基于GRU和LightGBM特征选择的水位时间序列预测模型

A WATER LEVEL TIME SERIES PREDICTION MODEL BASED ON GRU AND LIGHTGBM FEATURE SELECTION

水位时间序列受降雨量影响,在变化规律上呈现出季节性和复杂性。传统模型结构简单且很少考虑季节性因素的影响,对于汛期复杂的水位时间序列预测精度欠佳。提出一种基于GRU和LightGBM水位时间序列预测模型。利用GRU提取水位数据建立水位数据预测的基础模型,将预测结果分为非汛期与汛期两个阶段分别与LightGBM特征选择后的环境因素结合建立最终模型,解决了模型对于不同季节预测值简单叠加导致的精度丢失的情况。预测模型以射阳河流域站点为例,对水位时间序列进行预测。实验结果表明,该模型能更有效处理水文数据复杂的季节性变化,提高了预测的精确度。

The time series of water level is affected by rainfall,showing seasonality and complexity in the law of change.The traditional model is simple in structure and seldom considers the influence of seasonal factors.It is not accurate for the prediction of complex water level time series in flood season.This paper proposes a water level time series prediction model based on GRU and LightGBM.The basic model of water level data prediction was established by using GRU to extract water level data,and the prediction results were divided into two stages:non-flood period and flood season,respectively,and the final environmental model was combined with the environmental factors after LightGBM feature selection.It solved the problem of accuracy loss caused by the simple superposition of forecast values in different seasons.The prediction model took a site in sheyang river basin as an example to predict the time series of water level.Experimental results show that the proposed model can deal with the complex seasonal variation of hydrological data more effectively,and it improves the accuracy of prediction.