基于BP神经网络和SVM的江口流域洪水预报研究 毕业论文.doc

基于BP神经网络和SVM的江口流域洪水预报研究 摘要：在MATLAB7.8软件基础上，利用江口流域1981年1月1日到1985年12月31日的降雨、蒸发、径流数据，应用BP模型和SVM模型进行洪水预报模拟。结果显示，这两种模型在训练期的NASH系数都达到了0.9以上，且在预测期的NASH系数都达到了0.8以上。而且，SVM模型的训练期NASH系数达到了0.9770，BP模型达到了0.9588。SVM预测期的NASH系数达到了0.8865，BP模型为0.8197。结果分析表明SVM模型对洪峰的预报精度较BP模型好。从总体上看，这两种模型对于该流域的洪水预报效果都是比较好，并且就这两种模型相对而言SVM模型的模拟的精度较高，其预报的效果较好。但是就不同流域而言这两种模型的选择还需进行模拟对比才能确定。 关键词：洪水预报 BP神经网络 SVM Flood forecasting for the Jiangkou river basin based on BP neural network and SVM Abstract: Flood forecasting simulation for Jiangkou river basin was conducted based on the SVM and BP-ANN model on MATLAB7.8 platform,using the rainfall, evaporation, runoff data from January 1,1981 to December 31,1985. Results show that theNASH coefficients for both the two models in the training period have reached more than 0.9, and in the forecast period, they have reached more than 0.8. Moreover, SVM model’s NASH coefficient of training has reached 0.9770 and BP-ANN model is 0.9588. SVM model’s NASH coefficient of forecast period has reached 0.8865 and BP-ANN model is 0.8197. The results suggest that the peak forecast accuracy for SVM model is better than that for the BP-ANN model. In general, these two models’ effect of flood forecasting for the basin is relatively good and the SVM model’ simulation precision is higher and the forecast effect is better than BP model. But for the other different basins, it needs further simulation and comparison to make sure which model is better. Keywords: flood forecasting BP neural network SVM 1 前言 我国的防洪实践证明，工程措施与非工程措施的结合应用是建立和完善江河防洪系统的有效措施。工程措施包括：建筑堤防、防洪墙、分洪工程、河道整治工程、水库等。工程措施在我国已经经历相当长的一段时间了，也已经积累了相当丰富的经验了。非工程措施通常包括：加强防洪设施管理，保证防洪设施的防洪能力；在工程设施中充分考虑适应洪水短暂淹没的措施，以尽可能减少洪灾损失；建立健全通讯系统和预警系统；改进和发展洪水预报技术，提高防洪调度水平等。其中改进洪水预报方法、提高预报精度、实施洪水预报调度是非工程措施中最行之有效的办法[1][2]。 洪水预报是根据现时已经掌握的水文、气象资料，预报河流某一断面在未来一定时间内(称预见期)将要出现的流量、水位过程。研究水文预报方法开发水库实时洪水预报防洪调度决策支持系统，实现雨情水情资料的采集、传输、预处理、预报、调度的自动化，即联机实时洪水预报调度系统，是当前水情自动测报系统的发展方向[3][4][5]。 误差反向传播神经网络（Back Propagation ANN-Artificial Neural NetworkB