The new method of flatness pattern recognition based on GA–RBF–ARX and comparative research

Abstract

Radial basis function (RBF) network combines with AutoRegressive eXogenous (ARX) model to create RBF–ARX model. RBF–ARX model can describe the global nonlinear dynamic process of the object, and its function coefficients are approximated by data-driven method. Structured nonlinear parameters optimization method is generally used to optimize model parameters, but this method is very complicated and hard to be mastered by engineers. However, genetic algorithm (GA) is relatively simple and widely used. So thought of GA optimizing RBF–ARX is generated, called GA–ARX–RBF, and applied to nonlinear dynamic flatness recognition model. Flatness pattern recognition is the basis and core of flatness control. Considering disadvantages of traditional pattern recognition method, such as limitation of recognition accuracy and poor fault tolerance, a new intelligent method of flatness pattern recognition based on the GA–RBF–ARX model is proposed. Meanwhile, in order to examine the effectiveness of RBF–ARX, another intelligent flatness recognition model based on dynamic feedback Hopfield NN is set up. Simulation proves that RBF–ARX is more suitable for flatness pattern recognition than Hopfield NN. And optimization effect by GA can achieve the desired requirements and is more precise than structured nonlinear parameters optimization method, when RBF–ARX is applied to rolling mill. GA–RBF–ARX can be a new modeling method for complex nonlinear dynamic system.