The Research Of Particle Swarm Optimization Algorithms Based On Surrogate Model

Whether in the field of engineering or in our daily life, optimization problems widely exist. Over the years, people put forward many methods related to optimization problems, such as differential extremum method, conjugate gradient method, quasi-Newton method and so on. These methods all belong to the traditional classic optimization methods, and they are local optimization method. The common characteristics of these methods are only as local optimization method single peak, low-dimension simple function to achieve better optimization results. But to the practical problems, they are usually high-dimensional, multi-extreme, non-linear, so the traditional method is difficult to effective.Particle swarm optimization is a class of complex problems optimization(solving) method. Different from the conventional method, the method is based on imitation bird in the nature, they search for each individual, then repeated information process between individuals, adaptive guide each individual to the group moving in the direction of the optimal solution, and finally to search a better local optimal solution. Because the algorithm has good robustness and strong ability of global optimization, so it has achieved very good results in the field of that is difficult to solve in many traditional methods, such as biological feature extraction, data mining and large-scale combinatorial optimization problem, has been widely researched and applied in the field of science and engineering.Although the scholars have the study of particle swarm optimization algorithm repeatedly, but the algorithm in the aspects of operational mechanism and convergence theory still needs to be further improved. Especially for the evaluation of individuals(fitness), still no unified evaluation criterion, only i s the concrete analysis for concrete problem. Therefore algorithm often rely on specific issues and the experience knowledge, make the wide application of the algorithm is limited by a certain. On the other hand, in many engineering applications, because of the cost of fitness calculation is too big, so make the algorithm of calculation efficiency is low, it also hindered the widely used of the algorithm.In this paper the structure characteristics of particle swarm optimization(PSO) algorithm and the search process has been carried on the thorough analysis, and we also analy zed the distribution characteristics of fitness function and the advantages and disadvantages of existing fitness model according to the related theories and methods of machine learning. This paper process a cannot only reduce the computational cost fitness, and can guarantee the fitness of the model calculation accuracy and the stability of the algorithm based on existing fitness model—Homeomorphism manifold agent model based on the feedback mechanism, and analyzed the theoretical basis of the algorithm and the complexity of the algorithm. The agent model used the Riemannian manifold learning method, polynomial regression method, and the method based on feedback mechanism. Finally, we tested our algorithm on the set of benchmark functions, the standard particle swarm algorithm and two other different agent model s. Results show that our proposed algorithm can get very good results. Fully demonstrated the reliability and effectiveness of our proposed algorithm. In experiment and outlook section, described the particle swarm behavior should pay attention to the aspects of the analysis, and put forward the opinions of the agent model need to be further improved.