【作者】 张俊；

【导师】 程春田；

【作者基本信息】 大连理工大学 ， 水利水电工程， 2009， 博士

【摘要】 水资源的开发和利用,特别是水电系统资源的开发和利用,是一项综合运用水利水电工程、水文学及水资源和电力系统自动化等多门学科知识的系统工程,涉及洪水预报、水文预报、水库调度、库群补偿发电优化调度等一系列问题。其中,水文预报和水电站(群)优化调度是两个最为核心的问题。及时可靠的水文预报信息能够为调度提供科学依据,是调度成功的基础保障;水电站(群)调度方案则直接关系到资源的优化配置和电网的稳定运行。由于降雨通过产流和汇流形成径流的复杂过程、水力发电的特殊形式、水力元素和电力元素(水头、流量、出力、电量等)之间的动态耦合关系以及梯级水库群之间的水力联系等,导致水文预报和水电站(群)调度成为具有随机性、多维性、多阶段性、非线性、非凸性、离散性等特点的复杂数学问题,也使得它们长期以来成为水电系统中的研究热点和难点。本文依托国家自然科学基金“电力市场环境下省级电网水火电协调竞争优化风险分析方法(50679011)”和教育部高校博士点基金“面向省级电网复杂水火电系统建模方法研究(20050141008)”,以福建电力调度通信中心“福建电网跨流域水调高级应用系统”为工程背景,从模型与方法的角度,对这两个问题开展研究。针对中长期水文预报问题,主要从智能算法耦合和引入数值天气预报信息的角度改善预报效果;针对水电站(群)优化调度问题,主要从智能算法改进的角度研究能够适用于实际工程需求的解决方案。文中将神经网络、支持向量机、蚁群算法、遗传算法、粒子群算法等智能算法及其耦合算法应用到上述两个问题中,取得了一些有价值的研究成果,主要内容包括:(1)提出了基于蚁群优化算法参数优选的支持向量机水文预报模型(ACO-SVM),并将其应用于长期水文预报。SVM建模过程中,选择径向基核函数作为SVM的核函数,利用蚁群算法进行参数优选。以福建省安砂水电站月径流预报为例,对该模型进行建模仿真计算,并与时间序列方法(ARMA)、人工神经网络方法(BP-ANN)所获得的预报结果进行对比分析,结果表明,在拟合精度方面,ACO-SVM模型相比ARMA模型和BP-ANN模型有不同程度的提高,且增幅较大;在模型的泛化性能方面,ACO-SVM模型优于BP-ANN模型。(2)提出了耦合定量降水预报(QPF)的BP神经网络中期水文预报模型。将地区QPF信息转化为流域降雨信息,结合其他预报因子,建立基于多层感知器神经网络的中期水文预报模型。采用自相关函数和交叉相关函数确定预报因子有效阶数。针对常规BP算法收敛速度慢和易陷入局部最优的不足,将网络误差函数的改变量引入权值和偏移值的调整。采用自适应学习速率和自适应动量因子调整策略,改善算法性能。以福建省水口水电站日径流过程的中期预报为例,对该模型进行建模仿真计算,通过与标准BP神经网络模型、ARMA模型所获得的预报结果进行对比分析,表明改进BP算法比标准BP具有优越性,并且,耦合定量降水预报的中期水文预报能够延长预见期,提高预报精度。(3)提出了基于病毒进化遗传算法的水电站优化调度模型,首次将病毒进化遗传算法应用于水资源领域的优化问题。该模型通过引入生物的病毒感染机制改善种群多样性,提高遗传算法的全局寻优能力。以福建省棉花滩水电站的年发电调度为例,进行了建模和求解,分别对丰水、平水、枯水不同典型年进行了优化计算,并将调度结果与标准遗传算法和动态规划方法比较。结果显示,对于各种典型年,病毒进化遗传算法获得的调度结果均优于标准遗传算法,与经典方法动态规划方法获得的结果十分接近,因此,基于病毒进化遗传算法的水电站优化调度模型是可行、有效和优越的。(4)提出了基于混合改进粒子群算法的梯级水电站群优化调度模型。从三个方面进行改进标准粒子群算法:(a)提出一种新的惯性权重系数策略——自适应指数惯性权重系数。(b)借鉴遗传算法中染色体交叉、变异的思想,将其引入粒子的更新策略,提高粒子的多样性。(c)建立粒子精英集合,将适应值高的粒子选拔进入精英集合,用于代替进化过程中适应值低的粒子,实行优胜劣汰。以福建省闽江流域梯级水电站群优化调度为例,建立基于混合改进粒子群优化算法的水电站群长期优化调度模型,计算结果表明,该模型获得的调度结果优于常规粒子群优化算法(PSO),与逐步优化算法(POA)获得的结果达到相当水平,但求解时间却大幅缩短,计算效率大幅提高,是可以应用于工程实践的有效方法。(5)基于Oracle双机平台,采用J2EE架构,使用Java、EJB、Servlet、Web、面向对象等技术,设计并开发了福建电网高级水调自动化系统——中长期水文预报及调度系统,为福建电网水电系统经济运行提供决策支持。重点阐述了系统架构设计、功能设计和程序设计等方面的关键技术问题,并介绍了系统的特色和主要功能模块。最后对全文做了总结,并对有待于进一步研究的问题进行了展望。更多还原

