【作者】 陈潇；

【导师】 张勇传； 周建中；

【作者基本信息】 华中科技大学 ， 水利工程， 2022， 博士

【摘要】 随着我国水电能源集中开发已进入尾声阶段,诸多装机容量大、调峰能力强的巨型水电站现已坐落于祖国的大江大河之上,在防洪、发电、生态等方面发挥了举足轻重的作用。现阶段,风电、光伏等新能源发展势头迅猛,水电能源调峰压力与日俱增,给在运水电站精细化调度运行水平提出了更高的要求,如何在“以电定水”调度模式下精准计算水电站尾水位、出库流量并探明与之匹配的水量联系描述方法,是亟待解决的关键问题。此外,新形势下,在水电站日计划编制和实际调度中考虑分厂约束已是大势所趋,如何在分厂约束限制下制定机组启停与负荷分配策略是另一关键问题。针对上述问题,本文以三峡水电站为主要研究对象,运用水电能源学、人工智能等多学科理论与方法开展了较为深入的研究,取得了若干有价值的成果,可为最终实现梯级水电站短期精细化调度提供模型与方法支撑。本文主要的研究内容和创造性成果如下:(1)针对水电站尾水位计算精度问题,分析了三峡尾水位的非线性特征和后效性演变规律,构建了基于水电站监测数据的尾水位神经网络模型,对比了尾水位经验公式、顶托曲线以及尾水位神经网络模型在三峡尾水位单步和滚动计算中的精度差异。研究表明:经验公式和顶托曲线具有相似的计算精度,且因其输入参数未知而在“以电定水”调度模式下存在实用性不足的问题。相比较而言,以三峡水电站前一至前三时段库水位和尾水位、以及当前至前三时段三峡出力为输入的尾水位神经网络模型,能在一定程度上兼顾模型的计算精度和实用性,可为三峡尾水位精准计算提供模型支撑。(2)针对水电站出库流量计算精度问题,提出了以水电站毛水头为参数的负荷分配表编制方法,结合现行的日计划编制业务流程,分别从水电站发电流量最优、水电能转换特性以及数据挖掘的角度提出了七种出库流量计算方法,对比了上述方法在三峡出库流量单步和滚动计算中的精度差异。研究表明:与尾水位经验公式相比,尾水位神经网络模型能在一定程度上提高三峡出库流量计算精度,且对出库流量计算偏差不敏感,具有较强的鲁棒性。另外,具有与尾水位神经网络模型相同输入因子的出库流量神经网络模型,能够在一定程度上克服巨型水电站机组负荷分配策略难以总结的困难,是三峡出库流量计算的首推模型。(3)针对三峡出库流量与葛洲坝入库流量间水量联系精细化描述问题,分析了出入库流量历史运行数据的对应规律,构建了入库流量神经网络模型,设计了葛洲坝入库流量滚动计算敏感性分析实验,检验了出库流量神经网络模型与不同水量联系描述方法的匹配效果。研究表明:汛期期间与非汛期期间,与出库流量神经网络模型相匹配的水量联系描述方法与模型分别为现行的折算系数法、以及以当前和前一时段的三峡出库流量为输入的入库流量神经网络模型。(4)针对分厂约束限制下水电站厂内经济运行建模与求解问题,在分析机组运行特性的基础上构建了考虑分厂约束的厂内经济运行模型,并提出了成套求解方法。研究表明,所提模型及方法能灵活应对三峡水电站母线联络开关问题,有效协调各分厂出力以满足分厂稳定运行区约束,确保机组处于稳定、高效的运行状态,且具备良好的适用性。另外,厂内经济运行模型与出库流量神经网络模型互为补充,与尾水位、入库流量神经网络模型也具有较高的匹配度。更多还原

【Abstract】 As the centralized development for hydroelectric power has entered the final stage in China,many giant hydropower stations epuipped with large capacity and strong peak load regulation capacity are placed on big rivers,and are playing pivotal roles in flood control,power generation,ecology and other aspects.At present,the developments of wind power,photovoltaic and other new energy sources are strong,resulting in a growing demand in peak load regulation by hydropower,which puts forward higher requirements for the refinement dispatching of hydropower stations;how to accurately calculate tail water level and outflow of hydropower stations under given output and explore the corresponding characterization method for the water volume connection are the key problems to be solved.Besides,under new situations,it is a general trend to consider substation constraints when formulating daily plans and conducting actual dispatching of hydropower stations,and how to formulate start-stop and load distribution strategies for units restricted by substation constraints is another key problem.For the above problems,the Three Gorges hydropower station was taken as the main research object,and a series of studies were conducted by multi-disciplinary theories and methods,such as hydroelectric enelgy and artificial intelligence.Some valuable results were obtained,which could provide model and method supports for ultimately short-term refined dispatching of cascade hydropower stations.The main research contents and creative achievements of this paper were as follows:(1)For the calculation accuracy problem of tail water level,the nonlinear characteristics and aftereffects for tail water level of Three Gorges were analyzed,and the artificial neural network(ANN)models based on monitoring data for tail water level were established.The accuracy differences among the empirical formula,the jacking curve and ANN models for tail water level were compared in single period and rolling calculation.Results showed that the empirical formula and the jacking curve owned similar calculation accuracy,and showed poor practicability caused by their unknown input parameters in the dispatching mode under given output.In comparison,the ANN model for tail water level,whose parameters were the forebay and tail water level of Three Gorges during the last three periods,and output of Three Gorges from the last three periods to the current period,could combine the calculation accuracy with practicability to a certain extent;it could provide model supports for accurate calculation of tail water level of Three Gorges.(2)For the calculation accuracy problem of outflow,a method was proposed herein for formulating the optimal load allocation table with gross head as parameters.According to the current processes for formulating daily plans,seven kinds of methods for outflow were proposed from aspects of optimizing power generation streamflow,water-energy conversion characteristics and data mining respectively,and their accuracy differences in single period and rolling calculation were compared.Results showed that compared with the empirical formula,the ANN model for tail water level could slightly improve the calculation accuracy of outflow,and showed strong robustness to the calculation deviation for outflow.Besides,the ANN model for outflow,which owned the same input features as the ANN model for tail water level,could overcome the difficulty in summarizing the load distribution strategy of giant hydropower stations to a certain extent,and was the first choice to calculate outflow of Three Gorges.(3)For the problem of characterizing the the water volume connection between outflow of Three Gorges and inflow of Gezhouba,the corresponding rules in historical operation data between outflow and inflow were analyzed,and the ANN models for inflow were established.Then,sensitivity analysis experiments about rolling calculation for inflow of Gezhouba were designed,and the matching effects between the ANN model for outflow and different water volume connection characterization methods were checked.Results showed that during the flood and non-flood season,the water volume connection characterization method and model matching with the ANN model for outflow were the current convert coefficient method and ANN model for inflow respectively,whose parameters the outflow of Three Gorges from the last one periods to the current period.(4)For the problem of modeling and solving for the internal economic operation restricted by substation constraints,the internal economic operation model considering substation constraints was established on the basis of analyzing unit operation characteristics,and a set of solving method was proposed.Results showed that the proposed model and method could flexibly handle the bus connection problem,and effectively coordinate output of substations to meet the constraint of substation stable operation zones;simultaneously,the model and method could ensure that units were in a stable and efficient operation state,and owned a good applicability.Besides,the internal economic operation model and the ANN model for outflow complemented each other,and owned a high degree of matching with the ANN model for tail water level and inflow.更多还原