【作者】 王文川；

【导师】 程春田；

【作者基本信息】 大连理工大学 ， 水文学及水资源， 2008， 博士

【摘要】 随着水电建设的快速发展和工程规模的不断拓宽,水电系统中的问题规模和复杂度也越来越大,表现出高维、非线性、非凸等复杂特性,利用传统方法和单一智能方法对问题进行求解时结果往往不够理想,鉴于这种现状,算法混合的思想就成为提高算法性能的一个重要而有效的途径。因此根据研究问题的特点,如何有机结合各种算法的优点,构造更高效的算法,对提高水文预报的精度及水电系统经济、高效运行具有重要意义。本文首先系统介绍了概念性水文模型参数优选及不确定性分析、中长期水文预报、水电站(群)优化调度的研究现状,结合湖南双牌水库、云南澜沧江流域和贵州乌江流域的水电站(群)等工程,综合利用遗传算法、混沌系统、人工神经网络、自适应模糊推理、遗传程序设计、支持向量机等多种人工智能技术,深入研究了这些方法及其混合在概念性水文模型参数优选、概念性水文模型参数不确定性分析、中长期水文预报、水电站(群)优化调度的建模方法及应用,取得了一些有价值的研究成果。主要内容如下:(1)以湖南省双牌水库2000-2006年水情自动测报系统积累的48场1小时洪水过程资料为基础,预报模型采用三水源新安江模型,提出三水源新安江模型参数率定的模糊多目标优化混合启发式遗传算法。该算法应用混沌变量生成初始种群,应用退化混沌变异操作代替标准的变异操作,应用SA技术提供局部邻域搜索,从实际应用出发,采用洪峰流量、峰现时间和洪水总量的合格率作为场次洪水模拟的评价目标,对三水源新安江模型参数进行模糊多目标优选,模拟和检验的比较结果表明,提出的方法能较好地获得短期洪水预报模型参数。(2)应用概念性水文模型进行水文模拟时,由于模型本身的不足及参数多、信息量少等原因,会出现率定的最优参数组不唯一,不稳定等问题。考虑到以往的参数优选,都只得出一个参数组,无法评定优化所得参数组的不确定性的影响,存在一定的片面性和局限性,使模型的实际应用受到限制。针对这一问题。提出应用基于马尔可夫链蒙特卡罗(MCMC)理论的SCEM-UA算法,通过双牌流域不同历时的典型洪水数据对新安江模型参数进行随机优选和不确定性评估。结果表明,该算法能很好地推出新安江模型参数的后验概率分布;率定和检验结果比较分析也表明,应用SCEM-UA算法对新安江模型进行优选和不确定评估是有效和可行的。(3)对比研究了人工神经网络(ANN)、自适应模糊推理系统(ANFIS)、遗传程序设计(GP)和支持向量机(SVM)在径流中长期中的预报建模。为了评价它们在月径流时间序列的预报效果,采用了几个标准的执行评价措施,如相关系数(R)、确定性系数(E)、均方误差(RMSE)和平均绝对百分误差(MAPE)。应用了时间序列分析建模技术ARMA作为参考的标准。通过漫湾水电站52年月径流系列及洪家渡水电站54年的月径流系列预测结果比较分析表明人工智能技术是个强有力的工具,能获得比传统时间序列分析方法更好的预报精度。分析结果也表明根据不同的统计评价指标,ANFIS、GP和SVM分别能获得最好的预报结果。这说明ANTIS、GP和SVM相对于ARMA和ANN能提高月径流时间序列预报的精度。(4)利用混沌运动的遍历性、随机性和规律性等特点,提出了一种求解水电站水库优化调度问题的基于浮点数编码的混沌遗传(CGA)算法。该算法的思想是采用混沌优化进行改善初始种群质量和利用混沌退化变异算子代替常规算法中的变异算子,避免搜索过程陷入局部极值。利用两个著名的测试函数对提出的算法进行验证,分析结果表明提出的混沌遗传算法不但具有传统遗传算法的全局多点搜索,占用内存少等优点,而且还较好地克服了传统遗传算法的易“早熟”、易陷入局部最优和停滞的问题。通过对典型径流水电站优化调度,长系列历史径流资料的优化计算和梯级水电群的优化调度,并与传统方法相比,结果说明了该算法能提高计算结果,并具有计算速度快,搜索效率高,收敛性能好等优点。这说明提出的方法对求解复杂的水库优化调度问题是有效的和可行的。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】 With rapid development of hydropower construction and project scale widened constantly, the scale and complexity of problem is becoming bigger and bigger in hydropower system and posses properties of multi-dimension, non-linear, non-convex and etc. This is a reason that results are not ideal through the traditional method and single intelligence method to solve, and the idea of hybrid algorithms is very important way for improving effective and efficient algorithms. Therefore, how to efficiently utilize the advantages of various algorithms to develop more efficient algorithm is very significant to improve hydrological forecasting accuracy, economical and effective operation of hydropower system. In this dissertation, first, the conceptual rainfall-runoff (CRR) model parameters calibration and uncertainty analysis, the long-term forecasting and operation of hydropower system are reviewed in detail. Then, based on the project background of hydropower station(s) in Hunan Shuangpai reservoir, Yunan Lancangjiang River basin and Guizhou Wujiang River basin, the hybrid intelligence methods are studied for CRR model parameters calibration and uncertainty analysis combined with case of project application. The dissertation studies long-term forecasting and optimization operation of hydropower system using Artificial intelligence techniques such as genetic algorithms (GA), chaos system, artificial neural networks (ANNs) approaches, adaptive neural-based fuzzy inference system (ANFIS) techniques, genetic programming (GP) models, support vector machine (SVM) method and hybrid intelligence optimization methods. The major research work is outlines as follows:(1) The hybrid metaheruistic genetic algorithm with fuzzy multiobjective optimization is presented for solving the Xinanjiang model parameters based on 48 historical floods from hydrological telemetry system with one hour routing period for 7 years (2000-2006) in Shuangpai Reservoir. The proposed method takes advantages of the ergodic and stochastic properties of chaotic variables, an annealing chaotic mutation operation which is employed to replace standard mutation operator in the evolutionary process of GA and the local optimal search capability of SA method. The three statistical ratios of acceptable criteria relative to the peak discharge, peak time and total runoff volume among the calibrated and validated historical flood events, are used to evaluate the parameter calibration performance for rainfall-runoff model by fuzzy multiobjective optimization from actual application. The results of calibration and validation indicate that the proposed method can obtain better model parameters for short-term flood forecasting.