Improved Particle Swarm Optimization

Nonlinear optimization is a cross subject of computational mathematics and operation research. It is not only widely used in national defense, economic, engineering and management, but also many problems in other scientific fields can be come to it, such as pattern recognition problems in information science and protein folding problems in life science etc. Since the problems in engineering field are often large and highly nonlinear, they are difficult to solve by traditional algorithms. In addition, most of traditional methods need gradient information, thus they are not suitable for nonsmooth optimization problems. Different from traditional methods, swarm intelligence algorithms are less depend on smoothness of the objective function and don’t need gradient information. So they have very strong robustness.As a kind of swarm intelligence algorithm, particle swarm optimization algorithm has widely applied in function optimization, neural network training, nonlinear system identification and other fields. Because it is simple, less parameters are required to adjust and they can be adaptive control etc, But the algorithm is derived from biological phenomenon, it suffer from the drawbacks of premature convergence, slow convergence speed in the late time. To overcome these defects, particle swarm optimization algorithm is studied further and two kinds of improved algorithms are proposed in this dissertation.First, an improved algorithm is designed by introducing the mutation operator to overcome particles’single diversity in the late time. The new algorithm changes update way of particles’position, thus improve the diversity of population.Second, a new hybrid algorithm is designed by introducing differential evolution algorithm to overcome slow convergence speed of particle swarm optimization algorithm in the late time. The new algorithm keeps the common advantages of the two algorithms.Finally, several benchmark functions are tested and the experimental results show that the proposed algorithms can overcome premature convergence of particle swarm optimization algorithm and greatly improve the convergence precision of the algorithm.