Research And Development Of Soft-Sensor Based On RBF Network In The Poly Industry

Neural Network is a very convenient and valid technique to establish the model of soft-sensor, and Radial Basis Function is the best one in Neural Network by its advantages , such as rapid learning speed, convergence in whole range, good self-adapting ability and could be modified on-line and so on.To the problem of electing the soft-sensor theory to estimate the product quality in the industry of PET, which is illustrated by the viscosity of PET, the technique of Principle Component Analysis combining the SPE and T~2 control limits is used to detect gross error, on the basis of this, improved PCA dividing SPE into PVR and CVR two parts is used to diagnose if the bad data is caused by system exceptions. Then soft-sensor models based on RBF Neural Network is established, and the software package is developed by VC++.Detecting and getting rid of the gross error data is the critical...