| Continuous casting is an important part of the steel industry's production process.In the casting process of continuous casting production,the slag in the large package is easy to flow into the tundish with the molten steel,which is essential for improving the quality of cast steel and extending the service life of equipment.Aiming at the problem of high cost and low recognition accuracy of the current continuous casting ladle slag prediction method,an intelligent prediction method combining deep learning theory was proposed based on the state characteristics of ladle slag,and the research of continuous casting ladle intelligence system was further carried out.The main research work of the slag prediction system was as follows:1)Based on the dynamic relationship between the casting rate and the weight change of the ladle,the dynamic model of the slag prediction system was established,the appropriate data module selected to collect the signal and Visual Basic used to preparing the software platform for predicting the slag under the ladle.2)According to the time series characteristics of the slag process,the combination of LSTM and local weighted regression filtering method was used to predict the experimental results of the slag time.Comparing the LSTM model with the ARIMA and RNN models,the accuracy of the slag prediction using LSTM could reach 95%,which was the highest among the three prediction algorithms.The result shows that the deep learning method has practical value in the prediction of ladle slag.3)Aiming at the problems existing in the prediction model,the LSTM model was simplified and improved according to the characteristics of the data collected in the slag process,and three simplification algorithms was proposed.By comparing the test results of standard LSTM with the three simplified variant models,the simplified variant model not only reduces the demand for sample size,but also improves the prediction rate and the accuracy.Figure28;Table9;Reference 60... |
PDF Full Text Request