水闸扬压力混合预测模型构建与解释  

Construction and Interpretation of a Hybrid Prediction Model for Sluice Lift Pressure

作　　者：胡璟[1] 王豹[1] 王璐 龙俊[1] 贝欣 孙远 曹文翰 HU Jing;WANG Bao;WANG Lu;LONG Jun;BEI Xin;SUN Yuan;CAO Wenhan(Hongze Lake Water Conservancy Project Management Office,Huaian 223100,Jiangsu,China;Nanjing Sanchahe River Estuary Sluice Management Office,Nanjing 210036,Jiangsu,China;College of Hydrology and Water Resources,Hohai University,Nanjing 210024,Jiangsu,China)

机构地区：[1]江苏省洪泽湖水利工程管理处,江苏淮安223100 [2]南京市三汊河河口闸管理处,江苏南京210036 [3]河海大学水利水电学院,江苏南京210024

出　　处：《水力发电》2025年第3期50-56,共7页Water Power

基　　金：国家自然科学基金资助项目(52379122);中央高校基础业务费(2019B69814)。

摘　　要：针对现有“数据驱动”模型对水闸扬压力预测结果的可解释性低且未解译影响因素对扬压力不同成分的贡献程度,提出了一种基于逐次变分模态分解(SVMD)和集成学习算法的扬压力可解释混合预测模型。该模型采用SVMD将扬压力分解重构为趋势项、周期项和波动项;然后,基于轻量级梯度提升机(LGBM)对不同分量逐一建立预测模型并汇总结果。此外,采用SHAP方法分析渗流影响因素对水闸扬压力不同分量预测结果的影响程度及相关关系。工程实例表明,所提模型与单一算法模型相比,性能平均提升了87.1%,与混合模型相比,精度平均提升了84.6%,验证了模型有效性并提高了模型的可解释性。Aiming at the low interpretability of the existing“data-driven”models for sluice uplift pressure prediction results and the undeciphered contribution of factors to different components of the uplift pressure,an interpretable hybrid prediction model for sluice lift pressure based on the successive variational modal decomposition(SVMD)and ensemble learning algorithm is proposed.The model firstly adopts SVMD to decompose the uplift pressure into trend,seasonal and fluctuation terms,and then establishes corresponding models for different components based on the lightweight gradient boosting machine(LGBM)and summarizes the prediction results.In addition,the SHAP is used to analyze the influence and relationship of seepage influencing factors on the different components prediction results of the sluice uplift pressure.The engineering examples show that the proposed model improves the prediction performance by 87.1%on average compared with the single-algorithm model,and by 84.6%on average compared with the hybrid prediction model,which verifies the effectiveness of the model.At the same time,the interpretability of the model is improved.

关 键 词：水闸扬压力预测 逐次变分模态分解 集成学习 SHAP 模型解释 

分 类 号：TV698.1[水利工程—水利水电工程]