In order to improve the forecast accuracy of film screenings in each cinema in the mobile digital film projection, this paper proposes a new method combining nearest neighbor method and spatio-temporal sequence together based on the spatio-temporal correlation of cinema screenings. Firstly nearest neighbor method was used to find cinemas which correlated the predicted cinema closely. Secondly, spatio-temporal sequences used for a neural network were constructed to combine the spatio-temporal characteristics together, getting more accurate forecast model. Experiments compared the forecast result of this method with the tradition one, demonstrating its higher accuracy.