基于XGBoost和SHAP的海滩波浪爬高预测研究  

A study on beach wave run-up prediction based on XGBoost and SHAP

作　　者：张建 丁佩 刘楷操 路川藤[3] ZHANG Jian;DING Pei;LIU Kaicao;LU Chuanteng(Zhuhai Institute of Urban Planning&Design,Zhuhai 519000,China;Guangdong Coastal Area Disaster Prevention and Mitigation Engineering Technology Research Center,Zhuhai 519000,China;Nanjing Hydraulic Research Institute,Nanjing 210029,China)

机构地区：[1]珠海市规划设计研究院,广东珠海519000 [2]广东省滨海地区防灾减灾工程技术研究中心,广东珠海519000 [3]南京水利科学研究院,江苏南京210029

出　　处：《海洋预报》2025年第2期1-8,共8页Marine Forecasts

基　　金：水利部重大科技项目(SKS-2022087)。

摘　　要：海滩波浪爬高预测是海岸侵蚀防护和防灾减灾的关键技术支撑。针对现有经验公式在精确度、泛化性等方面的不足,将极限梯度提升模型XGBoost引入到波浪爬高预测中,利用1400多个来自实验室和现场观测的海滩波浪爬高数据,通过贝叶斯优化进行超参数调整,建立基于XGBoost的海滩波浪爬高预测模型。此外,还将可解释机器学习框架SHAP与XGBoost模型结合,以挖掘波浪爬高预测结果的关键特征。评估结果表明:XGBoost模型的决定系数为0.957,均方根误差为0.384 m,显著优于其他经验公式,整体预测可靠稳定;SHAP分析也表明XGBoost模型的预测趋势符合真实走向,且Iribarren数在海滩波浪爬高预测中起着关键作用。Beach wave run-up prediction is a key technical support for coastal erosion protection,disaster prevention and mitigation.In view of the shortcomings of the existing empirical formulas in terms of accuracy and generalization,the XGBoost model is introduced into wave run-up prediction,and more than 1400 labora tory and field observations of beach wave run-up are used as a dataset,and hyperparameter tuning is carried out by using Bayesian optimization,which in turn establishes an XGBoost-based wave run-up prediction model.The XGBoost model is used to predict beach wave height,and SHAP,an interpretable machine learning framework,is combined with the XGBoost model to explore the key features of the wave height prediction results.The evaluation results show that the R-squared of the XGBoost model is 0.957,and the root-mean-square error is 0.384 m,which is significantly better than other empirical formulas,and the overall prediction is reliable and stable,meanwhile SHAP shows that the XGBoost model predicted trend is in line with the true value direction and Iribarren number plays a key role in beach wave run-up prediction.

关 键 词：机器学习 波浪爬高 极限梯度提升模型 贝叶斯优化 可解释机器学习框架 

分 类 号：P731.33[天文地球—海洋科学]