| Sintering production is one of the important procedures in the iron and steel metallurgy process.Whether the sintering production process is stable directly affects the sinter quality and the normal progress of subsequent production processes.The sintering production process is a dynamic system with long processes,many influencing factors,and complicated mechanisms.It is difficult to achieve accurate prediction and optimal control of the sintering process using traditional mechanism models and classic control theory.By deeply analyzing the characteristics of sintering process,the big data technology was used to establish a sintering process parameter prediction and optimization system with functions of sintering end point state prediction,sintering ore chemical composition and quality index prediction,and sintering batching adjustment and optimization.The main research results obtained are as follows:According to the actual situation of sintering production,determine the sintering end point,sinter ore chemical composition and physical properties as the system’s prediction targets.Aiming at the problems of outliers,missing values,large dimensional differences,and inconsistent frequencies in the original data,a set of relevant methods for data extraction,cleaning,standardization,and integration suitable for the sintering process of steel was proposed.It is confirmed by visualization technology that the end point state is significantly related to the yield and quality index of sinter.Aiming at the problem of poor accuracy and lagging of sintering end point detection,the accurate sintering end state was obtained by improving the bellows exhaust gas temperature method.Based on this,a sintering end point prediction model was established using the gradient boosting tree algorithm and decision rules.The test results prove that the end point position prediction model shortens the end point position judgment error from 3.64 m to 1.25 m,and the model prediction hit rate can reach 85.6%,and the goodness of fit(R~2)of the end point temperature prediction model can reach0.809.The sintering end point state prediction model was put into use on the on-site 360m~2sintering machine,which improved the sintering end point stability and sinter quality index.Aiming at the characteristics of sintering production data,such as noise,high dimensional and time correlation,the box chart method and the isolated forest algorithm were used to detect and filter the noise,and the key feature selection method and the Pearson correlation coefficient method were used to solve the problem of high-dimensional parameters.On this basis,an online component monitoring model based on deep neural networks and an advanced component prediction model based on long short term memory network were established.Through a large number of experiments,it is shown that compared with RF,MLP,ARIMA and SVR algorithms,the model built by deep learning method has better performance,the R~2of the better model reaches more than 0.92,and the mean square error and mean absolute error approach zero.Aiming at the problem of detection lagging of sinter quality,a comprehensive prediction model of sinter quality was proposed.In order to evaluate the quality of sinter ore intuitively and quantitatively,a comprehensive evaluation index of sinter ore quality based on clustering algorithm was established.The extra trees algorithm was used to establish a sinter comprehensive quality forecast model consisting of a quality comprehensive index classification model,drum strength and sieving index regression models.The results show that the F1-score of the quality comprehensive index classification model is 0.92,and the R~2of the drum index and the sieving index regression models are 0.89 and 0.85,respectively,which realizes the intuitive and accurate evaluation of the sinter quality index.In order to achieve rapid adjustment and multi-objective optimization of sintering proportioning,a proportioning adjustment and optimization model consisting of a proportioning calculation and optimization model and a mixture performance prediction model was developed.The test results show that,for the case of fine adjustment of the raw material ratio,the model can achieve rapid adjustment and optimization of sintering batching under the conditions of considering sinter chemical composition and quality index requirements,as well as raw material inventory and cost constraints.It realizes scientific and rapid decision-making guidance for small change material operation in sintering production.According to the actual application requirements of the sintering process parameter prediction and optimization system,a detailed system hardware and software structure design was carried out.The application results show that the system can effectively improve the stability of the sintering end point and the sinter ore composition,increases the sinter quality index,and makes the enterprise obtain obvious economic and social benefits.Figure 58;Table 39;Reference 178... |
PDF Full Text Request