基于VMD-NGO-LSTM的融雪洪水汛期非平稳性极值径流预测模型及应用  

Non-Stationary Extreme Runoff Prediction Model of Snowmelt Flood in Flood Season Based on VMD-NGO-LSTM and Its Application

作　　者：周霞 周峰[1,2] ZHOU Xia;ZHOU Feng(College of Hydraulic and Civil Engineering,Xinjiang Agricultural University,Urumqi 830052,China;Xinjiang Key Laboratory of Water Conservancy Project Safety and Water Disaster Prevention,Urumqi 830052,China)

机构地区：[1]新疆农业大学水利与土木工程学院,新疆乌鲁木齐830052 [2]新疆水利工程安全与水灾害防治重点实验室,新疆乌鲁木齐830052

出　　处：《人民珠江》2024年第6期127-137,共11页Pearl River

摘　　要：金沟河属于典型的融雪补给流域,受自然环境、气候变化和人类活动等因素的影响,汛期极值径流序列表现出非平稳性及复杂性特征,给流域内汛期极值径流精准预测带来新的挑战。为解决该地区汛期极值径流的非平稳性对于预测结果的影响,引入变分模态分解算法(Variational Mode Decomposition,VMD),提出一种基于北方苍鹰优化算法(Northern Goshawk Optimization,NGO)与长短期记忆神经网络(Long Short-Term Memory,LSTM)的组合预测模型(VMD-NGO-LSTM),应用于金沟河流域八家户水文站1964—2016年的汛期极值径流预测,采用均方根误差(RMSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE)、Nash系数(NSE)评价模型的预测能力。结果表明:(1)根据金沟河流域融雪洪水汛期径流极值序列的周期变化和趋势变化的水文特性变化结果表明径流极大值序列和径流极小值序列均具有非平稳性;(2)VMD-NGO-LSTM预测模型的NSE均大于0.97,且RMSE、MAPE、MAE值均处于偏小状态,与VMD-LSTM模型和VMD-NGO-BP模型相比,VMD-NGO-LSTM模型能够很好地预测八家户汛期极值径流的变化过程。该研究为汛期极值径流预测工作提供了新的思路,对新疆地区防洪减灾具有一定参考价值。Jingou River is a typical snowmelt recharge basin.Due to the influence of natural environments,climate changes,and human activities,the extreme runoff sequence in flood season shows non-stationary and complex characteristics,which brings new challenges to the accurate prediction of extreme runoff of the basin in flood season.In order to eliminate the influence of the nonstationarity of extreme runoff in the flood season on the prediction results in the basin,the variational mode decomposition(VMD)algorithm was introduced,and a combined prediction model(VMD-NGO-LSTM)based on northern goshawk optimization(NGO)and long short-term memory neural network(LSTM)was proposed.It was applied to the extreme runoff prediction of the Bajiahu hydrological station in the Jingou River Basin from 1964 to 2016.The root mean square error(RMSE),mean absolute error(MAE),mean absolute percentage error(MAPE),and Nash coefficient(NSE)were used to evaluate the prediction ability of the model.The results show that:①According to the change in hydrological characteristics including period and trend of the extreme runoff sequence of the snowmelt flood in the Jingou River Basin in the flood season,the maximum runoff sequence and minimum runoff sequence are non-stationary.②The NSE values of the VMD-NGO-LSTM prediction models are all greater than 0.97,and the RMSE,MAPE,and MAE values are all small.Compared with the VMD-LSTM model and VMD-NGO-BP model,the VMD-NGO-LSTM model can well predict the change process of extreme runoff of Bajiahu hydrological station in flood season.This study provides a new idea for predicting extreme runoff in flood season and has a certain reference value for flood control and disaster reduction in Xinjiang.

关 键 词：融雪洪水 极值径流预测 变分模态分解 北方苍鹰优化算法 长短期记忆神经网络 非平稳性 

分 类 号：TV121[水利工程—水文学及水资源] P338.4[天文地球—水文科学]