双超产流模型参数敏感性分析与率定

发布时间：2018-11-08 15:34

【摘要】：随着洪水预报科学的发展,水文模型已被广泛地用来解决包括水文水资源、环境和生态等社会和人类发展问题。南方湿润地区雨洪配套资料丰富,在进行流域水文模型研究时可做的工作较多,计算方法也比较成熟。而对于占我国领土面积52%的半干湿地区,水文模型方面的工作做的较少,流域水文模型的使用在国内外在半干湿地区都存在问题。双超模型尤适用于半湿润半干旱地区,作为洪水预报首选模型,应给予充分重视,而现阶段对双超模型参数敏感性以及率定的研究贫乏,不能为各级政府和防汛部门提供准确的模型信息,难以满足洪水预报参数率定的需求。本文以双超产流模型为研究对象,对模型参数进行敏感性分析,识别模型输出响应的重要影响参数,减少模型参数率定过程中的盲目性,并且建立山西省小流域洪水分类参数率定系统,为参数率定提供依据,提高模型运行的可靠性与预报精度。本文以山西省内榆社、上静游与娄烦3个水文站控制流域作为研究对象,选取各流域内具有代表性的场次洪水进行双超模型参数敏感性分析。首先采用局部分析法,得出了双超模型参数在不同流域、不同等级洪水及多个目标函数下的敏感性与相关性情况,基于变异系数法确定模型参数的综合敏感性系数。再采用优化的LH-OAT法,通过对参数的定向改变,得出了双超模型参数在不同流域、不同等级洪水及多个目标函数下的敏感性与相关性情况,基于熵值法确定了模型参数的综合敏感性系数。并将两种方法所得结论进行对比分析,研究表明:(1)由局部分析法得到模型参数综合敏感性大小排序为Srbα0Ksσ≈C,参数Sr、Ks、b、α0为敏感性参数,C、σ为不敏感参数,由全局分析法得模型参数综合敏感性大小排序为KsbSrα0σ≈c,参数Sr、Ks、b为敏感性参数,参数α0为较敏感参数,C、σ为不敏感参数。对比二者成果可知不同的研究方法的到模型参数的敏感性大小排序有所不同。但是对于参数敏感性分级,仅α0受到分析方法的影响,其他参数敏感性等级具有较好的稳定性。(2)采用局部分析法与全局分析法对参数与目标函数相关性分析可知,在不同等级洪水、不同流域中,各敏感性参数与目标函数Wi、Qmi的相关性明确。表现为参数α0、b与Wi、Qmi正相关,参数Sr、Ks与Wi、Qmi呈负相关。但各参数并不是对所有的目标函数都有明确的相关性,当目标函数变为IVF、RE、RSS、PE时,相关性不明确。因此,有在实际运用中针对不同目标函数,在参数的调节方面需要区别对待,并不是都有规律可循。本文采用模糊ISOD ATA迭代模型对历史洪水进行聚类分析。因场次洪水过程的洪峰流量和洪水总量为洪水预报的主要目标,所以选定历史洪水的洪峰流量和洪水总量为聚类特征指标进行聚类分析。将历史洪水按照量级大小分为大洪水、中洪水、小洪水3种类型。由于洪水现象复杂多变,难以掌控,产汇流规律在不同类型洪水中也不尽相同,为降低仅用一组水文预报模型参数对全流域洪水进行预报的误差,本文确立了水文预报模型参数分类率定的思路,来寻找同类型洪水产汇流的规律。水文预报模型参数分类率定结果表明:(1)本文所建立BP神经网络分类模型,可准确判断流域洪水所属类型,在样本预测中精度达到100%。(2)本文所建立的流域洪水分类预报方法,将传统洪水预报的洪量合格率从73%提高到了82%,洪量相对误差从18.1%减少到了11.3%;洪峰合格率也从73%提高到了82%,洪峰相对误差从16.4%减少到了14.6%。提高了研究流域整体预报精度,为研究流域实时调度提供了可靠依据。前人曾对双超模型参数采用传统扰动分析法仅对模型参数进行了敏感性分类,并未对模型参数敏感性系数进行定量计算,本文通过对目标函数赋权,进行模型综合敏感性系数分析,对参数的敏感性进行了客观全面的分析,成果更加完善与可靠,对深入了解双超模型产流机理、减少模型率定过程与提高模型模拟精度等具有深远的实际意义;本文所建立的BP神经网络分类模型可以较为精确的判断流域洪水量级大小,对流域洪水分类可靠。另外,洪水分类及识别结果受洪水分类特征指标选取的影响,是应进一步研究的问题。
[Abstract]:With the development of flood forecast science, the hydrological model has been widely used to solve the problems of social and human development, including hydrology, water resources, environment and ecology. The data of the rain and flood in the wet area of the south is rich, and more work can be done in the study of the hydrological model of the river basin, and the calculation method is also mature. However, for the semi-arid area, which is 52% of the territory of our country, the work of the hydrological model is less, and the use of the hydrological model in the basin is a problem both at home and abroad in the semi-arid area. The double supermodel is especially suitable for semi-humid and semi-arid areas, and should be given full attention as the first choice model of the flood forecast. At the present stage, the sensitivity and the rate of the double supermodel parameters are poor, and the accurate model information can not be provided for all levels of government and flood control departments. it is difficult to meet the demand of the flood forecast parameter rate. In this paper, a double super-production flow model is used as the research object, the sensitivity analysis of the model parameters is carried out, the important influence parameters of the output response of the model are identified, the blindness in the process of determining the model parameter rate is reduced, and the system for determining the flood classification parameter rate of the small watershed in Shanxi Province is established, and the reliability and the prediction precision of the model operation are improved. In this paper, the two-supermodel parameter sensitivity analysis of the representative field flood in each basin is selected as the object of the study on the control of the basin as the study object in Yulin, Shangjing and Lou. The sensitivity and correlation of the two supermodel parameters in different river basins, different grades of flood and multiple target functions are obtained firstly, and the comprehensive sensitivity coefficient of the model parameters is determined based on the coefficient of variation method. By using the optimized LH-OAT method, the sensitivity and the correlation of the two supermodel parameters under different river basins, different grade floods and multiple target functions are obtained, and the comprehensive sensitivity coefficient of the model parameters is determined based on the entropy value method. The results of the two methods are compared and analyzed. The results show that: (1) The comprehensive sensitivity of the model parameters is determined by the local analysis method as