基于LSTM深度学习模型的华北地区参考作物蒸散量预测研究

Study on LSTM deep learning model-based prediction of reference crop evapotranspiration in North China

为有效提高华北地下水漏斗区参考作物蒸散量ET_0的预报精度,本文以华北地区7个气象代表站1958—2010年ET_(0-PM)(Penman-Monteith,P-M)的历史时间序列为训练集构建LSTM模型,以2011—2017年ET_(0-PM)的时间序列为验证集将LSTM模型与其他4种经验模型进行对比分析。结果表明:LSTM在华北地区预测的整体评价指标Gpi(Global performance indicator)排名第一,该模型可以作为华北地区逐月ET_0预测的推荐模型,为我国精准农业灌溉预报提供科学的依据。

In order to effectively improve the accuracy of prediction on the reference crop evapotranspiration(ET0) in groundwater funnel area of North China, the LSTM(Long Short-Term Memory) model is constructed through taking the historical time series of ET0-PM(Penman-Monteith, P-M) from 1958 to 2010 from seven meteorological stations in North China as the training set, and then a comparative analysis between LSTM model and the other four empirical models is made by taking the time series of ET0-PM from 2011 to 2017 as the validation set. The result shows that the global performance indicator(Gpi) of LSTM prediction in North China ranks number one, thus the model can be used as the recommended model for ET0 prediction in North China and provide a scientific basis for precise agricultural irrigation prediction in China.