%0 Journal Article
%T 基于机器集成学习的中长期径流预报研究
Research on Medium and Long-Term Runoff Forecasting Based on Machine Integrated Learning
%A 吕盼成
%A 王丽萍
%A 刘源
%J Journal of Water Resources Research
%P 44-52
%@ 2166-5982
%D 2021
%I Hans Publishing
%R 10.12677/JWRR.2021.101005
%X
中长期径流预报对水库优化调度及水资源优化开发利用都有着重要的意义。首先采用基于Boosting算法的梯度提升回归树(Gradient Boosting Decision Tree, GBRT)和极端梯度提升树(Extreme Gradient Boosting, XGBoost)、基于Bagging算法的随机森林(Random Forest, RF)和极端随机树(Extreme Random Tree, ET)四种算法作为预报模型对锦屏一级水库月平均入库流量序列进行预报，并对预测结果进行对比分析。结果显示，RF预测效果最差，XGBoost预测效果最好。进一步选用其中预测效果较好的三个方法ET、XGBoost、GBRT作为初级学习器，以Logistic回归作为次学习器，进行Stacking集成学习预测。结果表明，Stacking集成学习的预测效果要优于单一模型中预测效果最好的XGBoost，其预测值的结果和实测值更为接近，为中长期径流预报提供了新思路。
Medium and long-term runoff forecast is of great significance to the optimal operation of reservoirs, development and utilization of water resources. Firstly, the gradient boosting decision tree (GBRT) and extreme gradient boosting (XGBoost) based on boosting algorithm are selected. There is also random forest (RF) and extreme random tree (ET) based on bagging algorithm. These four algorithms are used as forecasting models to forecast the average monthly inflow of the Jinping-I Reservoir, and then the prediction results are analyzed and compared. The results showed that the RF prediction was the worst, and XGBoost was the best. Then, the three methods with better prediction effect are ET, XGBoost and GBRT as primary learners, logistic regression as secondary learners, and stacking ensemble learning to predict. The first mock exam results show that the prediction result of Stacking ensemble learning is better than that of XGBoost with the best prediction result in a single model. The predicted value is closer to the measured value, which provides a new idea for medium and long-term runoff forecast.
%K 径流预报，集成学习，机器学习，锦屏一级水库
Runoff Forecast
%K Ensemble Learning
%K Machine Learning
%K The Jinping-1 Reservoir
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=40392