Based Soft Instruments Of Of Drnn Polyester Viscosity

Soft Sensing Technology based on Neural Networks(NNs) has been widely investigated and utilized in process control in recent years. Prevenient soft sensor, however, mostly taking static feedforward NNs as modeling tools, is not able to acquire expectant performance constantly.In this paper, research on a new dynamic NN-Diagonal Recurrent Neural Network(DRNN) is made, which shows that DRNN, with feedforward and feedback simultaneously, is provided with much closer behavior to nonlinear dynamic systems and of great advantage in the modeling of dynamic systems.The practical work of this paper is carried out according to the polyester production in Tianjin Polyester Factory, in which the online sensing of ployester's viscosity remains a headache. Simulation shows that the soft sensor for polyester viscosity is with sound estimation precision and dynamic tracing ability, which presents a promising prospect in practical use.In order to improve the competence of DRNN in dynamic mapping, some measures to modify the architecture of DRNN are put forward in this paper. Simulation shows those steps are advisable. Further study in theory and algorithm of the modification, however, is supposed to be carried on.