【作者】 徐晨光；

【导师】 黄强； 田峰巍；

【作者基本信息】 西安理工大学 ， 水文学及水资源， 2004， 博士

【摘要】 作为水电站运行管理的重要组成部分，水电站厂内经济运行能够增加水电站的经济效益，这部分效益增加对于我国这样的水电大国和能源供需矛盾日渐突出的形势非常有意义。随着电力市场化进程的加快，“厂网分开，竞价上网”的电力市场运营机制的实施，发电企业成为了竞争主体，水电厂在参与市场竞争的过程中，必须根据自身与竞争各方的运行特性，扬长避短，提高自身的竞争能力，适应电力市场改革，并要求水电站厂内经济运行从理论走向实用。 本论文通过研究水电站厂内经济运行的关键技术问题，运用新方法和新技术建立水电站厂内经济运行系统理论体系与模型体系，探讨了水电站厂内经济运行准实时系统的设计与开发技术，为其商业化奠定基础。论文的主要研究内容和取得的成果有以下几方面： (1) 提出一种空耗水量的简单计算方法。针对承担电网调峰任务的水电站，西安理工大学博士学位论文根据电网所需备用容量任务和水电站的实际运行情况，简化空耗水量的计算方法，其关键是确定合理备用Nx、最大可调出力坑ox和单机可调出力Nn，简化算法使空耗水量计算简便可行。实例计算出龙羊峡水电站在2001年，仅由于不合理调度造成的空耗水量为3.01亿m，，折算电量为7462万kwh，空耗水量.占到该年出库水量162.96亿m3的一540，0，折算电量占当年实际发电量(40.1226亿kwh)的一860，0，按国家1990年颁布的西北电网平均影子电价0.194元/kwh计算，经济损失达1447.628万元，相当可观。 (2)提出并建立了水电站厂内经济运行的准实时系统理论体系。从水电站厂内经济运行的特点和任务时限要求，提出了设计水电站厂内经济运行准实时系统的思路和方法。从水电站厂内经济运行过程和影响因素分析，指出实现水电站厂内经济运行准实时系统需要重点解决数据采集的速度和精度与负荷分配的速度问题。数据采集的速度与精度主要由机组效率曲线计算和机组耗流量在线计算解决，负荷分配问题需引入如进化算法等新方法解决。厂内经济运行系统的开发与设计则需要采用统一建模语言(UML)和嵌入式开发模式进行设计和开发。因此，本文建立了较为完善的水电站厂内经济运行的准实时系统的理论、模型和方法。 (3)建立了基于RBF神经网络模型的水轮机组的效率曲线和耗流量计算模型。采用高效的径向基函数神经网络(RBF)进行水轮机组效率曲线的快速计算，通过与BP网络对比，该RBF神经网络模型具有全局收敛性，网络结构易于确定，训练速度快，计算结果更接近机组实际情况，证实该方法是可行的。本文提出跳过水轮机组效率曲线计算，直接计算水轮机组耗流量的方法禾!神经网络模型，对于水轮机组耗流量的在线计算问题，采用RBF神经网络模型，建立具有自学习功能的计算模型，采取样本的滚动采集、更新，实现在线训练与在线应用，实现耗流量的在线计算、在线应用。水轮机组效率曲线计算和耗流量在线计算为负荷的快速准确分配提供基础条件，有助于厂内经济运行准实时系统的实际应用。 (4)针对不同类型水电站厂内经济运行建立不同的进化算法模型。本文选择了三种在其他优化领域应用较好的方法:遗传算法、进化规划和免疫规划方法，通过建立相应的模型求解厂内经济运行负荷分配问题。实例研究表明，对于中小型水中文摘要电站的厂内经济运行，可以采用动态规划法或进化算法作为负荷分配求解方法;对大型水电站则需选择进化算法作为负荷求解方法。径流式水电站是在来水一定的条件下，以水电站总出力最大为目标进行负荷分配，在其计算过程中，遗传算法在计算速度和优化结果方面显示出优越性;有调节能力的水电站是在完成负荷任务的条件下，以水电站总耗流量最小为目标进行负荷分配，在其计算过程中，免疫规划在计算结果方面显示出优势。 (5)建立了水电站厂内经济运行系统设计禾l开发模式。本研究引入统一建模语言(UML)进行嵌入式厂内经济运行准实时系统的设计和开发，以模型图方式完成系统的需求分析、详细设计与代码实现几个阶段，基于UML的系统开发易于维护与编译。通过探讨嵌入式厂内经济运行准实时系统的开发过程，为未来水电站自动化系统的发展奠定理论基础。关键词:进化算法，水电站，厂内经济运行，空耗水量，径向基神经网络模型，准实时系统，嵌入式实时系统交更多还原

