This paper presented an approach of decoupling the system represented by a time series of dam monitoring data into a series of subsystems and modeling these subsystems with non-linear models. The decoupling analysis of original system was achieved using wavelet packet energy spectrum (WPES). The non-linear models used here is augmented mixed regression (AMR) model. Some detail problems such as how to find the scale for fine wavelet packets, the modeling of sub system, how to choose a forecast model and the data de-noising were studied in this paper. Through the analysis on the monitoring data of a roller compacted concrete (RCC) gravity dam, it proved the good performance of this model in data prediction.