Cloud-Based Multidimensional Parallel Dynamic Programming Algorithm for a Cascade Hydropower System

The joint optimal operation of cascade reservoir system can greatly improve the utilization of water resources. However, the complex high-dimensional and non-linear features and calculated costs often hinder the refined operation and management of reservoirs. Recently, the local parallel computing has become an effective way to alleviate the "curse of dimensionality". Current local parallel computing has hardware limitations, which is difficult to adapt to large-scale computing. This study proposes a novel parallel dynamic programming algorithm based on Spark (PDPoS) via cloud computing. The simulation experiments are carried out for a comparative analysis of the solution efficiency, influence factors and stability of cloud computing. The results are as follows: (1) The efficiency of the cloud-based PDPoS is related to some factors; the number of CPU cores is the main influencing factor, followed by the operator, and the architecture has the least influence. (2) The runtime variance of cloud computing is 2.03, indicating cloud computing has high stability. (3) Under the same configuration (i.e., CPU and memory), the runtime of cloud computing is 41.5%?~?110.3% longer than that of physical machines. However, cloud computing has rich resources, good scalability, and good portability of online operations, which is an attractive alternative for optimal operation of large-scale reservoir system.