Improvements In Research, Based On Local Search Of The Inverted Oscillator Fuzzy Controller

Artificial Neural Network (ANN) is a significant new subject in the field of information science and technology. It is a system based on the study and simulation about the architecture and function mechanism of human brain. In the recent years, fuzzy neural network has become a new research hotspot of machine and aviation control field because of its special advantages. Now applications based on the fuzzy logic control increase quickly and the researches such as city subway system, nuclear reaction control system and automobile driving system go perfect step by step. Scientists are doing further research in the design and realization of fuzzy control network.In the neural network control area, the control of inverted pendulum is always a typical hot but difficult problem for its representative and complexity. Therefore, every new idea or algorithm of neural network inclined to check its validity by solving the inverted pendulum problem. Many methods have been presented to solve this problem such as Anderson' s AHC (Adaptive Heuristic Critic) method, Peng J' s dynamic programming method, BP network algorithm, traditional local search algorithm, Q-learning algorithm, intelligent control rules algorithm and Professor Li' s cloud model control algorithm etc. However, there are some disadvantages in those methods or algorithms. For example, the AHC method can not solve the local minimum problem effectively. The dynamic programming method is very slowly and you even can not bear with such a long time in the actual applications. The traditional local search algorithm can not solve the initial big angle problem successfully.In this paper, we borrowed the useful experiences of control models and algorithms from former researchers and at the same time we overcome some disadvantages of old methods. We proposed an improved network control model and algorithm to solve the inverted pendulum problem. Our main innovativework lies in two aspects. First, we reconstruct the fuzzy control rules based on the ULR unit. We changed the old model of two fuzzy states and four control rules and proposed a new model of four fuzzy states and ten control rules. It makes the control process more accurate, sensitive and more quickly. Second, we do some improved research on the control algorithm. We proposed an improved local search algorithm to solve the problem of trapping into local minimum in the traditional local search algorithm. The new algorithm solves the local minimum problem successfully by changing the fixed step to alterable step during the learning process. It can make the learning process more quickly and increase the successful learning rate.This paper will introduce the related neural network knowledge about the inverted pendulum problem firstly and analyze the international research situation and advantages or disadvantages of some methods about inverted pendulum control problem. We will explain how we construct the four fuzzy states and ten control rules network model in detail and at the same time we will give the mathematical theory of how we can decide the control parameters of all control layers. After that, we will illustrate the shortcomings of local search algorithm by contrastive data. And then we will pay more attention in the main idea of our new algorithm. Finally, we will verify the validity and efficiency of our method through several groups of data.