基于随机森林的复坡堤越浪量预测研究

Overtopping prediction for composite slope breakwater based on random forest method

针对复坡堤越浪量的计算问题,提出了采用随机森林算法预测越浪量的方法。首先,通过对欧洲CLASH数据集进行筛选,挑选出符合复坡堤越浪量预测的数据;其次,对数据做无量纲化处理,建立以随机森林为基础的复坡堤越浪量预测模型,并通过网格搜索(GridSearchCV)方法对模型进行调参以改善模型的性能;最后,利用决定系数R^(2)来评估模型的精度,并将随机森林模型与集成神经网络模型做了预测能力的对比,同时还给出了随机森林模型各个特征参数对预测精度的重要性。结果显示,随机森林模型的决定系数为92.7%,集成神经网络模型的决定系数为87.7%,表明随机森林模型对越浪量具有更强的学习和预测能力。通过对特征重要性的分析,墙顶高程对模型预测精度的影响最大,堤顶高程次之,堤脚宽度影响最小。

Aiming at the problem of calculating overtopping of the composite slope breakwater,a prediction model of the overtopping for the composite slope based on the random forest method is proposed.Firstly,by filtering the European CLASH data set,the data consistent with the prediction of overtopping of the composite slope breakwater are selected.Secondly,after dimensionless processing of the data,overtopping prediction model is established based on random forest method,and improved by adjusting the model parameters according to GridSearchCV.Finally,the coefficient of determination R^(2) is used to evaluate the accuracy of the model,and the prediction ability of the model is compared with the ensemble neural network model.The effect of each feature parameter of the ran-dom forest model on the prediction accuracy is assessed.The results show that the coefficient of determination of the random forest model is 92.7%,and the coefficient of determination of the ensemble neural network model is 87.7%,indicating the random forest model has a stronger prediction ability for predicting overtopping.Wall height with respect to static water level has the greatest influence on the prediction accuracy of the model,the height of the top of the embankment is the second,and the width of the foot of the embankment least.