| 摘要: |
| 本文利用2015、2016年5-10月赤水河沿岸的茅台站、二郎站、赤水站三个水文监测站以及赤水站、习水站两个气象站逐小时采集存储的水文气象数据，基于多元线性回归(Multiple linear regression, MLR)、岭回归(Ridge regression)和套索回归（Least absolute shrinkage and selection operator，LASSO）三种机器学习方法，构建预测赤水河中下游未来六小时水位趋势的模型。结果表明，基于机器学习的方法可以较好的预测赤水河中下游未来6小时的水位情况，而利用72小时滞后量作为输入集的LASSO回归模型能取得RMSE为0.192m的预测效果。 |
| 关键词: 水位预测；机器学习；LASSO回归 |
| DOI： |
| 投稿时间：2018-02-28修订日期：2019-04-10 |
| 基金项目: |
|
| Research on Early Warning of Streamflow in the Middle and Lower Reaches of Chishui River Based on Machine Learning |
| wengling |
| (GuiYang Meteorological Bureau) |
| Abstract: |
| In this paper, we investigate the application to forecasting of 6 hours
waterlevel at Chishui station on Chishui river using machine learning models such as MLR, ridge regression and LASSO. The data used in the study were water level time series data measured at the Chishui station, Maotai station and Erlang station and other three rainfall time series collected by the automatic monitoring station in the middle and lower reaches of Chishui river at 2015-2016 May to October. The results show that machine learning based method can well predict the water level of the middle and lower reaches of Chishui River in the next 6 hours, and the optimal prediction effect can be obtained by using the LASSO regression model which constructs the 72 hour lag as the input set, and the RMSE is the predictive effect of 0.192m. |
| Key words: waterlevel forecast; machine learning; LASSO |
