A NPLS modeling method of outer-inner polynomial model based on variable selection

Multi-sensing information modeling usually has problems of multi-correlation and nonlinearity, a bi-polynomial NPLS modeling method based on variable selection is proposed. The method first uses PLS modeling method for variable selection, in the PLS variable selection, VIP indexes and PLS regression coefficients are used as criteria of variable selection to eliminate some unimportant or uninformative variables, and to obtain an optimal variable set. Then based on PLS variable selection, the proposed NPLS modeling method uses polynomial to realize double nonlinearization of PLS outer-inner model, it solves information coupling and nonlinear problem of multi-sensing information. Simulation result demonstrates the proposed NPLS modeling method can realize dimensionality reduction and nonlinear modeling efficiently, and its prediction accuracy of the established models heighten 56.2% and 24.7% respectively.