Aiming at the center wavelength shift problem in the fiber Bragg grating(FBG) sensing system, an FBG wavelength shift predict and generalization method based on extreme learning machine(ELM) is proposed. The method trains the neural network by constructing the mapping relationship between the center wavelength and the wavelength shift. It can realize the prediction and generalization of the wavelength shift. The experiment results show that the predicted error of FBG wavelength shift is less than 1 pm and the generalized error of FBG wavelength shift is less than 2 pm, which provides a useful exploration for the field calibration of FBG sensor system.