SVM Classification For Rotary Kiln Based On Time Series Trend Characteristics

Rotary kiln sintering is a complex industrial process in the kiln production area. Because of features with large time delay, strong coupling,multi-interferences, time-varying and nonlinear, appropriate system model is difficult to be established and the detection and control for sintering temperature is also a tough work.In recent years, there are some progresses on detection and control of sintering temperature, but soft sensing for sintering temperature based on flame image is easy to be disturbed by dust and smoke, which limited the control system's application. In addition,the methods depend on field signals without further extraction of trend characteristics would overlook the dynamic trend of thermal signals, and the accuracy and generalization ability are also decreased correspondingly.Compared to the methods directly using field signals, a scheme of Support Vector Machine model based on trend characteristics is presented in this paper. The dynamic trend characteristics of multi-sample are used to predict the coal feeding trend. Manual operation would be realized under the direction of accurate prediction. In this way, single signal rotary kiln control which only depends on sintering temperature will be avoided, and the robustness of kiln control is improved.Firstly, the rotary kiln thermal data is preprocessed. The rotary kiln sample is segmented on the basis of piecewise linear representation by key change points or sliding window. The trend characteristics are extracted to form the training sample. Clustering and Rough Sets are used to reduce the train sample.Secondly, parameters of SVM are optimized by heuristic optimization algorithm. It was proved that the proposed model with high prediction level is achieved.Finally, classifications of coal feeding based on single characteristics of sample points and based on trend characteristics of key change trends are designed to verify that the latter has higher accuracy. The models based on trend characteristics are used to explore the method which can improve the classification accuracy of coal feeding. It was proved that the proposed model with high prediction level which enhanced the robustness of the control of rotary kiln could be applied on the actual situation.