【作者】 覃晖；

【导师】 周建中；

【作者基本信息】 华中科技大学 ， 水利水电工程， 2011， 博士

【摘要】 流域梯级水库群的联合优化调度受气象循环、水文过程、发电控制、电网潮流、用水需求为等多种因素制约,是一类高维度、多目标、强耦合的复杂非线性约束优化问题,一直是水电能源科学与复杂性科学交叉发展的前沿研究领域之一。随着我国水电能源的持续大规模开发,流域梯级水电能源系统规模与拓扑结构日趋复杂,大型水利枢纽综合利用要求也晶益提高,从而给梯级水电站群的联合优化调度带来巨大挑战。流域梯级电站间复杂的水力、电力耦合联系以及不同调度目标间的竞争与冲突关系,使得传统优化调度理论与方法已无法满足大规模梯级水电能源系统整体优化与水能资源高效利用的需求,亟需研究并发展新的优化理论与方法。本文围绕水电能源大规模开发背景下梯级水电站群的联合优化调度决策问题,以梯级水电能源安全和高效利用为目标,结合系统工程理论、现代智能优化理论和多目标决策理论,研究了梯级水电站群联合优化调度决策的先进理论与方法,解决了制约优化理论在复杂水电能源多目标优化调度实际工程应用中的关键技术瓶颈,取得了一些有理论意义和工程实用价值的研究成果。本文的主要研究内容和创新包括：(1)针对流域梯级电站群联合优化调度决策变量多且相互耦合的特点,提出了适应于多变量耦合优化问题的文化差分进化算法。该算法从群体空间和知识空间的构造出发,以差分进化算法为群体空间优化驱动机制,并在知识空间中定义3种知识结构,以指导种群的进化过程,避免了算法早熟收敛,提高了算法全局搜索能力；文化差分进化算法应用于梯级水电能源系统的实例研究结果表明,该算法能有效处理复杂约束条件且计算速度快,为梯级水电站群联合优化调度实际工程问题的求解提供了一种有效的新方法。(2)为有效求解多目标优化问题,结合多目标优化理论与方法对文化差分进化进行改进、拓展和完善,首先提出了一种多目标文化差分进化算法。将优化方案的形势知识结构视为外部档案集,提出一种“μ+1”循环选择方式对形势知识结构进行更新和维护,提高了非劣解在非劣前沿的分布均匀性；进而运用Pareto支配原理对差分进化算法选择操作进行修正,使其能够处理多目标优化问题；数值仿真计算结果及其与多种流行多目标优化算法的对比分析表明,多目标文化差分进化算法能有效处理非凸、非连续、多模态的多目标约束优化问题,且非劣解集均匀分布于真实非劣前沿,是一种性能优异的多目标优化方法。(3)通过分析梯级电站群各调度目标间的竞争与冲突关系,研究并建立了均衡考虑上下游及大坝防洪安全的水库多目标防洪优化调度模型以及兼顾梯级电站发电效益和容量效益的多目标发电调度模型,并运用提出的MOCDE算法对模型进行求解,获得一组满足实际工程需求的非劣调度方案集,解决了梯级电站群多目标优化调度方案快速生成的难题,为三峡梯级防洪、发电调度决策提供了数据支撑；此外,研究了水火电混联系统的短期多目标优化调度问题,提出了系统负荷平衡约束处理的动态调整策略,仿真研究结果验证了MOCDE的合理性和高效性。(4)研究并发展了梯级水电能源系统风险分析和多属性风险决策的理论与方法。将预报入库径流拟合为服从正态分布的白噪声随机变量,运用蒙特卡洛模拟方法研究了径流不确定条件下梯级电站群的防洪、发电风险分析问题,并将最高水位、最大下泄流量、末水位、年发电量、最小出力等调度目标值描述为有限区间内服从正态分布的随机变量,建立了含有随机变量的风险型多属性调度决策模型；提出了基于相对优势可能度和综合赋权的多属性风险决策方法,突破了传统确定性决策方法难以有效处理随机决策变量的瓶颈；该方法运用主客观综合赋权的方法确定属性权重,并改进了随机变量相对优势度的计算方法；最后通过三峡梯级的实例研究,验证了风险分析模型和多属性风险决策方法的有效性。更多还原

