Soft sensor system of coke oven flue temperature based on CBR and PCA-RBFNN

Wentao Xiao¹, Gongfa Li^1,2, Honghai Liu², Guozhang Jiang¹, Ze Liu¹, Disi Chen¹, Weiliang Ding¹, Wei Miao¹, Zhe Li¹

¹College of Machinery and Automation, Wuhan University of Science and Technology, Wuhan 430081, China

²Intelligent Systems & Robotics Group, School of Computing, University of Portsmouth, Portsmouth, PO1 3HE, United Kingdom

The key process indicator - coke oven flue temperature – is difficult to detect online with instruments in the coke oven heating process, thus an intelligent forecasting model is developed which is composed of four parts: the data gathering and handling unit, the optional forecasting unit, the online amendment unit and the effect evaluation unit. The optional intelligent forecasting model and its corresponding algorithm are established for different categories of practice operating conditions. In normal operating condition, the nearest neighbor clustering algorithm based on the principal component analysis and neural network with the radial basis function is selected. In unconventional operating condition, the case-based reasoning technology is selected. The models of different conditions are validated and applied according to the actual data in a steel enterprise coke production, the results show that the established forecasting model can reflect different practice conditions and meet the real-time control requirements.