| Evolutionary algorithm is a random-serching method that imitate biological natural evolution process,Initially, there are three branches: genetic algorithm, evolution planning and evolutionary strategy. On the basis of genetic algorithm ,American scholars Koza proposed a new structure description method—Genetic Programming(GP) in the early 1990s.The main difference between GP and evolutionary algorithm is the expression of chromosome structure ,and chromosome is represented by dynamic tree in GP. GP algorithm has many characteristics ,such as describing problems hierarchically, dynamic coding length.etc.So it has been widely used in artificial intelligence, machine learning, symbol processing, etc.Human-simulated intelligent control(HSIC) theory based on Sensor-motor intelligent Schema(SMIS) shows great advantages in solving diffcult control problems that provides an effective method to solve complicated system problems.But Human-simulated intelligent controller based on SMIS is designed too much depending on expert experience, once the control problem changes, the controller structure need to redesigned, which brought great difficulties for designers. There hasn't a maturer and owning pervasive sense method for automatic designing of Human-simulated intelligent controller structure. While genetic programming is a kind of general optimization method that imitating natural biological evolution,so in this paper we try to solve the problem of automatic designing Human-simulated intelligent controller structure with GP algorithm thoughts.Pendulum is a typical application for the HISC based on SMIS,So through the result of simulation experiments,we proved that the GP can solve the double inverted pendulum.The main research contents of the dissertation are as follows:(1) Improving genetic programming algorithm. In this paper we first introduce the basic genetic programming and various improved methods.Then we mainly describes the improved strategies that we proposed,and gaves the results of it's performance test.(2) Applying the improved genetic programming to solve the structure automation design of double inverted pendulum.Whether the controller objectives can be realized or the results is good or bad is influenced directly by controller structure. In this dissertation we gave a detailed design process and the simulation results, and did comparative experiments, and the results show that the improved genetic programming we proposed has certain advantages.(3) Combining the improved genetic programming with Parallel computing. When applying the improved genetic programming to solve the structure automation designing of double inverted pendulum,it takes so long time for fitness computing,so we combine it with Parallel computing . In this paper we describes the design process of the genetic programming based on MPI in detail, and did experiments. And the results show it's efficiency for improving calculation speed. |
