基于小波神经网络的东江流域径流模拟多模型比较研究

基于小波神经网络的东江流域径流模拟多模型比较研究 Multiple Models Comparative Study of the East River Basin Runoff Simulation Based on Wavelet Neural Network 张正浩 ZHANG Zhenghao 1 2 3 first-author 张强 ZHANG Qiang 1 2 3 江涛 JIANG Tao 1 2 3 陈晓宏 CHEN Xiaohong 1 2 3 中山大学水资源与环境系 中山大学水资源与环境系 Department of Water Resources and Environment,Sun Yat-sen University Department of Water Resources and Environment,Sun Yat-sen University 中山大学华南地区水循环与水安全广东省普通高校重点实验室 中山大学华南地区水循环与水安全广东省普通高校重点实验室 Guangdong University Key Laboratory of Water Cycle and Security in South China,Sun Yat-sen University Guangdong University Key Laboratory of Water Cycle and Security in South China,Sun Yat-sen University 中山大学广东省城市化与地理环境空间模拟重点实验室 中山大学广东省城市化与地理环境空间模拟重点实验室 School of Geography and Planning,Guangdong key laboratory for Urbanization and Geo-Simulation,Sun Yat-sen University School of Geography and Planning,Guangdong key laboratory for Urbanization and Geo-Simulation,Sun Yat-sen University 东江流域为香港及广州等特大城市的重要水源地,水文过程模拟及预测对流域水资源开发与管理具有重要理论与现实意义.本研究针对小波神经网络中最佳母小波和最佳等级选择问题,在与传统模型比较基础上,研究得出东江流域径流模拟的最佳小波神经网络模型,并以此进行东江流域的径流预测分析.结果表明:1)选择恰当的母小波可以有效捕捉信号统计特征,该流域蒸发量、降水量、温度和湿度的最佳分解母小波分别为Db4,Sym2,Db9及Db4小波,其小波分解最佳等级为5;2)小波神经网络作为新型混合优化模型,在母小波选择和分解等级确定后,经东江博罗站径流模拟分析,确定为东江流域最佳小波神经网络模拟模型.该研究用于东江径流的预测,效果在满意范围内. As the critical water source of large cities such as Hong Kong and Guangzhou,hydrological process simulation of East River Basin has important theoretical and practical significance on water resources development and management.Therefore,this research focuses on the optimum mother wavelet and level of a new hybrid model wavelet artificial neutral network（WANN）.The relative performance of wavelet artificial neutral network is compared with the regular traditional models for runoff simulation and prediction in East River Basin.The results indicate that:1)the mother wavelet is useful for capturing statistical characteristics of time series,and the optimal mother wavelet for evaporation,precipitation,temperature and humidity is Db4,Sym2,Db9,Db4,the best decomposed level is five;2)through selecting the best mother wavelet and level,the prediction accuracy of WANN is better than other traditional models.Using this result to predict runoff in Dongjiang Basin is within the expected range. 小波神经网络 径流模拟及预测 母小波 分解等级 模型比较 wavelet artificial neutral network runoff simulation and prediction mother wavelet decomposed level model comparation 10.14188/j.1671-8836.2015.03.011 P333 国家杰出青年科学基金项目（51425903）;香港特别行政区研究资助局项目（CUHK441313） 2015-03-01 2021-03-31 3 61