梯级水库短期优化调度模型的精细化与GPU并行实现

目前制约梯级水库短期优化调度在实际工程中应用的主要瓶颈有:所构建的优化模型存在不合理的简化策略,所选择的求解算法无法保证解的质量以及模型的计算时间远超规定时长。为解决上述问题,本文首先构建精细至水电站各机组工作特性的优化调度模型,接着通过二重嵌套动态规划(DP)计算给定模拟精度下的高质量解,并针对算法固有的"维数灾"问题,一方面通过数据压缩与数据库技术降低程序的内存占用量,另一方面将GPU并行加速技术首次引入水库调度领域,通过OpenACC实现算法的GPU并行以减少计算时间。最后通过潘口、小漩梯级水库日优化调度的实例研究与对比分析得出:精细模型较传统模型能更好地贴合电站的实际工况,提高梯级系统的发电效益;内存占用缩减策略的引入能有效降低算法的空间复杂度;GPU并行较传统的CPU并行能大幅提升算法的求解速度。由此为短期优化调度的理论发展与算法"维数灾"的处理提供借鉴。

Model precision and GPU parallelism for short-term optimal operation of cascade reservoirs

There are currently three main bottlenecks restricting the application of cascade reservoir short-term optimal operation in practical engineering-unreasonable simplification strategies in an optimiza-tion model, difficulty in guaranteeing the quality of a solution and calculation time much longer than re-quired. To address these problems,a precise optimal operation model is firstly developed in this paper,fin-ing down to the operating characteristics of each generator set of a hydropower station, and then a high-quality solution with given simulation accuracy is worked out by double-nested Dynamic Programming (DP). The inherent "dimensionality curse" of the algorithm is also tackled. On the one hand, the memory footprint of the computation program is reduced by data compression and database technology. On the other hand, OpenACC is used to realize the GPU parallel acceleration of the algorithm, a technology first intro-duced into the field of reservoir operation,to shorten the computation time. Finally,the daily optimal opera-tion of Pankou and Xiaoxuan cascade reservoirs is taken for case study,and the following conclusions are drawn. Compared with the traditional model,the precise model can better fit the actual operating conditions of the power stations and enhance overall power generation benefits; the space complexity of the algorithm can be effectively lowered by employing the approach of memory footprint reduction; and in contrast to the traditional CPU parallelism, GPU parallel acceleration can considerably improve the speed of solving the model. The research results provide some reference for the theoretical development of reservoir short-term optimal operation and the processing of the "dimensionality curse" issue.