Intelligent Adaptive Decoupling Control And Its Application In Integrated Control Of Modernized Strip Flatness And Gauge

Strip Flatness and Gauge are two very important quality control indices in the modern strip production line. The quality control level will directly influence the quality of finished products as well as the sale markets. Therefore, the most important function of automatic control system of strip rolling mill consists of the control of the flatness and the gauge.The flatness and the gauge control, with the interlocking, intercrossing and strong coupling relation, constitute a complex multivariable strong coupling control system. By analyzing coupling relationship and variation law among all the variations in the system, the main factors of affecting the flatness and the gauge are analyzed, then the control model for strip quality integrated control system is established, and finally the three control methods are designed according to the flatness and the gauge control requirements of finishing mill 6 stand for Shan Dong Laiwu Steel 1500mm hot strip.The strip quality integrated control system has several features such as difficult to calculate exactly and to establish the accurate mathematical model, many control variations and strong coupling, etc. According to all these features, the neural network adaptive decoupling control theory is employed to the quality integrated control system, and several neural network decoupling controllers are designed as follows:(1) Through studying on the structural of the multivariable neural network controller, this paper presents a new multivariable tandem connection compensation neural network decoupling control method based on the invariance principle. Through adapting the neural network decoupling compensation control, the couple among the multivariations can be decoupled successfully and the performance indices are optimized under the conditions of maintaining the system intrinsic control function and intrinsic performance.(2) Special analysis and study is made on the standard genetic algorithm (SGA) for its principle, control parameters setting, genetic operation and coding method. Due to the standard GA with some deficiencies such as tendency to get into local optimization, prematurity, low calculation precision and operation efficiency owing to the existence of the too long binary character string in multi-valued parameter optimization process, etc, an improved real-code GA, namely Adaptive Competitive Genetic Algorithm (ACGA), is proposed. Comparing with the standard GA and the adaptive genetic algorithm (AGA), the simulation results demonstrate that ACGA have better performance. The several developed controllers are optimized by ACGA, the simulation results further show that ACGA not only overcomes the deficiencies of standard GA but also have better convergence rate and learning efficiency than standard GA and and AGA.(3) Due to difficult to determine softness factor and the poor adaptability of reference trajectory in prediction control, a new reference trajectory, namely adaptive reference trajectory algorithm, is presented. The prediction control algorithm based on neural network is described, which can be employed to the multivariable nonlinear objects due to the integration between neural network and prediction control. By adjusting the step, optimizing the weight and improving reference trajectory algorithm of Prediction control, the real time and control precision of the control system are improved, and the stability of the system is enhanced.(4) Although the conventional PID control algorithm is easy to achieve and good control performance, the adaptability for the variation of system parameters is weak. Therefore, the ACGA is applied to the online optimization for the parameters of the PID controller, and the simulation results show that the proposed algorithm not only achieves the online self-tune of PID parameters, but also improves the control precision and response speed. Moreover, the validity of proposed algorithm is further demonstrated by the application to the real finishing mill.Finally, the proposed intelligent control methods have been applied to the flatness and the gauge integrated control system of finishing mill 6 stand for Shan Dong Laiwu Steel 1500mm hot strip project, and the results show that the multivariable neural network adaptive decoupling predictive controller has the advantages of better adaptability for parameter variety, stronger interference rejection, better tracing input, and higher calculation and control precision. Furthermore, the controller can be easily design due to not relying on precise mathematic model and simple control algorithm, and can also fulfill strip quality integrated control requirements very well.