Abstract: Different strength and interaction (hydrolysis, association) varing from different batches of dyes causes the first spline results of computer color matching system inaccurately, you need to manually adjust formula time and again to achieve customer requirements. For these issues, I first dyed according to the orthogonal experiment, and took the result as a standard sample, then analysis the error between standard sample and prediction one which corresponds to the initial formulation by CCMS（Computer color matching system） forcasting, and established a non-linear model between them. Based on the model and error analysis, and combining CCMS prediction, I chose particle filter to estimate the actual recipe, thus got PF algorithm prediction formula. Experiments show that the nonlinear model is correct, the color error（CIELAB） which using PF algorithm to predict will reduced more than 50% than the CCMS prediction. Thus, the nonlinear model is accurate and the method of using PF algorithm to predict the actual formula is feasible.

Key words: computer color matching, fabric, modeling, particle filtering, forecasting