基于机器学习的排涝闸站雨后水位预测

Abstract

【目的】精准预测排涝闸站雨后水位。【方法】在分析为期1 a的田间实测水位数据的基础上，收集了四湖流域2个典型闸站（习家口站、田关站）为期10 a（2010—2020年）的历史水情资料，利用2种机器学习算法（支持向量机回归算法、回归树算法）对排涝闸站的雨后水位进行预测分析。【结果】支持向量机回归算法和回归树算法均较好地预测了习家口站和田关站的雨后最高闸上水位，R2基本大于0.80；2种机器学习算法在习家口站的表现均优于田关站，核函数的选取对支持向量机回归算法的预测结果有一定影响，线性核函数表现较为稳定。回归树算法的效果略优于支持向量机回归算法。【结论】基于闸上水位、降水量、降水时间、泵站排水流量预测雨后最高闸上水位是可行的。不同闸站应分开进行训练，并寻找最优的机器学习算法，未来有必要结合降水预报数据实现农田涝灾情况的实时预报。

Alternate abstract:

【Objective】The aim of this paper is to propose a new method to improve the accuracy of the prediction of water level at drainage sluice stations after rainfalls.【Method】The analysis was based on field-measured water level data spanning over one year. We collected hydrological data measured from 2010—2020 from two sluice stations at Xijiakou andTianguan, respectively, in the Sihu Basin, and used two machine learning methods -support vector machine regression (SVR) and regression tree - to predict the most unfavorable conditions of the sluice stations after rainfalls. 【Result】Both SVR and the regression tree model are able to predict the maximum sluice water level after rainfalls in the two stations with R2>0.80. On average, the two models worked better for the Xijiakou station than for the Tianguan station. The selection of the kernel function has a consequence for the SVR model, with the linear kernel function working better. The regression tree model was slightly better than the SVR model.【Conclusion】The maximum sluice water level following rainfalls can be predicted reasonably well using the characteristic variables of the sluice water level, rainfall intensity, rainfall duration, and drainage flow of the pumping station. It is necessary to train the machine learning methods for different sluice stations to find the most accurate one.