The Study Of Computational Intelligence Problems

Learning from life phenomenon and the human being's intelligence activity, People create computational intelligence methods to solve some complicated problems. There are three typical embranchments in the study of computational intelligence, such as artificial neural networks, evolutionary computation and fuzzy logic. This paper gives a comprehensive study on some theory and practice problems in the fields of artificial neural networks, evolutionary computation. The main contents are as following:1. The feature, theory, and methods of the modern artificial neural networks–neural fields theory is introduced and summarized.2. An easy-to-test criteria for global exponential robust stability of a class of reaction-diffusion uncertain neural networks with time-varying is established by the means of creating new Lyapunov-Krasovskii functional and linear matrix inequality (LMI).3. A new hybrid genetic algorithm is proposed. It applies the strategies such as chaos series to produce initial population, multi-offspring competition, and adaptive mutation to improve the genetic operation. The hybrid algorithm can generate new individuals by the methods such as the elite reservation, quadratic interpolation and improved genetic operator, so that it can overcome the shortage of premature and slow convergence speed of the standard genetic algorithm.4. A synthesize fitness function is proposed in the process of the training of artificial neural networks by using hybrid genetic algorithm in order to improve its learning ability.5. A new adaptive particle swarm optimization algorithm which based on the measurements of population diversity is presented. Two measurements are proposed to indicate the swarm population diversity. The algorithm applies a special mutation operator to increase the swarm population diversity. New velocity term and dynamic inertia weight are also provided in the adaptive particle swarm optimization algorithm to balance the exploration and exploitation of the global optimization.6. An improved particle swarm optimizer is applied in some constrained optimization problems by using the dynamic punishment functions method and the tag punishment functions methods to deal with the complicated constrain conditions.