Research On Coevolutionary-Based Hybrid Intelligent Optimization Algorithms And Their Applications

The optimization problem has always been recognized as one of the most difficult but important problems. Inspired by the natural phenomena, social phenomena or biological intelligence, computational intelligence provides simple, versatile, robust and parallel methods that can effectively solve most of the optimization problems. In response to an increasing number of complex optimization problems in morden society, traditional intelligent optimization methods have many limitations, we need know more about hybrid intelligent optimization in hybrid mechanism and strategies, and there is no thorough study of inherent relationship and inner machanism in combining intelligent optimization algorithms. As a new algorithm introducing co-evolution into traditional computational intelligence, co-evolution provides an abstraction algorithmic model and can be flexibly constructed according to the real problems to be solved. Because co-evolutionary algorithm can effectively overcome the problems of premature phenomenon and low optimization precision etc., hybrid intelligent optimization algorithms based co-evolution has recently become a hot research topic in the field of artificial intelligence.In introducing the relative knowledge of computational intelligence methods, this thesis mainly analyzes the state-of-art of the co-evolutionary algorithm and hybrid intelligent optimization algorithm, and recent applications of the hybrid intelligent optimization algorithm to the Traveling Salesman Problem (TSP), the railway passenger volume forecasting and fault diagnosis. Focused on the existing problems, this thesis has achieved the following research work:(1) In reflection of the co-evolutionary strategies and the characteristics of existing intelligent optimization algorithms, a parallel co-evolutionary (PCEGP) algorithm is proposed by introducing the co-evolutionary mode and parallel evolution mechanism into genetic algorithm and particle swarm optimization. The PCEGP algorithm divides the individuals into two equal-sized groups according to their fitness values. The subgroup of the top fitness values is evolved by GA and the other subgroup is evolved by the PSO algorithm. The optimal solution is found out in whole group. The theoretical analysis proves that the PCEGP algorithm can100percent converge to the global optimal solution under conditions of the bounded closure and continuousness.(2) To optimize the structure and parameters of RBF neural network, a parallel hybrid intelligence optimization (PHIO) algorithm based on PCEGP algorithm and RBFNN is proposed. By designing a switching function, the RBFNN optimization is translated into a simple function optimization problem, the PCEGP algorithm is then used to find the global optimal solution and thus PHIO algorithm with high performance is constructed. The experiments with the given function optimization show the PHIO algorithm has the characteristics of the quick convergence, strong global search ability, good stability and high solving accuracy.(3) Based on phased hybrid intelligence, a two-stage hybrid intelligent optimization (TSHIO) algorithm based on PCEGP and ACO is proposed. The whole process of the TSHIO algorithm is divided into the rough searching and the detailed searching. The TSHIO algorithm is better than ACO in time efficiency and PCEGP in the refining efficiency. Various scale TSPs are tested to validate the effectiveness of the TSHIO algorithm, and the simulation results indicate that the TSHIO algorithm has better convergence, higher accuracy and stronger global search ability.(4) On the basis of hybrid intelligent optimization strategy, a hybrid intelligent optimization (RSBPNN) algorithm based on rough set and BP neural network is proposed, and a new railway passenger volume forecast method based on RSBPNN algorithm is presented. The RSBPNN algorithm uses the knowledge reduction ability of rough set to deal with and reduce the staple data in order to determine input layer variables and the number of neurons of BPNN. Then arbitrary function approximation and learning ability of BPNN is used to construct the railway passenger volume forecast method. The experiments show that the forecast results are closer to the real statistical values.(5) In order to form the fusion and complementary of the intelligent optimization algorithms, a new intelligent fault diagnosis (RGBNFD) method based on combining hybrid intelligent optimization strategy, rough set, genetic algorithm and neural network is proposed. The RGBNFD method takes full advantage of the knowledge reduction ability of rough set, the maintaining population diversity of genetic algorithm and the classification ability of BP neural network. In order to verify the effectiveness of the proposed RGBNFD method, the RGBNFD method is used to diagnose the motor rolling bearing fault. The results show that the RGBNFD method can not only effectively solve the fault diagnosis, but also obtain high accuracy rate, and take on the certain fault tolerance ability.