| Pre-decomposition and suspension preheating are being regarded as the core technology of the new dry process for cement production, which have been highly developed and widely used. The decomposing furnace is the core equipment of pre-decomposing system, which functions as burning fuel, heat transferring between gas and solid and pre-decomposing the carbonate before entering into the rotary kiln. The stability of decomposing furnace has an important effect on the decomposing rate of raw material, the stable operating of rotary kiln and the quality and output of final production. The decomposing furnace temperature is a key processing parameter of pre-decomposition system. However, the decomposing furnace temperature process has some characteristics, such as complex mechanism process, various production conditions, strong nonlinearities, strong couplings, and uncertainties and so on. It is difficult to establish exactly the process mechanism model and the data-driven model, thus the control and optimization methods based on model are very difficult to be applied in the decomposing furnace temperature process. Therefore, it is very important to research on modeling approach of decomposing furnace temperature process, which promotes to study optimized control methods of decomposing furnace temperature process, to achieve optimization control and product quality and production efficiency.This dissertation is supported by the National Hi-tech863/CIMS."Intelligent Control System of Large Scale Rotary Kiln", In order to deeply study the optimization technology of decomposing process, this dissertation proposed a dynamic model based on multi-step ahead prediction error criterion approach, the main work is as follows:l)An improved ARX model of the decomposing furnace lemperature process which based on multi-step ahead prediction error criterion is proposed to tackle the problems of one-step ahead prediction error criterion that often results in a poor performance on multi-step ahead prediction, which is also not suitable for control and optimization. To solve the problems of highly computation cost and low efficiency of the grid search method, comes up with the Nonnegative Garrote regularization method of the model order selection. The input variables consist of single pipe rotational speed in the kiln tail, the raw material flow and three winds volume. The output variable is the decomposing furnace temperature. Choice of sequential quadratic programming method to solve nonlinear constrained optimization problems. Finally, through verifications and analysis of the experiment, the results show that there is a fast and efficient performance of the Nonnegative Garrote regularization method, and a better multi-step ahead prediction performance of the decomposing furnace temperature process ARX model based on the multi-step ahead prediction error criterion.2) On a class of nonlinear process, a Hammerstein model based on multi-step ahead prediction error criterion is proposed. For the problems of computational complexity and large quantity parameters in the static nonlinear part of Hammerstein model, extreme learning machine(ELM) method is used to simplified the nonlinear model parameters. Through the Nonnegative Garrote regularization multi-step ahead prediction error criterion, the model parameter estimation and model parameter selection are integrated into one optimization problem, and use ’three-stage solution algorithm’to estimate the parameters of the linear part, the parameters of the nonlinear part and weights of Hammerstein model respectively. Finally, the established model is applied to the decomposing furnace temperature process. Experimental results illustrate a better multi-step ahead prediction performance compared with the Hammerstein model based on one-step ahead prediction error criterion, and also the better multi-step ahead prediction performance to the ARX model based on multi-step ahead prediction error criterion. |
PDF Full Text Request