基于随机森林回归算法的山洪灾害临界雨量预估模型

Critical Rainfall Prediction Model for Mountain Torrent Disaster Based on Random Forest Regression Algorithm

针对山洪灾害临界雨量影响因素众多、计算流程复杂的问题,以山东省临朐县山丘区237个沿河村落为研究对象,选取有实测资料、计算条件较好的沿河村落,采用水文水力学法,计算山洪灾害临界雨量作为原始数据集;选取降雨特征、流域特征、沿河村落特征、河道特征等参数,基于随机森林回归算法,构建山洪灾害临界雨量预估模型,利用网格搜索法及K折交叉验证法调整预估模型参数;确定无实测资料的沿河村落山洪灾害临界雨量指标,并验证预估模型精度,分析特征参数重要度。结果表明:所构建的预估模型具有较高的准确性与泛化性,训练集与测试集的决定系数均大于0.9,预估效果较好;流域面积、流域最长汇流路径比降、降雨均值、流域平均坡度等特征参数的重要度相对较高。

In view of the problem of numerous influencing factors of critical rainfall of mountain torrent disaster and complicated calculation process,237 villages along river in the hilly area in Linqu county,Shandong province were taken as research objects.The villages along river with measured data and good calculation conditions were selected,and the critical rainfall of mountain torrent disaster was calculated by using hydrologic hydraulic method as an original data set.The parameters of rainfall characteristics,basin characteristics,characteristics of villages along river,and river channel characteristics were selected to construct a critical rainfall prediction model for mountain torrent disaster based on random forest regression algorithm.The parameters of the prediction model were adjusted by using grid search method and K-fold cross validation method.Critical rainfall indexes of mountain torrent disaster in villages along rivers without measured data were determined,and the accuracy of the estimated model and the importance of the analysis characteristic parameters were verified.The results show that the constructed prediction model has high accuracy and generalization.All of the coefficients of determination in the training set and the test set are greater than 0.9,so the prediction effect is good.The importance of characteristic parameters such as basin area,gradient of the longest confluence path of the basin,average rainfall value,and average slope of the basin is relatively high.