(2) While application Xinanjiang model to simulate hydrograph, the "best" parameter set calibrated may be not unique and uncertain because of model limitation, more parameters and limited information. Considering previously parameter optimization of Xinanjiang model, there is only a unique "best" parameter set to be found and it doesn’t describe uncertainty of parameter. There is a certain one-sidedness and limitation for Xinanjiang model used. Aiming at this problem, this paper presents using SCEM-UA algorithm based Markov Chain Monte Carlo (MCMC) methods for optimization and uncertainty assessment of Xinanjiang model parameters with different routing periods for Shuangpai basion. The results demonstrate that SCEM-UA algorithm is well suited to infer the posterior distribution of Xinanjiang model parameters. The results of calibration and validation indicate that it is a feasible and effective for optimization and uncertainty assessment of Xinanjiang model parameters.(3) Developing a hydrological forecasting model to apply Artificial intelligence (AT) technology which include artificial neural networks (ANNs) approaches, adaptive neural-based fuzzy inference system (ANFIS) techniques, genetic programming (GP) models and support vector machine (SVM) method. The main purpose is to investigate the performance of several AI methods for forecasting monthly discharge time series. The four quantitative standard statistical performance evaluation measures, the coefficient of correlation (R), Nash-Sutcliffe efficiency coefficient (E), root mean squared error (RMSE), mean absolute percentage error (MAPE), are employed to evaluate the performances of various models developed. The autoregressive moving-average (ARMA) model is also employed as reference benchmark. The results obtained in this study indicate that the AI methods are powerful tools to model the discharge time series and can give good prediction performance than traditional time series approaches through 52 years discharges series in Manwan hydropower and 54 years discharges series in Hongjiadu hydropower. The results indicate that the best performance can be obtained by ANFIS, GP and SVM in terms of different evaluation criteria. The results of the study are highly encouraging and suggest that ANFIS, GP and SVM approaches are promising in modeling monthly discharge time series comparation with the ANN and ARMA.(4) By use of the properties of ergodicity, randomicity, and regularity of chaos, a chaos genetic algorithm (CGA) based float encoding is proposed to solve optimal operation of hydropower reservoir. CGA adopts chaos optimization of the initialization to improve species quality and utilizes annealing chaotic mutation operation to replace standard mutation operator in order to avoid the search being trapped in local optimum. Comparison of results among the dynamic programming, the standard GA and CGA showed that CGA can significantly reduce the overall optimization time and improve the convergence quality through complex function optimization, hydropower station optimization operation with typical annual runoff, hydropower reservoir optimization operation with a series of monthly inflow and cascaded hydropower optimization operation. The analysis results show that CGA has obvious advantages in convergence speed and solution quality. Thus, the proposed approach is feasible and effective in optimal operations of complex reservoir systems.Finally, a summary is given and some problems to be further studied are discussed.更多还原