Research On The Method Of Hot Rolling Gauge Control Based On The Improved Neural Network

The output and quality of steel are very important norm to judge the level ofindustrialization and modernization in a country. Iron and steel industry is the foundationalindustry for modern development. Nowadays, with the improvement of production’s level,many industry, such as motor industry, aircraft industry and real estate, have improved theirrequirement for strip steel’s quality. The gauge accuracy of strip steel is the most importantnorm to judge the quality of strip steel. So, how to control the gauge and improve the gaugeaccuracy becomes the most important task for metallurgy industry all over the world. The AGCsystem, shortened for automatic gauge control system, is a control system has some featuressuch as multivariable, strong coupling and nonlinearity. Now most AGC system takes theadvantages of artificial intelligence algorithm to realize the high-level control and improve theflexibility of control system.Nowadays, the hot rolling gauge control system is the most extensive used control systemin strip steel production. This paper introduces the manufacturing technique and process of hotrolling; the basic theory of gauge control system, such as the mathematical model of gaugecontrol and the reasons caused to gauge’s change and so on; the principles of different AGCsystems and set up a mathematical model for hydraulic pressure AGC system. Through theanalysis of the main components of hydraulic AGC controller, servo amplifier, servo valve,hydraulic cylinder and sensor, and the mathematical model is built for each part, establishedmathematical model of hot rolled AGC, the research focuses on adaptive controller, so thereduction of the mathematical model of complex, the five order model of complex thesimplified mathematical model of two order.This paper pays more attention to the improvement of neural network and its application inhot rolling gauge control. Neural network has some advantages, such as high-level nonlinearityinformation processing capability and self-learning capability. But sometimes it also has somedisadvantages, such as slow convergence speed and easily getting in local minimum. So how toovercome these disadvantages and improve it is our focal point. This paper aims at thesedisadvantages, optimized and improved neural network by using a specific method----addingmomentum item. This paper also studies some aspects, such as how to choose the structure of neural network, how to choose the nonlinearity error function and how to choose the factor ofmomentum and so on. As a result, I determined a BP neural network structure with good qualityof tracking performance, and proved its effectiveness by simulation experiment. Finally, thispaper designs a PID controller with improved BP neural network and applies it to hot rollingcontrol system. The result of simulation experiment proved its good quality in trackingperformance and information processing.