Predicting the newsprint tear strength in MD on the effective variables (Case study: Mazandaran Wood and Paper Company)

In paper making process، different variables are measured to improve the product runnability. These variables affect the paper properties. The main question is، which one of these variables is the most effective variable on the newsprint tear strength in MD? Therefore، 145 variables of online data from 2009 to 2011 periods were chosen by modeling methods of liner regression and modeling of Gamma Test (GT)، M-Test (MT)، Genetic Algorithm (GA) and Artificial Neural Network (ANN). 7 variables were selected on the base of liner regression of stepwise and then ANN models were created on the basis of these variables. Selected variables were as following; 1. Machine Cons. including additives and stock، 2. Slice open headbox، 3. Rush/Drug-wire، 4. No. 8-Vacuum pumps، 5. No. 10-Vacuum pumps، 6. Press 4-clothing tension، 7. Fan output-Heat recovery1. Generally، results of GT، MT and GA were showed that all of 7 variables have been the most effective on the newsprint tear strength. The best model of BFGS neural network has shown that mean absolute percent error and the correlation coefficient are equal 0. 011% and 0. 97، respectively.