Abstract: In order to efficiently forecast the elevator interfloor traffic distribution state,a method to predict the interfloor traffic O-D matrix is presented.The method makes use of the advantages of both grey forecasting and neural network,and organically combines grey forecasting and radial basis function neural network to construct the grey method based radial basis function neural network(GM-RBFNN) forecasting model.The accumulated generating operation(AGO) in grey forcasting is used to converse the initial observed traffic data to obtain the accumulated traffic data with strong regularity which are employed to model and train the GM-RBFNN.Meanwhile,a technique which modifies the abnormal traffic data is presented to further reduce the randomness of the observed traffic data.The proposed method not only avoids the theoretical error of grey forecasting,but enhances greatly both the training speed and prediction accuracy of neural networks,so it is suitable for short period forecasting of elevator interfloor traffic distribution.Simulation experiments prove the validity of the proposed method.