【Abstract】 As the importance of hydropower plant operation and management and the conflict between energy supply and demand is considered, the Economical Operation of Hydropower Plant (EOHP) is becoming dramatic for the reason of increasing the economic performance of hydropower plant. With the opening of power market and the separation of power plant and power networks and the implementation of electricity price bidding, the power generation enterprise will participate in the competition. Therefore, hydropower plant must improve the competition ability to suit the reform of electricity market. Moreover, the reform of electricity market requests the economicaJ operation of hydropower plant used widely.This thesis tries to study several technical problems of EOHP, and to rebuild its’ theory system and model system by using new theory and methods. The study of the design and development technique of quasi real-time EOHP prepares for its commercial popularization. The main results are as follows:(1) A calculation method of the spill water owing to improper dispatch for large-scale hydropower plant, which must undertake the spinning reserve of electrical network, was designed and applied. Considering of the operation of hydropower plant and the spinning reserve needed by the electrical network, the spill water may be calculated from the reasonable spinning reserve Nx; the maximum adjustable output of the plant Nmax; and the maximum adjustable output of each unit. The spill water owing to improper dispatch of Longyangxia hydropower plant in 2001 was calculated. The result showed the spill water owing to improper dispatch is remarkable. The spill water is 3.01×108m3, which is 1.84% of the output water for the hydropower plant in 2001. The converted electricity is 7462×104kWh, which is 1.86% of the output electricity for the hydropower plant in 2001. According to the shadow price of Northeast pool 0.194yuan/kWh, the economical lost is 1447.628 Yuan.(2) Quasi real-time EOHP theory system is put forward. The quasi real-time EOHP should be designed according to the characteristic of EOHP and the system deadline. By analyzing the process of EOHP and its’ influence factors, two improvements should to be done to realize quasi real-time EOHP, one is to improve accuracy of data collection, and the other is to increase the speed of load dispatch. The data collection accuracy is depends on the efficiency calculation of water turbine generator units; and the evolutional method should be used to solve the load dispatch problem. At the same time, the EOHP system should be designed and be developed by unified model language (UML) as embedded system. Therefore, this thesis set up the theories, model and method for EOHP systematically.(3) In order to gain the correct state of each water turbine generator units, a radial basis function (RBF) neural network was proposed to calculate efficiency curve and forecast the flow consumption for water turbine generator unit. Compared with BP neural network, the RBFNN avoids the problem of local minimum and raises the learning. Thus, this method is feasible. This thesis tried to calculate the flowconsumption by neural network instead of the efficiency curve. A self-learning RBF neural network was modeled to calculate the flow consumption online with the training sample collected rolling. Applying this method in load dispatch of EOHP can calculate exactly and help to the application of EOHP.(4) Aiming at realizing quasi real-time EOHP of different kinds of hydropower plant, different evolutional method were proposed to solve the load dispatch problem. Three kinds of evolutional method, the Genetic Algorithms, the Evolutionary Programming and the Immune Algorithm as well as Dynamical Programming, were selected to use in the optimization. The case study shows that both the Dynamical Programming and the evolutional methods can be used to solve the load dispatch problem for run-of-river hydropower plant, and the evolutional method should be selected to solve the load dispatch p更多还原