The operation of cascaded reservoirs is a complex problem, and lots of algorithms have been developed for optimal cascaded reservoir operation. However, the existing algorithms usually have disadvantages such as the ‘curse of dimensionality’ and prematurity. This study proposes a grey discrete differential dynamic programming (GDDDP) algorithm for effectively optimizing the cascaded reservoir operation model, which is a combination of the grey forecasting model and discrete differential dynamic programming (DDDP). Additionally, a modification of the grey forecasting model is presented for better forecast accuracy. The proposed method is applied to optimize the Baishan-Fengman cascaded reservoir system in the northeast of China. The results show that GDDDP obtains more power generation than DDDP with less computing time in three cases, i.e., dry years, wet years and the whole series. Especially in the case of the whole series, the power generation of GDDDP is 2.13 MWH more than that of DDDP, while the computing time is decreased by 66,161 ms. Moreover, the power generation of GDDDP is comparable with that of dynamic programming but the computing time is much less. All these indicate GDDDP has high accuracy and efficiency, which implies that it is practicable for the operation of a cascaded reservoir system.

You do not currently have access to this content.