Abstract: In this paper, a grey discrete differential dynamic programming (GDDDP) algorithm based on the grey system prediction is presented for solving the problems of poor global convergence and low calculation efficiency in application of the discrete differential dynamic programming (DDDP) to optimal operation of cascade reservoirs. As an improved DDDP method for better accuracy and efficiency, this new algorithm adopts the grey system prediction and makes forecasting before each iteration, and then starts the iteration using the forecasted trajectories of all the reservoirs and taking the data sequences of optimal trajectories and forecasted trajectories obtained in previous iterations as an input of the ongoing iteration. It also adopts a recursive arithmetic progression formula to eliminate the low accuracy defect of the grey system prediction method in its forecasts of oscillatory sequences. A case study of the operation of Baishan-Fengman cascade reservoirs is given to verify the effectiveness of GDDDP.