| With the development of both industrialization and computer technology, the age of big data comes, and many engineering optimization problems with higher dimensions, stronger coupling and larger data size occurred. Instead of traditional deterministic optimization algorithms, Particle Swarm Optimization (PSO) and Bare Bones PSO (BBPSO) are more powerful algorithms to these new problems. Because of their simple structures, low restrictions to objective problems, low computation consumptions and parallel computing, they have appealed to researchers in various areas. However, there are still some shortcomings for them. Firstly, early maturing problem together with unstable optimization procedure make them far away from perfect robustness; second, they are not global convergence; finally there are little theoretical works about how they function to different objective functions. Several efforts have been achieved to overcome these problems in this thesis and listed as follows:(1) An introduction of optimization problems and methods are given, followed by the research backgrounds of PSO and BBPSO algorithms. The state-of-arts concerning the development, theoretical analysis and application of BBPSO are reviewed. Then the general BBPSO form, which is essential for theoretical analysis, is proposed after the analysis of several BBPSO variants.(2) Objectivity, swarm diversity and optimizing characteristics of BBPSO are discussed. Experimental results indicate that BBPSO-I is a rotational invariant algorithm with poor swarm diversity, while BBPSO-II is rotational variant with better swarm diversity and general performance. The using of Gaussian, Exponential, Cauchy or Uniform distribution makes particles of BBPSO-II tend to move along the axes. Afterward, these features are clarified by theoretical analysis. Based on these results, some advices on the application of BBPSO are given.(3) A new bare bones particle swarm optimization algorithm with particle pruning strategy (NPSO) is proposed to improve both global exploration and local exploitation. Firstly a new evolution equation is proposed. The instant search domain analysis indicates better swarm diversity is obtained by the new equation. Also global convergence of the new evolution equation is analyzed, and how to choose a proper parameter to ensure swarm convergence is discussed. Secondly, a particle pruning strategy is introduced:each time when a particle reaches a new best position of the swarm, it would be pruned and inserted to another position. Not only can pruning strategy improve global exploration, theoretical analysis shows it also improves local exploitation when certain criterion is met. Finally experiments on benchmark problems show NPSO obtains significant improvement when compared to some classical PSOs.(4) The application of NPSO to power transformer design optimization (TDO) problem is presented. The design procedure of TDO is introduced, and the mathematical model, solution space and constrains of TDO are discussed. NPSO together with a new multi-constrains handling method is used to solve TDO. Meanwhile, exhaustion method and heuristic method which are suitable for low complex TDO problems, is proposed. They are compared with NPSO on high complex TDO problems, showing that NPSO obtains excellent performance.(5) The application of NPSO to transformer tank design is presented. Finite element method is employed to analyze the tank deformation distribution. A novel transformer tank finite element model, employing3-D springs to simulate the behavior of joint deformation, is proposed to improve the computation precision. Experiments are implemented to get the deformations of three tanks. NPSO is used to fit the spring stiffness in the tank FE model. The comparisons of calculated and experimental results show the proposed method’s effectiveness. Finally the deformation and stress distributions of two typical tanks are calculated with the proposed method, and some engineering suggestions are given. |
PDF Full Text Request