沙漠与绿洲气象2025，Vol.19Issue(3)：157-164,8.DOI:10.12057/j.issn.1002-0799.2310.13003
基于LSTM和KNN组合模型的黄河源区日径流量模拟研究
Simulation of Daily Runoff at the Yellow River Source Region Using a Combined LSTM-KNN Model
摘要
Abstract
This study develops a hybrid model integrating Long Short-Term Memory(LSTM)and K-Nearest Neighbors(KNN)algorithms to predict daily runoff at the Jimai Station,Jungong Station,and Tangnaihai Station in the Yellow River source region.Dynamic attributes of the watershed were constructed using meteorological variables(e.g.,temperature and precipitation),while static attributes were derived from historical hydro-meteorological and geographic data.Feature selection was performed using the LSTM model,and the optimized TOPO_CLIM_SOIL_LSTM model was applied for daily runoff prediction,followed by real-time correction via the KNN algorithm.Results indicate that the TOPO_CLIM_SOIL_LSTM model effectively captures rainfall-runoff relationships and stabilizes low-flow predictions.After KNN correction,the accuracy of next-day runoff forecasts exceeds 93%at all stations,with the Nash-Sutcliffe Efficiency(NSE)increasing by 18.07%,6.45%,and 12.5%for Jimai Station,Jungong Station,and Tangnaihai Station,respectively,demonstrating significant improvement in prediction precision.关键词
黄河上游/日径流量预测/LSTM模型/特征量/KNN模型Key words
Yellow River source region/daily runoff prediction/LSTM model/feature/KNN model分类
天文与地球科学引用本文复制引用
李永花,代青措,马玉芳,刘玮,张静..基于LSTM和KNN组合模型的黄河源区日径流量模拟研究[J].沙漠与绿洲气象,2025,19(3):157-164,8.基金项目
国家自然科学基金区域发展联合基金(U22A20556) （U22A20556）
第二次青藏高原科考项目(2019QZKK0105) （2019QZKK0105）
青海省温室气体及碳中和重点实验室开放基金项目(MSXM-2023-02) （MSXM-2023-02）
