Machine learning water level prediction feature construction method based on rainfall hysteresis effect

The invention provides a machine learning water level prediction feature construction method based on a rainfall lag effect, which determines the time length of backward movement of rainfall data through a method of moving the rainfall data forward and calculation of correlation coefficients in the time dimension. Machine learning water level prediction features of the rainfall hysteresis effect are constructed; the method specifically comprises the steps of rainfall lag feature construction, rainfall lag feature data processing, and correlation coefficient and feature selection of rainfall sequence data and upstream and downstream water level data. By applying the technical scheme, the precision of the machine learning water level prediction model can be improved. 本发明提供了一种基于降雨滞后效应的机器学习水位预测特征构造方法，在时间维度上，通过将降雨量数据往前移动的方法，并通过相关系数的计算，来决定降雨量数据往后移动的时间长度，从而构造降雨量滞后效应的机器学习水位预测特征；具体包括降雨量滞后特征构造、降雨量滞后特征数据处理、雨量序列数据与上下游水位数据的相关系数及特征选择。应用本技术方案可实现提升机器学习水位预测模型的精度。