Predictive Control On Rotary Kiln And Soft Computing Modeling For Its Parameters

Rotary kiln is a complex plant with nonlinear, strong coupling and long time-delay features.It is a tough work for automation of rotary kiln.There are three problems researched in this paper to improve the performance of automation system of rotary kiln.In order to improve the performance of the control system of rotary kiln in normal state, two improved predictive control theory are introduced.One is a predictive control algorithm based on error prediction with gray theory.Another is a kind of simple GPC algorithm.The former strengthens The DMC and MAC system's capability of anti-distubance by using gray theory to predict the error caused by model mismatch and interference.The latter improved predictive control method avoids the computation of Diophantine equations by transforming the predictive model. Compared with traditional GPC algorithm, it can run more quickly. The above improved predictive control methods are applied in the control system of the rotary kiln.The experiments show that they can improve the performance of control system of rotary kiln.The forecast of the temperature of burning zone is significant for the design of control system of rotary kiln.Therefore,a kind of mixtures of experts is introduced, which needs only a small number of iterations for training, further more,each iterations takes up less time.So it can match the requirement of on-line training in some field.Compared with the fuzzy neural network and BP neural network, it can be found that the mixtures of experts model is more suitable for forecast the temperature of rotary kiln'burning zone.When the rotary kiln is in the non-normal state, the expert control or the fuzzy control is usually used to deal with the harder problem.To improve the quality of knowledge used by the expert control system or the quality of fuzzy rules used by fuzzy control system,a dynamic committee machine model is introduced.First, the model is trained with the historical data, i.e.historical operation records.Then, from its parameters and structure,knowledge or rules are extracted.This knowledge or rules can be used as the knowledge of the expert control system or can be used as the rules of the fuzzy control system.Several committee machines can also be jointed to work as one part of the expert control system.This article uses three different classifying tools to design intelligent machines for control of the coal.The result shows that the performance of the intelligent machine based on dynamic Committee machine is similar to the those performances which made by the intelligent machine based on BP neural network and the intelligent machine based on static Committee machine, but the intelligent machine base on dynamic committee machine displays relatively more explicit physical meaning, and is more suitable for engineers to analysis and correction it, or to extract knowledge and rules.