Hot continuous rolling strip steel head width prediction method fusing rolling mechanism and deep learning

Abstract

The invention provides a hot continuous rolling strip steel head width prediction method fusing a rolling mechanism and deep learning. The method comprises the following steps of firstly obtaining production data of a hot continuous rolling site, and using a Pauta criterion to remove outlier data, and obtaining sample data; and screening influence factor data according to influence factors of rolling broadening, then constructing a rolling mechanism prediction model of each machine rack, calculating a prediction reference value of the hot continuous rolling strip steel head width according to the influence factor data, and constructing a deep belief neural network model to predict a correction value of the strip steel head width, and finally, adding the predicted reference value of the strip steel head width to the predicted correction value to obtain a final predicted value of the measurement position of the strip steel head width at an outlet. According to the method, the rolling mechanism and the deep belief neural network are fused to predict the strip steel head width, the problems that a prediction model based on a traditional single hidden layer neural network has low prediction precision and is easy to fall into a local extreme value are solved, and a good foundation is provided for optimization of a process automation level setting model.

Description

technical field [0001] The invention relates to the technical field of automatic control of steel rolling, in particular to a method for predicting the head width of a hot continuous rolling strip that combines rolling mechanism and deep learning. Background technique [0002] Width accuracy is one of the most important dimensional indicators in strip production. Although most of the width control methods of the hot continuous rolling production line are concentrated in the rough rolling area, the setting of the width model of the rough rolling area is affected by the change of the width of the finishing rolling area. When adjusting the setting model parameters of the process automation level, if the head width after finishing rolling can be accurately predicted, it can guide the adjustment of model parameters and provide a basis for correcting the width setting model. [0003] The prediction of the width parameter is very complicated due to the non-linear interaction and d...

AI Technical Summary

Problems solved by technology

[0003] Prediction of the width parameter is very complicated due to the nonlinear interaction, dynamic coupling process in the steel production process

Although the width prediction model established according to the rolling mechanism conforms to the general rolling law, it will inevitably be simplified and approximated in the derivation process, and the theory-guided modeling idea will ignore many field factors, which is different from the actual production conditions. Therefore, the error in predicting the width by the rolling mechanism prediction model alone is relatively large, which cannot meet the increasingly accurate rolling requirements

[0004] With the development of intelligent technology, some width prediction methods based on rolling data and neural network have appeared in recent years. Although the accuracy of these methods has been improved, due to the black box characteristics of neural network, the prediction model based on neural network The width is poorly interpretable and reliable, and the prediction accuracy of the ordinary single hidden layer neural network structure is still lacking, the generalization ability is not strong and it is easy to fall into local extremum

Images