基于随机森林和RBF神经网络的长期径流预报  

Research on the Long-Term Runoff Forecast Based on Random Forest Model and RBF Network

作　　者：赵文秀 张晓丽[2] 李国会 

机构地区：[1]翁牛特旗水利局,内蒙古赤峰024500 [2]黄河水利科学研究院,河南郑州450003 [3]中国电建集团中南勘测设计研究院有限公司,湖南长沙410014

出　　处：《人民黄河》2015年第2期10-12,共3页Yellow River

基　　金：国家自然科学基金资助项目(51409112);水利部公益性行业科研专项(201301062)

摘　　要：基于我国南方某河流1965—1999年每年7月的实测流量资料,首先采用随机森林模型筛选预报因子,之后利用筛选的预报因子作为RBF神经网络的输入层,利用RBF神经网络对2000—2008年每年7月的流量进行了"滚动式"预报,并与实测结果进行了对比。结果表明:随机森林模型能有效地筛选影响因子,利用这些因子采用RBF神经网络进行径流预报的相对误差均在10%以内,拟合效果很好;"滚动式"长期径流预报结果相对误差的绝对值均在20%以内。Based on the measured flow data of each July during the period of 1965 to 1999 of a river in southern China,the random forest model was used to filter the impact factors. Then,the RBF network input layer was trained with the selected factors and was utilized to fore-cast the flow data annually in July during 2000—2008 using rolling type pattern. The results show that the random forest model can effective-ly filter out the main factors. Based on the factors filtered by the random forest model,the relative error of RBF networks prediction results is within 10%. The fitting effect is good. Besides that,the relative error of the long-term prediction results using rolling type pattern is within 20%.

关 键 词：随机森林模型 RBF神经网络 滚动式预报 长期预报 径流预报 

分 类 号：P338[天文地球—水文科学]