Abstract: Abstract ：Modification of linen fabric is performed via UV initiated photografting of acrylic acid in this design. BP neural network and least squares regression modeling methods are used to predict the relationship between grafting ratio and air permeability under different conditions of photografting time, photoinitiator amount and concentration of acrylic acid, respectively. A three-layer BP network model with architecture of 1-10-1 is established after extensive discussion, including one node in one input layer representing grafting ratio, one node in one output layer representing variation of air permeability and ten nodes in one hidden layer. The activation function of sigmoid is selected. The optimum parameters, training step of 100 and training goal of 0.001, are determined. The correlation coefficient of BP neural network model is higher than least squares regression model, while the percentage error is lower. Therefore, BP neural network has higher simulation precision, which provides an effective predictive model for the relationship between graft ratio and air permeability.

Key words: air permeability, grafting ratio, mathematical model, BP neural network, least squares regression method