基于智能流态分析的水泵水轮机甩负荷过渡过程旋转失速特性研究

During load rejection transient processes in pumped storage power stations, the hydraulic characteristics of pump-turbines, including the pressure pulsations, hydraulic forces and outer performances, always experience significant or even sudden variations. These variations can be attributed to the rotating stall flow patterns because these two phenomena occur at the same operating regions, show similar low-frequency features, and rotating stalls have been proved to be the main source of hydraulic instabilities in steady operating sates of pump-turbines. This project aims to reveal the formation and evolution mechanism of rotating stalls in a model pump-turbine during load rejections and their influences on pump-turbine hydraulic performances by one-dimensional and three-dimensional coupling numerical simulations. Firstly, an intelligent model based on artificial neural networks will be built to identify and analyze the rotating stalls. Then, the evolutions of the rotating stalls with the main control factors of load rejection processes, including the machine inertia, the water inertia in the whole system and the guide vane closing laws, will be analyzed. Finally, the connections between the rotating stalls and the hydraulic performances of the pump-turbine as well as the reasons of the significant variations in pressure pulsations, hydraulic forces and outer performances, will be investigated. Through the studies above, we strive to answer how the phenomenon of rotating stalls occur and evolve in load rejection processes, and how they influence the hydraulic performances of the pump-turbine during these processes. The innovation points of this project are that the load rejection controlling parameters are considered as the influencing factors of the rotating stall evolutions and an intelligent model which embodies the convolutional and recurrent neural networks will be proposed to analyze the rotating stalls.

水泵水轮机在甩负荷过渡过程中容易出现压力脉动、转轮受力和外特性等水动力特性的剧烈变化甚至突变，可能导致机组振动、破坏、抬机或损毁等极端事故。此过程中剧烈水动力特性的根源极可能是旋转失速，其依据是两种现象出现的工况区基本重合，同样表现出低频特征，且旋转失速已被证明是水泵水轮机稳态工况一系列不稳定特性的主因。本项目以一维输水系统与三维水泵水轮机耦合的数值模拟为主要研究手段，用人工神经网络构建旋转失速的智能分析模型以解决现有方法的不足，重点研究不同甩负荷控制参数下水泵水轮机的旋转失速特性，目的是揭示其在甩负荷过程中的形成机理、演化规律及对水泵水轮机水力特性的影响，以找到抽水蓄能电站在甩负荷过程中容易发生严重事故的原因。特色为突破原有流态固定于工况区的观念，尝试建立旋转失速特性与甩负荷控制参数之间的关系，创新是将人工智能概念引入水泵水轮机流态分析，提出结合卷积和递归神经网络的旋转失速智能分析模型。

抽水蓄能电站的水泵水轮机在甩负荷过渡过程中容易出现压力脉动、转轮受力和外特性等水动力特性的剧烈变化甚至突变，可能导致机组振动、破坏、抬机或损毁等极端事故。这些水力特性的突变可能与水泵水轮机转轮流态的失稳有关，而最典型的转轮失稳流态即为旋转失速。本项目以水泵水轮机的转轮流态为主要研究对象，探索过渡过程中转轮流态失稳以及旋转失速演化规律对水泵水轮机一系列水力特性的影响机理。主要研究内容为建立基于人工神经网络的转轮失速流态智能分析模型、研究甩负荷过渡过程中转轮流动失稳及旋转失速演化特性和探索转轮流态失稳对水泵水轮机水动力特性的影响规律。. 本项目构建了一个卷积神经网络模型，能够以100%的准确率对测试数据集中的转轮流态样本是否失速进行分类，具有高精度和强泛化能力，可以尝试将其作为水泵水轮机转轮流态诊断的标准工具之一。在对不同参数的甩负荷导叶拒动（飞逸）过渡过程中转轮流态的研究中发现，在由转轮流态不够平顺的部分负荷工况开始的飞逸过程中，容易出现流态失稳；随着机组转动部分转动惯量的增大，水泵水轮机工作点改变的速度减慢，水泵水轮机在不稳定的S形特性区的运行时间增长，更容易出现转轮流态失稳；无论是由流态不佳的部分负荷工况开始的飞逸过程，还是工作点改变更慢的飞逸过程，都容易在转轮叶道内形成不均匀分布的失速团并可能形成旋转失速。形成不均匀分布的失速团是转轮流动失稳的典型标志之一，也是导致水泵水轮机一系列水力特性突变的根源。另外，本项目还对抽水蓄能系统两种极端甩负荷过渡过程工况——“一管—两机—一隧”系统两机同时甩负荷和甩负荷过程中由空化导致的水柱分离——的转轮流态演化过程进行了研究，相关发现可更新现有对水泵水轮机过渡过程工况动态特性的认知，也可为抽水蓄能电站极端工况下的安全性评估提供更多指导。