基于LS-SVM的河道洪水预报研究

发布时间：2018-06-07 14:32

本文选题：洪水预报 + LS-SVM　；参考：《昆明理工大学》2016年硕士论文

【摘要】：我国的洪水灾害发生频繁,严重的影响着我国国民经济的平稳健康发展。洪水预报作为一项非常重要的防洪减灾的非工程措施,在制定防洪减灾方案的工作过程中一直发挥着至关重要的作用。本文对四川省自贡市富顺县地区水文对象以及水系关系进行了分析,应用了河段洪水预报中的两种方法：水位预报和流量预报。另外,因为一些外在因素产生的影响,天然河道的水位和流量的数据在采集的过程中会产生一些误差,所以就需要剔除这些误差数据来提高河道水位和流量的预测精度。本文根据上述目标主要做了以下的研究工作：(1)分析了样本数据中误差数据产生的原因、误差的分类以及水文测验中误差的主要特点,采用了误差处理方法中的改进的拉依达准则法(3σ)和肖维勒准则法(Chauvenet)来处理样本数据中存在的误差数据。(2)比较了最小二乘法(LS)、支持向量机(SVM)、BP神经网络和最小二乘支持向量机(LS-SVM)这四种算法,得出了最小二乘支持向量机(LS-SVM)的预测能力最好的结论,所以本文中采用最小二乘支持向量机(LS-SVM)构建预测模型。(3)通过比较最小二乘支持向量机(LS-SVM)的四种核函数的预测效果,得出了RBF核函数的预测效果最好的结论,所以本文中采用的核函数为RBF核函数,并且应用网格搜索法优化模型的参数。(4)结合四川省自贡市富顺县地区的相关水位和流量数据资料构建了三种预测模型,通过对历史样本数据的学习和训练,得到所建立的三种预测模型的输出数据,根据《水文情报预报规范》(SL250-2000)为依据来评定此三种模型的预测精度。根据实验结果和《水文情报预报规范》(SL250-2000)得到：建立的三种预测模型的预测效果都比较好,并且双输入单输出(水位和流量—水位)预测模型要比单输入单输出(水位—水位)预测模型和单输入单输出(流量——流量)预测模型的预测精度高,说明：如果输入的影响因子增多,那么得到的预测模型就会越精准。该结果可以为研究河段的水位和流量的变化提供一定的参考。
[Abstract]:Flood disasters occur frequently in our country, which seriously affect the steady and healthy development of our national economy. Flood forecasting, as a very important non-engineering measure for flood control and disaster reduction, has been playing an important role in the process of making flood control and disaster reduction plan. In this paper, the hydrological object and the relationship of water system in Fushun County, Zigong City, Sichuan Province are analyzed, and two methods of flood forecasting in river reach are applied: water level forecast and discharge forecast. In addition, because of the influence of some external factors, the data of water level and discharge of natural river will produce some errors in the process of collecting, so it is necessary to eliminate these error data to improve the prediction accuracy of water level and discharge. In this paper, the following research work is done according to the above objective: 1) the causes of the error data in the sample data, the classification of the errors and the main characteristics of the errors in the hydrological test are analyzed. The improved Laida criterion (3 蟽) and the Shawville criterion (Chauvenetett) are used to deal with the error data in the sample data.) the least square method (LSN), support vector machine (SVM) BP neural network and least square branch are compared. LS-SVM) these four algorithms, It is concluded that the least square support vector machine (LS-SVM) has the best prediction ability. Therefore, the prediction model of LS-SVM is constructed by using least squares support vector machine (LS-SVM), and the prediction results of four kernel functions of LS-SVM) are compared by comparing the prediction results of LS-SVM (least squares support vector machine) and LS-SVM (least squares support vector machine). The best prediction effect of RBF kernel function is obtained, so the kernel function used in this paper is RBF kernel function. And the parameters of the model are optimized by using the grid search method. (4) combined with the data of water level and discharge in Fushun County, Zigong City, Sichuan Province, three kinds of prediction models are constructed, and through the study and training of historical sample data, three kinds of prediction models are constructed. The output data of the three prediction models are obtained and the prediction accuracy of the three models is evaluated according to the Hydrological Information Forecast Code (SL250-2000). According to the experimental results and the Hydrological Information Forecast Standard (SL250-2000), it is concluded that the prediction results of the three prediction models are good. Moreover, the prediction accuracy of double input and single output (water level and discharge water level) prediction model is higher than that of single input and single output (water level-water level) model and single input and single output (discharge-discharge) model. Explanation: if the input factors increase, the prediction model will be more accurate. The results can provide some reference for studying the variation of water level and discharge.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TV122

【相似文献】