Floating Height Prediction And State Identification Of Strip In Air Cushion Furnace With Machine Learning

The floating height of the strip in an air cushion furnace is a key parameter for the quality and efficiency of production.However,the high temperature and high pressure of the working environment prevents the floating height from being directly measured.In recent years,the development of machine learning provides new approach.Furthermore,the strip has multiple floating states in the whole operation process,so it is difficult to accurately describe the floating height under different states by a single model.In addition,there is an inevitable transition state between the stable state and the vibration state in the working process of the air cushion furnace.The duration of transition state is short,but it has a large impact on the quality of the product.Therefore,the study of state identification and height prediction of transition state has vital theoretical significance and economic value.With the help of the machine learning technique,the paper studied the floating state and the floating height of strip in the air cushion furnace,especially the identification and modeling of the transition state.In order to research the prediction of floating height,a semi-supervised state recognition algorithm based on improved data gravitation classification is proposed.Firstly,the method takes advantage of semi-supervised thought to expand the labeled data set by self-training.Secondly,the maximum escape velocity is used to delimit the gravitational field to improve the classification performance.Then,considering local data features,the transition state is identified according to dynamic process characteristics on the basis of classification.Considering the time series and noise interference of data in industry,a state identification method combined adaptive k-nearest neighbors and principal component analysis is proposed.An adaptive k-nearest neighbor is used to search nearest neighbor data unit within a fixed search range.And the floating state is divided and identified by control limit of SPE on time series.The identification of floating state is helpful to the modeling and prediction of floating height of strip.Based on the identified results,the prediction models of strip floating height are established,involving hybrid model for the stable state,double-random forest model for the vibration state and soft-transition model for transition state.In the hybrid model for the stable state,the mechanistic model combined thick jet theory and the equilibrium equation of force to cope with the lower floating height.In particular,in view of crossing and gradually changing dynamic characteristics of transition state,a novel soft-transition floating height prediction method is proposed.The membership degree is developed to dynamically describe the relationship between transition data and neighbor states,which further reflects the internal process characteristics of transition states.