基于深度学习的河流径流值预测

Deep Learning Based River Runoff Value Prediction

在传统预测径流值方法中,由于径流序列具有非常复杂的非线性特征,所以无法充分对径流值进行信息提取并进行预测。本文提出一种基于LSTM (长短期记忆网络)和注意力机制的模型来对径流值进行预测。对于气候原因引发的自然灾害——洪水的预测来说,该模型考虑到降雨量与上游水文站径流值对目标地区的径流值的影响,并通过加入注意力机制对各个影响因素添加距离影响。以四川省乐山市五通桥区径流值数据为例,通过对比2015年径流值的预测值和真实值进行对比。验证结果表明:添加了注意力机制的LSTM模型具有误差小,准确度高的特性,可以较为显著提高径流值预测的能力。

In the traditional method of predicting runoff values, because the runoff sequence has very complex nonlinear characteristics, it is impossible to fully extract and predict the runoff values. In this paper, a model based on LSTM (Long Short-Term Memory Network) and attention mechanism is proposed to predict runoff values. The model takes into account the effects of rainfall and upstream hydrological station runoff values on the runoff values in the target area, and adds distance effects to each influencing factor by adding attention mechanisms. Taking the runoff value data of Wutongqiao District in Leshan City, Sichuan Province as an example, the predicted and true values of the runoff value in 2015 are compared. The verification results show that the LSTM model with the addition of attention mechanism has the characteristics of small error and high accuracy, which can significantly improve the ability to predict runoff values.