO algorithm and the CC-PSO/GA algorithm,are proposed.In the EMPSO algorithm,an evolutionary strategy DS for discrete variables is proposed,which effectively solves the application of the particle swarm optimization algorithm in discrete problems.Additionally,an update strategyIDeb based on constraint is proposed,which receives the infeasible solutions as a personal best position by probability and effectively utilizes the useful information contained in infeasible solutions.The CC-PSO/GA algorithm attempts to combine the PSO algorithm and the GA algorithm to solve the nonlinear mixed-integer programming problem.The APSO algorithm is used to deal with the continuous variable and the TGA algorithm is used to deal with the discrete variable.Cooperative cross evolution based on a small population is used to organically combine the two algorithms.Finally,the two algorithms are tested on 14 benchmark problems.The numerical results show that the two algorithms have their own advantages and can effectively solve mixed-integer programming problems.(3)For solving the multi-objective optimization problem,a multi-objective particle swarm optimization algorithm,based on Gaussian mutation and an improved learning strategy,is proposed(MOIPSO).The approach adopts different learning strategies for non-dominated and dominated solutions to guide the particles to search for the global best solution.Additionally,to further improve the uniformity of external archives and current populations,a Gaussian mutation strategy is adopted to drop points at sparse and boundary positions,so as to increase the diversity of the solution.An indicator--DM is presented to measure the distribution width of the non-dominated solution set,which is produced by various algorithms.Combined with the data and figures,it can be seen that the distance width indicator is reasonable.Finally,in order to verify the effectiveness of the MOIPSO algorithm,numerical experiments are carried out on 12 multi-objective optimization test problems.(4)For solving the many-objective optimization problem,a multi-objective particle swarm optimization algorithm based on Tchebycheff decomposition(NMOPSO)is proposed.This algorithm constructs the idea of updating the personal best position with the weight vector,not with the particle.To improve the efficiency of the algorithm and escape from the local best solution,the evolution operation is performed in the external archive.Additionally,a dynamic updating method of the weight vector is proposed to improve the uniformity of the non-dominated solution set.The DTLZ and WFG test suites with 5,10 and 15 objectives are used to assess the performance of NMOPSO.The experiments indicate that NMOPSO has superior performance over six current algorithms for the adopted test problems.(5)For solving the prediction problem of the stock price,a hybrid adaptive PSO and BP neural network algorithm(APSO-BP)is proposed.The APSO-BP algorithmeffectively integrates the global search ability of the PSO algorithm and the local search ability of the BP algorithm and further improves the prediction accuracy.Two sets of real stock data of China's stock market are applied to empirical analysis,and the results show that the algorithm is more effective than the standard BP algorithm in solving this problem and can provide timely risk warning information for investors.(6)For solving the multi-objective flexible job-shop scheduling problem,a multiobjective particle swarm optimization algorithm based on the discrete variable learning strategy(AMOPSO)is proposed.According to the characteristics of the flexible job-shop scheduling problem,the position vectors of particles are composed of operation and machine coding.Additionally,a particle learning strategy based on the operation is constructed,which combines the processing mechanism of discrete variables and the principle of the multi-objective problem.The strategy not only guarantees the effective learning of particles to the best solution,but also guarantees the feasibility of particles.Finally,numerical experiments are carried out on four standard FJSP problems,and the results show that the AMOPSO algorithm can achieve the non-dominated solution with better convergence and distribution.