Research On Application Of Bayes Classification Algorithm Based On ICA For Rotary Kiln Control

Rotary kiln is very important industrial equipment in the process of alumina production. It has a big volume, and it has high energy consumption of which thermal data is hard to measure. The process in rotary is characterized with large inertia, long time delay, nonlinearity, time-dependent and multi-disturbance and the kiln body keeps rotating during the production. The transformation of aluminum ore to alumina is a very complex physical reaction and chemical reaction process. As a result, the rotary kiln is very hard to control well. Therefore, in order to reduce operator's workload, improve product quality and save energy, it is of great theoretical and practical value to research and develope high performance control system of rotary kiln. Based on analyzing both the advantages and disadvantages of current control technology of rotary kiln, we designed a coal feed predictive module in rotary kiln based on bayesian classification, and applied it to field control. The major tasks and innovation points of this paper are as follows.First, we develop a bayesian classification algorithm based on independent component analysis method (ICABC). The method of ICA is applied to na?ve bayes classifier, which can transforms an observed multidimensional random vector into components that are statistically as independent from each other as possible to satisfy the attributes independence of na?ve Bayesian. It uses the fastICA algorithm as standard ICA model estimator. The maximization of non-Gaussian criterion with negative entropy is adopted to the computation process of the fastICA algorithm. The final contrast results of experiment using standard databases from UCI showing that, the ICABC is more effective than the na?ve bayesian classification algorithm.Then, we construct a coal feed predictive module based on ICABC using the field kiln data. The data is filtered, and then discreted into time series trends by line approximation method. Then the pre-processed data compose the training sample collection and validation sample collection. In this section, AdaBoost algorithm is adopted to boost classifiers for improving the classification accuracy. The final experiment results proved that it is feasible to apply the ICABC algrithm in control of rotary kiln.Finally, we add the coal feed predictive model based on ICABC to the actual rotary kiln former intelligence expert control system to assist the intelligent control system making real-time decisions of sintering temperature control. The application effect shows that this technology of data mining in rotary kiln control is of feasibility.