Research On Predicting Models Of Paper Tensile Strength On Line

It would be a waste of energy and less production with lower paper tensile strength, as well as a waste of materials and high energy consumption with larger paper tensile strength during papermaking process. Therefore, online prediction of tensile strength benefits instruction on controlling paper’s quality, avoiding breaks and energy saving.This dissertation proposed several paper tensile strength predicting models to cater real scale situation. This research were firstly proposed paper tensile strength predicting models with real production parameters as input parameters and the main research contents are as followed:(1) Partial-least squares (PLS) method was applied on predicting tensile strength based on22real production parameters through theoretical and real production analysis. The expression of PLS model was: ypre=β0+β1x1+β2x2+…+βpxp, where β0was a constant, p=1,2,...,22, each coefficient can be figure out by Matlab.Results showed that PLS model can be applied on paper tensile strength predicting which Person’s value r was0.732, Root Mean Square Error (RMSE) was276N·m-1, and Mean Relative Error (MRE) was5.17%. Besides that, the model had a good analytical ability that the6key elements could be interpreted very well from mechanism angle with variable importance in projection (VIP) value.(2) The input parameter’s number increased from22to30for modify on real production. Based on that, Support Vector Machine (SVM) method was applied on this research. Results showed that Linear-SVM model, whose penalty parameter C is3.3137had better precision. The expression of SVM model was, in which the number of support vector machine, n, was71, coefficient of support vector Wi was a71*1matrix and support vector xi was a71*30matrix.Results showed that SVM model can be applied on paper tensile strength predicting which Person’s value r was0.882, Root Mean Square Error (RMSE) was313N·m-1, and Mean Relative Error (MRE) was6.95%.(3) A hybrid method of PLS-SVM was approached. This hybrid model contained two parts, PLS and SVM. Principal component analysis of PLS reduced the dimensions of input data from30to6combined with linear SVM model. This PLS-SVM model’s expression wasin which the number of support vector machine, n, was153, coefficient of support vector Wi was a153*1matrix and support vector xi was a153*6matrix, and the rest parameters can be calculated by PLS model.Results showed that PLS-SVM model can be applied on paper tensile strength predicting which Person’s value r was0.912, Root Mean Square Error (RMSE) was277N·m-1,and Mean Relative Error (MRE) was6.05%.(4) According a model’s precision and practical application, three models were evaluated. Comprehensively speaking, the order of practical application of paper tensile strength model should be SVM model, PLS model and PLS-SVM model.(5) A measuring platform for practical application was approached based on the best model and the software of Visual Studio2010. The measuring platform successfully designed out through collecting data from DCS and loading SVM model by C#and ASP. NET. The system can query paper tensile strength curve of historical data and real-time data on line, which showed a basic realization of model of industrial application.