Roof separation is one of the main precursors of coal mine accidents at the workface. At present, indicator data are usually read in an attempt to understand the real-time changes of the roof separation, but the change trend cannot be predicted. Predicting the change trend of roof separation on time can play a very important role in roof control. In this study, we apply the gray theoretical analysis method to predict the change trend of roof separation. We use Universal Distinct Element Code (UDEC) numerical simulation software to simulate the generation processes of roof separation. For this purpose, we consider four scenarios with different immediate roof heights. Using experimental separation data as original data, we predict roof separation using the gray algebraic curve model (GAM). The GAM-predicted data show little error when compared with the experimental data and are also better compared with prediction data obtained with the traditional gray model GM(1,1). Finally, in a case study using fieldmonitoring roof separation data from a mine, the GAM method led to good prediction results, and the predicted development trend reflected the real conditions. Thus, we prove that GAM is a very effective prediction method, which will be very useful in predicting roof separation trends for the warning of roof fall accidents.