【Abstract】 With the constraints of hydrological cycle, generation control, network power flow and water supply, the optimal operation of cascade hydroelectric stations is a high-dimension, multi-objective, strong-coupling, nonlinear optimal problem, and it is also a hot interdisciplinary research field of hydroelectric science and complexity science. Nowadays, with the continued fast exploitation of river basin hydroelectric energy, the scope and topology of cascade hydroelectric stations increases day by day, which bring great challenges to the joint operation of cascade hydroelectric stations. The traditional optimization theories and methods can not satisfy the demands of joint optimal operation of large-scale hydroelectric energy systems and high efficient utilization of river basin water resources. Therefore, it is quite necessary to study interrelated new optimal theories and decision-making methods. In this thesis, concering on the joint optimal operation problems of cascade hydroelectric stations, we deeply study the joint optimal operation theories and decision-making methods, by adopting the system engineering theory, modern optimization methods based on swarm intelligence, and multi-objective decision-making theory. Some conclusions with theoretical and practical value are obtained. The main conclusions of research and innovation are as follows:(1) Considering that the joint optimal operation problems of cascade hydroelectric stations have a large number of variables and these variables are usually coupled, we present a cultural differential evolution (CDE), which is suited for the optimal problems with multiple coupled variables. CDE consistes of two parts:population space and belief space. In population space, differential evolution (DE) is adopted as the searching engine of the population, and in the belief space, three knowledge structures are defined to guide the evolution process of the population, which can avoid premature convengence and improve the algorithm’s global searching ability. CDE is tested by several benchmark functions and applied to the optimal operation problems of a hydropower system with 10 stations and the Three Gorges-Gezhouba cascade stations. The obtained results show that CDE can handle the complex constraints efficiently and has fast convergence rate and high convergence precision, thus it is an efficient method for joint optimal operation of cascade hydroelectric stations.(2) According to the characteristics of multi-objective optimization (MOO), we extend CDE to solve multi-objective problems (MOPs) and present a multi-objective algorithm named multi-objective cultural differential evolution (MOCDE). In MOCDE, the situational knowledge is used as archive set and an update strategy based on iterative "μ+1" selection is presented to improve the diversity of the solutions in archive set. Meanwhile, the select operator of DE is modified to suit for MOPs, according to the Pareto dominace principle. MOCDE is tested by a series of widely used multi-objective benchmark functions and it is compared with several popular multi-objective optimal algorithms, the results show that MOCDE can effectively deal with complex MOPs with the characteristics of non-convex, discontinuous and multi-modal, and it is an efficient solver for MOPs(3) Besed on analyses of the competition and conflict relation among different objectives, a multi-objective flood control operation model and a multi-objective annual power generation model are established, the former considers the flood control demands of upstream and downstream areas, and the latter considers the cascade stations’ generation benefit and capacity benefit. Then we use MOCDE to solve these two multi-objective models, the results show that MOCDE can get a set of non-dominated operation schemes in a single run and provice alternatives for decision makers. Furthermore, MOCDE is used to solve multi-objective short-term optimal operation of a hydrothermal power system. A dynamic adjustment strategy is presented to handle power load balance equivalence constraints. The results validate the rationality and availability of MOCDE.(4) Risk analysis and multi-attribute risk decision making theory and methods of cascade reservoir systems are deeply researched. Considering the uncertainty, the natural inflow is described as a stochastic forecast-dependent white noise process, and then we analyse the flood control and power generation risk of cascade stations. The operation objectives, such as the maximum water level, the maximum discharge, the final water level, the annual power production, the minimum power output, and so on, are described as normal distribution stochastic variables. To solve the decision making problems with random variables, a risk decision making method based on superiority possibility and comprehensive weights assignment strategy is presented. This method calculates attribute objective weights according to the deviation of decision variables, and then integrates attribute objective weights and subjective weights to get comprehensive weights. After that, a relative dominance matrix is constructed by comparing each two schemes, and finally we can sort the schemes and pick out the best scheme. Case studies of the Three Gorges Cascade Stations show the validity of the proposed risk analysis and decision making methods.更多还原