基于物联网和GCNN-LSTM的河流水文预测方法

River Hydrological Prediction Method Based on GCNN- LSTM

摘要:

针对河流水文存在预测精度不高的问题，利用物联网技术设计了分布式的降雨和水文信息自动采集系统，并提出了一种基于图卷积神经网络和长短期记忆网络模型对河流水位和径流量进行预测的方法。首先通过分析确定了影响河流水文的主要因素，将流域范围内的降雨量信息组成网格化的二维图形矩阵。然后提出了GCNN-LSTM预测模型，将含有降雨信息的二维图形矩阵作为网络模型的输入，获取该流域内降雨与水文变化的时空分布特征。最后采用所提出的GCNN-LSTM预测模型对河南省周口市段颍河的历史水文数据进行训练，再利用训练后的网络对测试集数据进行预测，得到了较高精度的径流量和水位结果，径流量预测结果的RMSE、MAPE和MAE分别仅为17.09m3/s、1.68%和8.57m3/s，水位预测结果的RMSE、MAPE和MAE分别仅为0.32m、0.65%和0.29m，与其他几种预测方法相比表现出了优越性，对科学合理利用水资源和防洪减灾具有重要意义。

Abstract:

Aiming at the problem of low prediction accuracy in river hydrology, a distributed automatic collection system for rainfall and hydrological information was designed using Internet of Things technology. A method for predicting river water level and runoff based on graph convolutional neural networks and Long Short-Term Memory (GCNN-LSTM) network models was proposed. Firstly, the main factors affecting river hydrology were identified through analysis, and the rainfall information within the watershed was composed into a grid based two-dimensional graphical matrix. Then, a GCNN-LSTM prediction model was proposed, using a two-dimensional graphical matrix containing rainfall information as input to the network model to obtain the spatiotemporal distribution characteristics of rainfall and hydrological changes in the watershed. Finally, the proposed GCNN-LSTM prediction model is used to train the historical hydrological data of the Yinghe River in Zhoukou City, Henan Province, then the trained network is utilized to predict the test set dataand and get a high-precision results of runoff and water level, and the RMSE, MAPE, and MAE of the runoff prediction results are only 17.09m3/s, 1.68%, and 8.57m3/s, respectively, while the RMSE, MAPE, and MAE of the water level prediction results were only 0.32m, 0.65%, and 0.29m, respectively, compared with other prediction methods, which demonstrates superiority and has a great significance for the scientific and rational utilization of water resources and flood control and disaster reduction.