改进型LSTM模型在西北干旱区径流模拟中的应用

——以祖厉河为例

打开文本图片集

关键词：GWO-LSTM-Attention模型;XGBoost模型；径流预测中图分类号：TV121 文献标识码：A 文章编号：1001-9235（2025）08-0050-08

Application of Improved LSTMModel in Runoff Simulation in AridRegion ofNorthwest China： A Case Study of the Zuli River

SONGHaiping，TANGYiran，DANGWentao，WANGYibo

1.GansuProvincialDepartmentofWaterResources WaterConservancyProjectConstructionCostandFeeManagementCenter， Lanzhou73o，ina;2.CollgeofarthndEnviromentaliences，LnouUnversityanzouOa）

Abstract：InviewofthescarcityofmeasureddataintheZuliRiverBasin，theupperreachesoftheYelowRiver，aswellasthe diffculthdrologicalsimulationproblemscausedbytheintenseinterferenceofhumanactivities，thisstudyfocusedontheZuliRiver Basinin theupperreachesoftheYellowRiver.XGBoost，LSTM-Attention，andtheimprovedLSTMmodel（GWO-LST-Aention model）wereemploydtosimulate monthlyunof.Usingobservedrunof，precipitation，andonthlymeantemperaturedatafrom1980 to2020，theresearchincorporatedfeatureengineering，combinedwithextreme-valuepost-procesingandmixedlossfunction optimization.Onthisbasis，thegreywolfoptimization（GWO）algorithmwasusedtooptimizetheparametersoftheLSTM-Aention model，andteGO-T-etioodelssucted，aingteodelsabilitytoptureeio'spef mechanisms. The results indicate that： ① The improved LSTM model demonstrates superior performance in simulating the Zuli River， validating the adaptability of the improved strategies for complex hydrological processes. ② The GWO algorithm significantly enhancesthemodel'spredictionacuracyinthebasin，confirmingtheeffectivenessofoptimizationalgorithmsinparameter calibration. ③ The GWO-LSTM-Atention model requires further validationand optimization for future daily-scale simulations and monthly runoff reconstruction.

Keywords：GWO-LSTM-Attentionmodel;XGBoostmodel; runoff prediction

全球气候变暖背景下，西北干旱区水资源短缺问题日益严峻，径流作为水资源的重要指标，其精准模拟对于水资源优化配置和管理至关重要。（剩余8994字）