中文摘要：

为了应对随机动态规划算法在解决梯级水电站群长期发电优化调度时的“维数灾”问题,并行化方法得到了广泛研究。单机多核并行算法扩展性不强;传统的分布式并行算法编程复杂,缺少负载均衡和容错机制。云计算平台作为新的分布式计算平台能够充分利用资源,具有诸多优势。为了探索云平台下的分布式并行随机动态规划模型,该文基于消息传递接口（message passing interface,MPI）和Spark框架分别实现了传统集群计算和云计算分布式随机动态规划算法,后者将计算模型转换为数据处理模型进行计算,并通过三库优化调度实验对算法进行了比较。算法分析及实验结果表明,基于云计算的分布式并行随机动态规划算法则可以有效利用云平台的优势,同时拥有完善的容错以及负载均衡机制,具有广阔的应用前景。

英文摘要：

In order to solve the problem of ＂curse of dimensionality＂ in long-term optimal operation of cascade hydropower stations, parallel methods for stochastic dynamic programming are widely studied. The scalability of multi-core parallel algorithms for single computer is limited; the traditional distributed parallel algorithms are difficult for programming; in addition, they lack the load balance and fault tolerance mechanism. As a new distributed computing platform, cloud computing platform can make full use of resources and has many other advantages. In order to explore implementations of distributed parallel stochastic dynamic programming models on cloud platforms, this paper implemented traditional cluster and cloud computing distributed parallel method on stochastic dynamic programming algorithms based on message passing interface （MPI） and spark framework respectively. The algorithm on spark framework transformed calculation model into a data processing model for computation. These algorithms were compared with each other through three reservoirs’ optimal experimentations. Analysis and Experimental results show that distributed parallel stochastic dynamic programming based on cloud computing can make full advantages of cloud platform and has effective fault tolerance and load balancing mechanisms.