首页|期刊导航|江西水利电力大学学报|基于 Hyperband-PatchTST模型的河流流量及水位预测

基于 Hyperband-PatchTST模型的河流流量及水位预测OA

Prediction of streamflow and water level based on Hyperband-PatchTST model

中文摘要英文摘要

河流流量与水位的精准预测是水资源管理、防洪减灾与流域生态调控的关键.本文以圣胡安河和科罗拉多河监测站实测数据为研究对象,构建了融合 Hyperband 和 PatchTST 的深度学习模型,并与 LSTM、GRU、ANN 等模型进行对比分析.通过 Hyperband 算法优化超参数,可以降低模型训练成本和计算资源消耗;借助 PatchTST 将水文数据划分为局部片段(Patch),并利用自注意力机制(Self-Attention)捕捉数据的时序局部特征与非线性关系,可提高模型计算效率、精度和稳定性,实现对流量及水位的精准预测.结果表明:Hyperband-PatchTST 在各评价指标上均取得最优表现,相较于次优模型,其 MAE 和 RMSE 降低幅度分别为6%~12%和2%~8%,R2 衰减速度更缓,预测稳定性与泛化能力显著提升,可为河流水文预测提供技术支撑.

The prediction of streamflow and water level is crucial for water resource management,flood control and disaster mitigation,and watershed ecological regulation.To improve the accuracy and efficiency of such predictions,this study uses measured data from monitoring stations on the San Juan River and the Colorado River as the research object.A deep learn-ing model integrating Hyperband and PatchTST is constructed and compared with models such as LSTM,GRU,and ANN.The Hyperband optimization algorithm tunes the hyperparameters of the deep learning model,reducing model training costs and computational resource consumption.PatchTST divides the hydrological data input into local segments(Patches)and u-ses the Self-Attention mechanism to capture local temporal features and nonlinear relationships in the data,thereby impro-ving the model's computational efficiency,accuracy,and stability,and enabling precise prediction of stream flow and water level.The results show that Hyperband-PatchTST model achieves optimal performance across all evaluation metrics.Com-pared with the suboptimal model,its MAE and RMSE are reduced by approximately 6%~12% and 2%~8%,respective-ly,and its R2 decays more slowly over time.It indicates that the model offers higher prediction accuracy,stronger stability,and better generalization ability,providing an efficient and reliable new method for prediction of stream flow and water level.

陈舒泰;彭友文

江西水利电力大学 水利工程学院,江西 南昌 330099江西水利电力大学 水利工程学院,江西 南昌 330099

建筑与水利

河流流量水位预测Hyperband-PatchTST模型超参数优化深度学习

streamflow predictionwater level predictionHyperband-PatchTST modelhyperparameter optimizationdeep learning

《江西水利电力大学学报》 2026 (4)

10-20,11

国家自然科学基金项目(52579124)

评论