【作者】 徐莹；

【导师】 程春田；

【作者基本信息】 大连理工大学 ， 水文学及水资源， 2014， 硕士

【摘要】 径流预报和水电站(群)调度是水电系统优化运行的关键问题。准确的长期径流预报可以为制定合理可靠的调度方案提供保障；而及时准确的中期径流预报可以针对不同的天气类型(如台风、暴雨等)和节假日特殊用水需求制定水库的蓄泄计划,对防汛抗旱和水资源的合理配置有着非常重要的意义。结合预报的中期调度能充分考虑极端天气(如高温天气)用电需求和工作日节假日用电差异,发挥水电机组运行灵活的特性充分削峰,使预留给火电的负荷过程尽量均匀,确保电力系统安全平稳运行。水文预报调度由于考虑因素众多、涉及目标复杂、求解过程困难,往往无法直接进行求解,需要深入分析问题特点,研究求解效率高的实用化方法。近年来,智能算法跳出传统模型需明确目标问题机理成因的束缚,以其强鲁棒性、自学习性、非可微要求等优势获得空前发展。本文通过深入分析中长期径流预报调度的特点,以南方地区水电系统为研究背景,结合智能算法进行了深入研究。具体工作如下：(1)针对长期径流预报中支持向量机寻优慢、精度低的问题,用遗传算法对支持向量机的参数进行寻优,建立了基于遗传算法寻参的支持向量机模型,应用于乌江流域某电站,与网格寻参的支持向量机模型进行对比分析。结果表明,前者无论在训练期还是检验期,各项评价指标都要好于后者,说明利用遗传算法进行参数寻优是合理可行的。(2)针对中期径流预报中BP模型非线性处理能力较差、需不断调整权重的问题,构建了基于广义回归神经网络的径流预报模型,结合降雨径流资料,对棉花滩电站91天的日径流进行预报,与BP模型进行对比。通过比较,广义回归神经网络较BP网络拟合能力和泛化能力都有所高,是一种有效的预报方法,可以用于水电站中期径流预报。(3)针对中期水电调度过程期望预留给火电站的余荷过程尽量平稳,构建了调峰出力最大模型。深入研究了基于遗传算法的调峰出力最大模型,确定了梯级电站的计算顺序,采用实数编码方式按顺序对各电站水位进行编码,通过选择、交叉、变异等遗传操作得到最佳水位运行序列。最后给出水电中期调度的调峰结果和流域内各主要电站的水位出力过程线。实例表明本方法可以有效安排水电站群的出力过程,保证预留给火电站的余荷过程尽量平稳。更多还原

【Abstract】 Runoff forecasting and reservoir group is a key issue for optimal operation of hydropower systems. Accurate long-term runoff forecasting can provide a reliable guarantee for the draft of operation scheme. Timely and accurate mid-term forecasting can formulate storage and discharge plans of reservoirs for different weather types (such as typhoons, storms, etc.) and holiday special water needs, which has a very positive meaning for controlling flood and drought, as well as rational allocation of water resource. Mid-term operation combined forecasting which considers electricity demand during extreme weather (such as hot weather) and electricity demand differences between weekdays and holidays, clip peak fully rely on characteristics of hydropower plant for flexible operating and strong climbing ability, which can reserved for thermal power load as evenly as possible and ensure the safe and stable operation of the power system. Usually, hydrological forecasting and operation can’t be directly solved due to many considerations and complex targets, so analysis of the problem characteristics and high efficiency of practical methods for solving models are necessary. In recent years, intelligent algorithms out of the bondage of mechanism and causes need to be clear at traditional models, are developed unprecedented due to strong robustness, self-learning and non-differentiable. This paper analysis of features at mid-long-term runoff forecasting and operation, study the south hydroelectric system combined with intelligent algorithms. Details are as follows:(1) For SVM optimized slow and low accuracy in long-term runoff forecasting, using GA for parameters optimization, SVM is established based GA, the model used the first stage of cascade stations developed on Wujiang River, comparative analysis with the grid-search SVM. The results showed that the former are better than the latter, indicating that the use of GA for parameter optimization is reasonable and feasible.(2) Considering poor nonlinear processing capacity and adjusting the weights constantly in BP, GRNN model is proposed and be applied to day-runoff forecasting in Mianhuatan hydropower station. Compared with BP, GRNN has a better fitting and generation ability, which is an effective forecasting method can be used in mid-term runoff forecasting.(3) Building the maximum output peaking model makes sure power plant having a steady load. Studying the maximum output peaking model based on genetic algorithms, calculating to determine the order of cascade hydropower stations, coding sequence encodes each station level with real number, solve the model through genetic manipulation, selection, crossover and mutation.Finally giving peaking results, water level curve and generated output curve of main power plants in the basin. Examples show that this method can effectively contribute to arrange the process of hydropower stations to ensure that power plant has a smooth process of the remaining load.更多还原