With the impact of climate change and human activities, river flow regime presents complex fluctuation in recent years, which brings limitation to the application of those normal statistical analysis flow forecast model. The multi-factor auto-regression model (ARX) can simulate the complex fluctuation system with higher precision and is improved to simulate and forecast the river flow of a highly impacted river reach -the Xiangyang station from 1960 to 2002. By comparing with some traditional simulation methods, the result proves that the improved RAX model performs the best. Based on the RAX simulation model, annual flow at the Xiangyang station from 2003 to 2012 was forecasted. The reliability rate of forecast is 90%, which indicates the applicability of the RAX model in mid-long term river flow forecast.