Study On Soft Sensor Of Viscosity For Petroleum Resin

A soft sensor is constructed by the research of equipment 400 unit of petroleum resin of CNPC LANZHOU CHEMICAL COMPANY, to solve the problem that the viscosity of petroleum resin can not be detected on-line.Based on mechanism-analysis of craftwork flow, combining with the factors that affect viscosity of petroleum resin and the operating conditions, secondary variables used for viscosity of petroleum resin soft sensor are determined. On sample data collected from real plant, the pretreatments are conducted by lapse-error-way, random-error-way and unitary-way with data statistic theory; On the basis of analysis of several methods for modeling, a soft sensor based on kernel principal component analysis(KPCA) and least square support vector machine(LSSVM) is proposed. KPCA is employed to compress data for extracting the nonlinear component, then LSSVM is used to develop regression model. By doing so, the complexity of calculation will be reduced and the generalization ability improved. Cross validation method is used to choose proper parameters of LSSVM method. Soft sensor is applied to predict viscosity of petroleum resin. Results indicate that this method can provide good approximation and high precision. The precision of prediction is improved, compared to the method used currently in plant. It is proved to be an efficient modeling method.The model proposed here possesses higher precision compared to other models which are built by the methods of partial least squares regression (PLS), RBF neural network, LSSVM, principal component analysis(PCA) combining with LSSVM.The correction function is provided, which includes the correction of output and parameters of the model. The precision and generalization ability are improved furthermore. The model has good value for wide application.