On-line Learning Algorithm Based On ADP And Its Application Research On The Looper System In A Hot Strip Mill System

This thesis focuses on a systematic treatment for developing a generic on-line learning control system based on reinforcement learning and dynamic programming. Adaptive dynamic programming (ADP) is a promising research field from multiple disciplines such as adaptive control,cognitive science,neuroscience,and psychology. which both can learn on-line and exhibit optimal.Over the past decades,the algorithm of ADP development quickly and walked out the lab,which has been successed application in the military and industry,shows its outstanding optimal control performance.While many of the existing researches focus on multiple inputs single output (MISO) system with steepest descent search.The looper systerm is an important device of the hot strip rolling mill which is a multi-variate complex control system. The looper systerm which is focused on the electrical transmission technology, computer technology, hydraulic servo control system, automatic control technology.So as to improve the study of the plate in the hot strip rolling mill output and quality index is very meaningful.The looper tension and angle control is important in hot strip mills. The looper tension and angle should be kept to a desired value simultaneously, which is called as a nominal operation point. Because they are not independent respectively, they are closely-coupled. The normal research is will live set of highly controlled and live set of tension control as two separate subsystems respectively and seen the couple as disturbance effect is not very ideal.The ADP is a kind of more general and on line learning of control algorithm to solve this problem has the huge potential.Based on the research on ADP,the main parts of my work are as follows:(1) Based on the classic Actor-Critic structure of ADHDP algorithm,improved the weights updated algorithm of ANN,and improved from the only applies MISO systems to MIMO systems, which is more applicable to a wide range of practical real world applications;(2) We study the looper systerm which is an important device of the hot strip rolling mill and establish hydraulic looper control mathematical model and simulation model;(3) We put the improved MIMO ADHDP algorithm applied to the looper height and tension control system, and MATLAB software are used to simulate.ADP on-line learning algorithm is effective to overcome a dynamic programming "dimension disaster" problem, and the actual physical model of "model disaster" problem.According to the multiple-input and multiple-output (MIMO) system of looper system, which is difficult about establishih the control system.We put the ADP on-line learning control model applied to looper system, through the simulation of Matlab software, the results demonstrate that the proposed approach can achieve effective and robust performance,show this algorithm has the strong ability of on line learning.