| 引用本文: | 江赜伟,杨士红,柳真杨,等.基于机器学习的排涝闸站雨后水位预测[J].灌溉排水学报,2022,():-. |
| JIANG Zewei,YANG Shihong,LIU Zhenyang,et al.基于机器学习的排涝闸站雨后水位预测[J].灌溉排水学报,2022,():-. |
|
| 摘要: |
| 【目的】排涝闸站雨后水位预测对于调节闸站不利工况和农田涝灾预警具有一定的参考意义，但目前尚未有简便可行的方法。【方法】在一年的田间实测水位数据分析的基础上，本研究收集了四湖流域两个典型闸站（习家口站、田关站）过去十年（2010-2020）的水情数据，使用了两种机器学习方法（支持向量机回归SVR、回归树）对雨后闸站最不利工况进行了预测。【结果】结果表明， SVR和回归树模型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；两种机器学习模型在习家口站的预测表现均要好于田关站，核函数的选取对SVR模型的预测结果有一定的影响，线性核函数表现较为稳定。回归树模型的效果略好于SVR模型。【结论】使用闸上水位、降雨量、降雨时间、泵站排水流量为特征变量预报雨后最高闸上水位是可行的。不同闸站有必要分开进行训练，并寻找最优的机器学习算法，未来有必要结合降雨预报数据实现农田涝灾的实时预报。本研究对易涝地区的水利工程设施调控和农田防涝减灾管理等具有一定的指导和借鉴意义。 |
| 关键词: 农田;机器学习;闸上水位;涝灾;预测 |
| DOI：10.13522/j.cnki.ggps.2021600 |
| 分类号:S276 |
| 基金项目:国家重点研发计划项目（2018YFC1508303）,国家自然科学基金项目（面上项目，重点项目，重大项目）;江苏省水利科技项目（2018065）;江西省水利科技项目（201921ZDKT06） |
|
| Prediction of water level of drainage sluice station after rain based on machine learning |
|
JIANG Zewei1, YANG Shihong1, LIU Zhenyang1, XU Junzeng1, Pang Qingqing2
|
|
1.College of Agricultural Science and Engineering,Hohai University;2.Nanjing Institute of Environmental Sciences,Ministry of Ecology and Environment
|
| Abstract: |
| 【Objective】 The prediction of water level after rain at the drainage sluice station has a certain reference significance for regulating the adverse working conditions of the sluice station and early warning of farmland waterlogging. However, there is no simple and feasible method at present. 【Method】 Based on the analysis of one year"s field measured water level data, this study collected the hydrological data of two typical sluice stations (Xijiakou station and Tianguan station) in the Sihu basin in the past ten years (2010-2020) and used two machine learning methods (support vector machine regression SVR and regression tree) to predict the most unfavorable conditions of sluice station after rain. 【Result】 The results showed that SVR and regression tree model could well simulate the maximum sluice water level after the rain of Xijiakou station and Tianguan station, R2 was basically greater than 0.80; the prediction performance of the two machine learning models in Xijiakou station was better than that in Tianguan station. The selection of kernel function has a certain influence on the prediction results of the SVR model, and the linear kernel function is more stable. The effect of the regression tree model was slightly better than the SVR model. 【Conclusion】 It is feasible to predict the maximum sluice water level after the rain by using the characteristic variables of sluice water level, rainfall, rainfall time, and drainage flow of pumping station. Moreover, it is necessary to train different sluice stations separately and find the optimal machine learning algorithm. In the future, it is necessary to realize the real-time prediction of farmland waterlogging by combining it with rainfall forecast data. This study has certain guidance and reference significance in the regulation of water conservancy facilities and the management of farmland waterlogging prevention and mitigation in waterlogging prone areas. |
| Key words: Farmland, machine learning, water level before the gate, waterlogging disaster; prediction |
