Modeling Methods For Temperature Process Of Decomposing Furnace Of Cement Raw Material

The technology of the new type dry process cement production is the main development direction of the modern cement industry, and the predecomposing technology is one of the core techniques of the new type dry process cement production. Decomposing furnace is the main equipment to realize predecomposing technology. The functions of decomposing fur-nace are burning fuel, transfering heat between gas and solid and decomposing carbonate. The decomposing furnace temperature is a key index of stable running condition of decomposing furnace, which has a great effect on the decomposing rate of the raw material and the stability of the final production quality and yield. However, due to decomposing furnace, rotary kiln and preheater are linked directly, there are many uncertainties of the disturbances. At the same time, the mechanisms of combustion and heat transfering in decomposition furnace are com-plex. Therefore, some characteristics of the decomposing furnace temperature process, such as strong nonlinearity, pure time delay, strong coupling, uncertainties and so on. It is difficult to establish the accurate dynamic mechanism model and data-driven model of the decompos-ing furnace temperature process. As a result, the control and optimization methods based on model are very difficult to be applied in the decomposing furnace temperature process. The research on the modeling methods of the temperature process of decomposing furnace, espe-cially the establishment of a smaller simulation error model, is of significance on the control and simulation research of the temperature process of decomposing furnace.This thesis is supported by "large rotary kiln intelligent control system" of the national high technology research and development program of China (863). Due to the difficuity to establish a smaller simulation error model with current system identification methods, this thesis proposes three modeling methods that can improve the model performance, and the dynamic models of the decomposing furnace temperature process based on the actual data are established. The main work of this thesis is as follows:(1)Aiming at the bigger simulation error of the NARX(Nonlinear Auto Regressive with eXogenous Inputs) model, an improved model selection method is suggested for support vec-tor machine(SVM). This method can be used to improve the model performance so that the model can be more close to the actual object. A SVM NARX model of the decomposing fur-nace temperature process is established by this method, which is based on the actual data. The simulation results shows that this method can reduce simulation error of the model and im-prove the model generalization. (2)Given the smaller simulation error of output-error (OE) model and more difficulty of establishing the nonlinear OE model, this thesis proposes an ensemble output-error model based on the AdaBoost(Adaptive Boosting) algorithm. The output-error model is chosen as the weak learning machine of the ensemble model. In order to get the weak learning machine models with different performance, AdaBoost algorithm is used to update the sampling weights. According to the great complexity of ensemble model, selective ensemble strategy is proposed to prune the ensemble model. This method has also been used to establish a ensem-ble output-error model of the decomposing furnace temperature process. The simulation re-sults shows that this model has good learning performance and generalization ability.(3)In view of the problem of too much parameters of the nonlinear module existing in Hammerstein model and high cost of learning, a Hammerstein output-error modeling method is proposed based on Extreme Learning Machine(ELM) in this thesis. The ELM model is chosen as the nonlinear module, and the output-error model is chosen as the linear module, then a two-step iteration optimization algorithm to estimate the parameters of the model is suggested. Finally, this method is used to establish the model of the decomposing furnace temperature process. The simulation results shows that the model structure is relatively simple, and learning complexity is quite low, and the simulation error of the model is reduced to a certain extent.(4)A comparative analysis is done on three modeling methods from the emphasis of the modeling, the modeling time, the complexity of the model, model accuracy and generalization performance. The results shows that there are different emphases about three algorithms. SVM NARX model is mainly a kind of model selection algorithm, while integrat output error model and Hammerstein output error model based ELM are model structures which can re-duce the simulation error; In terms of the complexity of the model, modeling time, model ac-curacy and generalization capability, three methods had their own advantages and disad-vantages. But on the whole, the precision of the SVM model is the lowest and the rest algo-rithms have nearly the same performances in study accuracy and prediction accuracy.