深度学习的LSTM-GRU复合模型在水文模拟中的应用  

The LSTM-GRU Combinatorial Model of Deep Learning and Its Application to Hydrological Simulations

作　　者：陆冠宇 刘文强 郝慧清 王奇 郝永红[2] LU Guanyu;LIU Wenqiang;HAO Huiqing;WANG Qi;HAO Yonghong(College of Geography and Environmental Sciences,Tianjin Normal University,Tianjin 300387,China;Tianjin Key Laboratory of Water Resources and Environment,Tianjin Normal University,Tianjin 300387,China;College of Mathematical Science,Tianjin Normal University,Tianjin 300387,China)

机构地区：[1]天津师范大学地理与环境科学学院,天津300387 [2]天津师范大学天津市水资源与水环境重点实验室,天津300387 [3]天津师范大学数学科学学院,天津300387

出　　处：《水文》2023年第3期52-59,共8页Journal of China Hydrology

基　　金：国家自然科学基金资助项目(42072277,41272245,40972165)。

摘　　要：在深度学习中,长短期记忆网络(LSTM)和门控循环单元(GRU)是两种模拟时间序列、循环神经网络(RNN)的主要基础结构,各有优缺点。为弥补二者的不足,提高河流流量的预测精度,建立了LSTM-GRU复合模型,并用于海河流域大清河水系白沟河流域流量的预测。基于东茨村水文站2006—2019年的日观测数据,以8个水文气象因子(气压、水温、相对湿度、降水量、日照、地温、风速、水位)的观测数据为输入,河流流量为输出,建立LSTM-GRU水文模型。为验证该模型的优势,将LSTM-GRU的模拟结果分别与LSTM和GRU的结果进行比较。结果表明,LSTM-GRU复合模型的稳定性和精确度明显优于单一的LSTM或GRU模型,为河流流量预测提供了一个更精准的方法。Recursive neural networks(RNN)with long and short-term memory(LSTM)or gated cycle units(GRU)are deep neural networks suitable for time series,each of which has respective advantages and disadvantages.In this paper,in order to make up for their respective deficiencies and improve the prediction accuracy of river flow,the LSTM-GRU combination model is established and applied to the Baigou River basin of Daqing River system of Haihe River Basin.Based on the daily observation data of Dongci Village Hydrological Station from 2006—2019,the LSTM-GRU hydrological model was established with the observation data of 8 hydrometeorological factors(air pressure,water temperature,relative humidity,precipitation,sunshine,ground temperature,wind speed and water level)as the input and the river flow as the output.To verify the model advantages,the simulation results of LSTM-GRU were compared with those of LSTM and GRU,respectively.The results show that the stability and accuracy of the LSTM-GRU combination model are significantly better than the single LSTM with GRU model,providing a better tool for river flow prediction.

关 键 词：深度学习 长短期记忆网络(LSTM) 门控循环单元(GRU) 移动平均 流量预测 

分 类 号：TV11[水利工程—水文学及水资源] P33[天文地球—水文科学]