基于LSTM-SVM模型的河流流量预测

Forecasting of river flow based on LSTM-SVM model

基于长短时记忆神经网络(LSTM)处理中长期时间序列方面的优势和支持向量机(SVM)泛化能力强的特点,提出了一种LSTM-SVM混合模型,并应用于海河流域拒马河逐日流量模拟和预测中,探究该模型的适用性.基于流域内逐日气象数据及水文数据资料,分别利用LSTM和LSTM-SVM构建水文模型,并通过性能评估指标对其进行分析.结果表明:①2种模型在预测河流流量时表现良好,具有可靠性;②时间步长为4 d时,模型预测结果的精度最高,混合模型效果优于LSTM模型,具体表现在LSTM-SVM模型模拟结果的离散度比LSTM模型小,更具稳定性.因此,LSTM-SVM混合模型在中长期河流流量预测中具有应用潜力.

Based on the advantages of long-short-term memory neural network(LSTM)in processing medium and long term-time series and the strong generalization ability of support vector machine(SVM),a hybrid LSTM-SVM model was proposed and applied to the daily river flow simulation and forecasting of the Joma River in Haihe River Basin to explore its applicability.LSTM and LSTM-SVM were used to construct hydrological models based on daily meteorological data and hydrological data in the basin,respectively,and the obtained results were analyzed through performance evaluation indicators.The results showed that:①The two models performed well in river flow forecasting and had good reliability;②The accuracy of prediction results was the highest when the time step was 4 days,and the hybrid model was superior to the LSTM model.Specifically,the sim-ulation results of LSTM-SVM model were less discrete and more stable than those of LSTM model.Therefore,the LSTM-SVM hybrid model had application potential in medium and long-term river flow forecasting.