【Abstract】 The development and utilization of water resources,especially the usage of hydropower resources,is a complicated system engineering which refers to the knowledge of several subjects such as Water Resources & Hydroelectric Engineering,Hydrology & Water Resources,Automation of Electric Power System,etc.,and contains many issues like flood forecasting,hydrological forecasting,reservoir operation,multireservoir compensation optimal operation and so on.Among these issues,hydrological forecasting and hydropower station(group) optimal operation are the most significant ones.Reliable and timely hydrological forecasting is the basis of operation,while hydropower station(group) optimal operation is more important which has a direct effect on resources collocation and operation of power grid.Due to the complicated processes from rainfall to streamflow,special form of power generation by hydropower stations,dynamic relationship among water head,discharge. power,and electricity energy etc.,as well as inherent connecting of upstream hydropower reservoirs and downstream reservoirs,both hydrological forecasting and hydropower station (group) optimal operation are characterized by stochastic,multidimensional,multistage, non-linear,non-convex and discrete characteristics,and both of them are still the most difficult issues in hydropower system up to now.On the basis of the National Natural Science Foundation of China(Grant No.50679011) and the Ph.D.Programs Foundation of Ministry of Education of China(Grant No.20060183043),as well as based on the project background of Fujian Power Grid Multireservoir System Advanced Application Software in the charge of Fujian Power Grid Dispatching and Communication Center,this dissertation researched deeply on aforementioned two issues in terms of model and method research.Regarding to the ftrst issue,main focuses are on the combination of algorithms and import of numerical weather prediction information for improving forecasting quality.With respect to the second issue,emphases are on the improvement of intelligent optimization methods for solution of practical engineering requirement.In this dissertation,main focus is on the application of intelligent algorithms including neural networks,support vector machine,ant colony optimization,genetic algorithm,particle swarm optimization,etc.,and their hybrid algorithms in above-mentioned two issues.The main contents and outlines are as follows:(1) An ant colony optimization(ACO) based support vector machine model (ACO-SVM) is proposed for long term hydrological forecasting.SVM algorithm is of reliable global optimality and good generalization,and it is suitable for mid-and-long term hydrological forecasting which contains the study of finite samples.However,the results considerably depend on relevant parameters in SVM and conventional choosing method by experience can not obtain satisfactory outcomes.Radial basis function is selected as kernel function and parameters of SVM are optimized by ACO.An’sha reservoir in Fujian province is selected as a case study to demonstrate the modeling of ACO-SVM and forecasting results are compared with that of conventional ARMA model and BP-ANN model.The experimental results show that ACO-SVM model is much more efficient in global optimization,the forecasting accuracy is better than that of the other two models,and the ability of generalization is superior to BP-ANN.(2) A back-propagation neural network model taking account of quantitative precipitation forecasting(QPF) is proposed for mid term hydrological forecasting.The model is based on multi-layer perceptron network and forecasting factors include basin QPFs transformed from region QPFs.A statistic method which makes use of auto-correlation function(ACF) and cross-correlation function(CCF) is employed to specify the most efficient predictors’ lags which were generally selected by empiricism.The standard back-propagation algorithm is improved by using self-adaptive learning rate and self-adaptive momentum coefficient which import the error function variation into the adjustment of weights matrix and biases matrix.Shuikou reservoir in Fujian province is selected as a case study to demonstrate the modeling and forecasting results are compared with that of conventional ARMA model and standard BP-ANN model.The experimental results show that the proposed model outperforms ARMA and standard BP-ANN model,and the utilization of QPF information can prolong lead-times as well as improve forecasting accuracy.(3) A virus evolution genetic algorithm(VEGA) based hydropower station operation model is proposed and this is the first time that VEGA is used for the optimal problems in the field of water resources.Virus infection mechanism is employed by VEGA for enhancing population diversity of GA,so as to improve the global searching ability of GA.Mianhuatan reservoir in Fujian province is selected as a case study to demonstrate the modeling of VEGA and operation results are compared with that of standard GA model as well as dynamic programming(DP) model.The experimental results show that the proposed VEGA model outperforms standard GA,in cases of different typical year,and the annual electricity energy are all close to that of classical DP model.Therefore,VEGA based hydropower station operation model is feasible,effective and advanced.(4) A hybrid improved PSO algorithm(HIPSO) is proposed for hydropower station group optimal operation.The improvement of PSO contains three aspects:(a) A new updating strategy of inertia weight coefficient-Self-adaptive Exponential Inertia Weight Coefficient (SEIWC) is proposed to replace the linearly decreasing inertia weight coefficient (LDIWC).(b):The crossover and mutation techniques of chromosome from genetic algorithm are imported into updating adjustment of particles to improve the global searching ability of PSO.(c):Particle elite set is established to reserve particles with better fitness value which are used to replace the bad ones during evolution process.The proposed HIPSO is applied to the optimal operation of hydropower station group of Minjiang basin in Fujian province,using energy maximization as optimized objective function,and results indicate that the HIPSO performs much better than standard PSO and optimal operation results of HIPSO is comparable with that of progressive optimization algorithm(POA) with much shorter time consumption.So,the proposed HIPSO is an effective method in engineering practice for hydropower station group optimal operation.(5) Based on Oracle clustered databases and J2EE framework,a mid- and- long term hydrological forecasting and operation system,the core part of Fujian Power Grid Multireservoir System Advanced Application Software,is deigned and developed utilizing several techniques such as Java,EJB,Servlet,Web,object-oriented and so on for the purpose of providing decision support for Fujian Power Grid.This section focuses on the key techniques of system structure design,function design,programming design,etc.,and the characteristics and major function modules are also introduced.Finally,a summary is given and some problems to be further studied are discussed.更多还原