Research on Soft Measurement Modeling for Industry Rotary Kiln Based on Flexible Neural Network

Combining with the control difficulty that output index of drying & calcination process of industry rotary kiln can not be measured on-line, this paper carries out a research on modeling of soft measurement for output index, adopting the flexible neural network which has excellent flexibility and study ability. During training process of the model, not only are the connective weights and thresholds trained by Levenberg-Marquardt algorithm, but also the shapes of nerve cell´s transfer functions are trained by gradient descent method too. Results of experiment show the modeling method has rather quick convergence speed, high fitting precision, and also achieves good predictive effect. It will be helpful and effective for drying & calcination process to set up the real-time feedback control system of output index, and also has reference and application value for complex process modeling and parameterr prediction in industry.