Srb-0Ks-C, the parameters Sr, Ks, b and {0} are sensitive parameters, and C and K are not sensitive parameters. The comprehensive sensitivity of the model parameters is determined by the global analysis method as the KsbSr-0-IGC, the parameters Sr, Ks and b are sensitive parameters, and the parameter {0} is the sensitive parameter, and the parameter C and the parameter are not sensitive parameters. The results show that the sensitivity and size of the model parameters are different from those of the different research methods. However, for the parameter sensitivity classification, only the sensitivity level of the other parameters is affected by the analysis method, and the other parameter sensitivity grades have good stability. (2) The correlation between the parameters and the objective function is analyzed by the local analysis method and the global analysis method, and the correlation between the sensitivity parameters and the target function Wi and Qmi is clear in different levels of flood and different river basins. The performance is that the parameters' 0, b 'are positively related to Wi and Qmi, and the parameters Sr, Ks are negatively correlated with Wi and Qmi. However, the parameters are not related to all target functions, and when the target function becomes IVF, RE, RSS, PE, the correlation is not clear. Therefore, there is a need to treat different target functions in the actual application, and the regulation of the parameters needs to be treated differently, and it is not all rules to follow. In this paper, the fuzzy ISOD ATA iterative model is used to cluster the historical flood. Since the flood peak flow and the flood volume of the field flood process are the main targets of the flood forecast, the flood peak flow and the total flood volume of the selected historical flood are cluster analysis. The historical flood is divided into three types of flood, medium flood and small flood according to the order of magnitude. Because the flood phenomenon is complicated and changeable, it is difficult to control, and the law of runoff generation is different in different types of flood. In order to reduce the error of forecasting the flood of the whole river basin by a group of hydrological forecasting model parameters, this paper establishes the idea of the classification rate of the parameter of the hydrological forecast model. so as to find the law of the same type of flood runoff and confluence. The classification rate of the model of the hydrological forecast model is as follows: (1) The classification model of the BP neural network is established in this paper, and the type of the basin flood can be accurately determined, and the accuracy of the model is 100% in the sample prediction. (2) The method of the watershed flood classification and forecast in this paper is to increase the qualified rate of the flood forecast from 73% to 82%, and the relative error of the flood volume from 18. 1% to 11. 3%. The qualification rate of the flood peak is also increased from 73% to 82%. The relative error of the flood peak was reduced from 16. 4% to 14. 6%. The whole forecast precision of the study basin is improved, and a reliable basis for studying the real-time dispatching of the river basin is provided. In this paper, the sensitivity classification of the model parameters is only carried out by the traditional perturbation analysis method, and the sensitivity coefficient of the model parameters is not calculated quantitatively, and the comprehensive sensitivity coefficient of the model is analyzed by the weight of the objective function. The sensitivity of the parameters is analyzed in an objective and comprehensive way, and the result is more perfect and reliable. It is of far-reaching significance to the deep understanding of the production flow mechanism of the double supermodel, the process of reducing the model rate and the improvement of the model precision, etc. The BP neural network classification model established in this paper can accurately judge the magnitude of the flood level of the river basin, and is reliable for the classification of the flood in the basin. In addition, the classification of the flood and the result of the identification are affected by the selection of the characteristics of the flood classification.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TV122

【相似文献】