为提高连通水库群水资源利用效率，耦合预报信息的调度方式、制定合理的调度规则和寻求高效的求解方法是供水水库群优化调度研究的前沿和热点。选取研究相对较少但具有典型性的双向连通水库群为研究对象，首先结合有物理基础的变预报因子，研制多模型贝叶斯组合预报方法，以提高预报精度；然后制定连通水库群调度规则，尤其是引水规则及其决策方式；在明确中长期预报信息利用方式的基础上，构建模拟优化调度模型，结合贝叶斯理论推导建立考虑中期预报信息不确定性的随机动态规划模型，从而实现预报信息与调度方法相耦合；系统破解线程通信方式、并行程序设计、存储性能优化等关键技术，构建CPU/GPU协同并行求解框架，以实现模型的高效求解；最后形成“径流预报-优化调度-模型求解”一套预报调度理论与方法体系。本研究将促进水利学科与气象、计算机等学科的交叉融合与发展，完善连通水库群预报调度理论与方法，为水资源高效利用提供理论和方法支撑。

In order to improve the utilization efficiency of water resources in complex connected multi-reservoir system, the foreland and hotspot research of optimization operation in water-supply multi-reservoir system are focus on the following issues: reservoir operation method coupled with medium- and long-term runoff forecast; reasonable water supply-transfer rule in two-way water-transfer project, and effective solution method. Two-way connected multi-reservoir system which is the typical and less studied is taken as a study case. Firstly, to improve the availability of rainfall prediction, multi-model combination forecast based on Bayesian method, in combined with the physical basis forecast factors, is established. Secondly, the pattern of water supply-transfer rule, especially the water-transfer rule and how to apply these rules in two-way connected projects are figured out, and then the simulation-optimization model is developed after specifying the use patterns of the medium- and long-term runoff forecast. At the same time, considering the medium-term forecast information, the stochastic dynamic programming model is derived using the Bayesian theory. Finally, for enhancing the model computation efficiency, key technologies such as thread communication mode in parallel computation, parallel programming, storage performance optimization in GPU, are to be settled, based on which the CPU/GPU parallel solution frameworks is set up. Finally, the “runoff forecast-optimization operation-model solution” of the theoretic and methodological system on forecast and operation are established. This study will promote the fusion and development in water conservancy science with meteorological science, system science, computer science, et al. Most importantly, the forecast scheduling theories and methods in connected multi-reservoir system will be further perfected, which can provide the theoretical and technological supports for the effective utilization of water resources.