基于随机森林算法的吴堡站测流断面形态预测

Prediction of Cross-Section Profile of Wubu Station Based on Random Forest Algorithm

在水文站洪水期流量测验中,受风浪和漂浮物的影响以及设施设备的限制,断面测量是一直以来的难点。传统的断面邻近借用法在断面发生较大冲淤变化时会造成较大的流量计算误差。根据水深与流速之间存在的相关性,使用随机森林算法,以流速分布、水位、河宽等作为输入参数建立断面形态预测模型,对吴堡站测流断面形态进行了预测。结果表明:使用基于随机森林算法的断面形态预测模型来确定测流断面形态是对传统的断面邻近借用法的有力补充;吴堡站流量在3 000 m3/s以上测次的流量预测标准差为13%,大于规范标准,模型仍需改进。建议从两方面来提高断面形态预测的准确性:一是增加特征垂线实测水深等附加参数;二是从断面变化角度出发分析断面冲淤与水沙过程的关系,进而找到更多的影响因子加入回归模型。

During the flood season discharge test of hydrologic station,the measurement of cross section is always difficult because of the influence of wind wave and floating matter and facilities and equipment. The traditional cross section borrowing method will introduce larger flow calculation errors when the cross section changes greatly. According to the correlation between water depth and flow velocity,the random forest algorithm was used to predict the shape of the cross section of Wubu Station,with the flow velocity distribution,the water level and the river width as the input parameters. The results show that the cross section profile based on random forest algorithm is a powerful complement to the traditional cross section borrowing method; the standard deviation of the flow of Wubu Station is more than 3,000 m3/s and the standard deviation is 13%,showing the model still needs to be improved. It is suggested to improve the accuracy of the shape prediction from two aspects： the first is to increase the additional parameters such as the measured water depth of the characteristic vertical line and the second is to analyze the relationship between the erosion and siltation and the process of water and sand from the angle of cross section and then find more influence factors to put into the regression model.