In this article we present a method for the on-line identification and modeling of full profile disturbance models for sheet forming processes. A particular principal components analysis technique called the Karhunen-Loeve expansion is used to adaptively identify the significant features of the profile. In addition, we show how the temporal modes of the reconstructed profile can be modeled using low-order linear autoregressive (AR) processes. By simulation examples, the effect of the order of the AR model is studied, as well as the window size of the data used in the on-line application of the KL expansion, the effect of data weighting, the importance of the correct selection of the number of modes, and the frequency of updating the parameters of the RR models. Identified disturbance models can be easily incorporated into model-predictive control algorithms.