Data-based Modeling Of Decomposition Furnace And Rotary Kiln Subsystems

The new dry cement production mainly includes raw material preparation,clinker burning and cement grinding.The clinker burning is the core part of the new dry cement production.It includes four subsystems: preheater,decomposition furnace,rotary kiln and grate cooler.The decomposition furnace is the core equipment of the precalciner kiln,and its main task is to complete the decomposition of carbonate.The temperature of the decomposition furnace has a crucial influence on the decomposition rate of carbonate and the quality of the final cement clinker.The rotary kiln is a device for completing the clinker burning process,the change of the kiln current can reflect the working state of the rotary kiln.Therefore,the decomposition furnace and rotary kiln are taken as research objects,and the modeling for decomposition furnace and the rotary kiln are based on the actual production data.Based on the “Data-Based of Identification and Modeling of Cement Production” project of Southeast University and Nanjing Kaisheng International Engineering Company,this paper has done the following work:?1?Analysis of key data for the decomposition furnace and rotary kiln systems.There are many variables associated with the decomposition furnace and the rotary kiln systems.In the modeling process,if all the variables are taken into consideration,the model will be very complicated.Therefore,it is necessary to analyze the key variables of the two systems and determine the input and output parameters.After analysis,the output variable of the decomposition furnace system is defined as the outlet temperature of the decomposition furnace.The input variables are the quantity of coal-feeding of the decomposition furnace,the tertiary wind temperature and the quantity of raw-material-feeding.The output variable of the rotary kiln system is the rotary kiln current,and the input variables are the amount of the coal fed to the kiln head,the rotation speed of the kiln and the amount of raw materials fed into the kiln.?2?Establishing the model of the decomposition furnace system.The single-input single-output model and the single-input multi-output model of the decomposition furnace are established by the ordinary least square algorithm and recursive least squares algorithm respectively.The experimental results show that the model prediction error obtained by the ordinary least squares algorithm is between-2?C¹⁰1.5?C,but it has a certain hysteresis,and the model prediction error established by the recursive least squares algorithm is between-1?C¹⁰1?C,and the phenomenon of predicting lag is obviously improved.Some data have strong correlation among them.Therefore,Principal Component Analysis should be performed on the highly correlated data to obtain the principal components,and thus establish the Principal Component Model.It has been found that the applicable duration of a model is not infinite.After a period of time,the prediction bias of the model tends to be larger and larger.In order to solve this problem,the model can be modeled by receding-horizon training base on the data and the model is updated regularly to ensure the accuracy of the model prediction.?3?Establishing the model of the rotary kiln system.First of all,it is necessary to judge the working conditions of the rotary kiln based on the data.The back propagation neural network model,the first-order input-output model and the high-order input-output model are established for different working conditions,and the advantages and disadvantages of the model under different working conditions are compared.For the data with strong correlation,the principal component model is established by principal component analysis.The simulation experiments show that the principal component model can achieve better results in predicting stability and accuracy.For the issue that a model is not always applicable,the method of receding horizon training can control the error within a certain range.For the rotary kiln system,in addition to the above model,an incremental model is also established.The experiment proves that the incremental model can also achieve good prediction results.?4?Simulation of on-line identification and the development of system identification software.In order to facilitate on-site debugging,this paper combines MATLAB and the OPC simulation server to simulate the online identification process and develop the system identification software.