Abstract: Considering the uncertainty of information and the nonlinearity and dynamics in large scale complex industrial processes, a system level process modeling and condition forecasting approach is developed. A time delay estimation method is proposed based on the combination of wavelet transform and mutual information theory. Then, a probabilistic time delayed signed digraph model is built, based on which, a combined prediction method is adopted to estimate the future status of key process variables. The proposed approach is verified on an air separation process. Preliminary results show that the approach is effective and has satisfying prediction performance. This study is expected to have a good prospect in industrial application.