The Quality Model Research Of Hot-rolling Plate Based On High-dimensional Wavelet Neural Network

The modern steel market has increasingly high demand for thequality of hot-rolling plate and traditional quality model cannotmeet the requirement of production. The production of hot-rollingplate has multi-channel processes and there are many factors thataffect the quality of the plate. They have a complex nonlinearrelationship between the factors and the quality of plate. In order tomeet the needs of enterprises for quality requirements, artificialintelligence(AI) technology is used to build a quality model of thehot-rolling plate, which can control the quality of hot-rolling plate.For this the following work has been done:Firstly, the basic structure and learning rules of artificial neuralnetwork (ANN) were analyzed. Network structure and the learningprocess of BP neural network-a typical of ANN were discussed.Relevant theoretical knowledge of wavelet neural network、hot-rolling process of plate and quality requirements were studied.Secondly, the quality model of hot-rolling plate based onlarge-dimensional wavelet neural network (WNN) was structured.Through analyzing the production sample data of hot-rolling plate，the key input and output variables of hot-rolling plate quality model were determined; The quality model algorithm of hot-rolling platebased on WNN was studied with the momentum factor; through theexperiment, it is determined that the number of hidden nodes inWNN is 60, and the Morlet wavelet function was used for hiddenexcitation function. The simulation results showed that the fittinghit rate of the quality model based on WNN reached to 90.1%, andthe testing hit rate was 81.5%.Finally, the quality model of hot-rolling plate based onlarge-dimensional WNN was improved. 1) LVQ neural network wasused to cluster for sample data, which can reduce the timecomplexity of the algorithm, and improve the convergence speed ofnetwork. 2) The structure of WNN with double input layer was used.According to the production process of Hot-rolling plate, therefining input parameters were placed in the first input layer ofnetwork, and rolling input parameters were placed in the secondlayer. The simulation results showed that fitting hit rate of theimproved model was 92.3%, and testing hit rate was 84.5%, whichcan meet the basic need of enterprise production.