摘要
针对中长期径流时间序列具有强非线性和非平稳性特点、致使模型精准预测较为困难的问题,提出了一种结合辛几何模态分解(SGMD)、样本熵(SE)、非洲秃鹫优化算法(AVOA)和长短期记忆神经网络(LSTM)的新的SGMD-SE-AVOA-LSTM 耦合模型.首先,采用SGMD和SE对历史径流数据进行预处理;再通过AVOA优化LSTM超参数;最后,将各序列的预测结果叠加重构得到月径流预测值.选取黑河流域的莺落峡水文站、乌江流域的洪家渡水电站的月径流数据进行实例验证,并与LSTM模型、AVOA-LSTM模型、EEMD-LSTM模型、SGMD-SE-LSTM模型和EEMD-AVOA-LSTM模型进行对比,结果表明:在莺落峡水文站SGMD-SE-AVOA-LSTM模型的NSE和R分别达到0.8961和0.9498,与对比模型相比,MAE 分别减少了45.26％、18.95％、26.33％、20.09％、14.07％;RMSE 分别减少了40.06％、25.74％、31.24％、19.24％、21.65％;在洪家渡水电站 SGMD-SE-AVOA-LSTM 模型的NSE和R分别达到0.7949和0.8935,与对比模型相比,MAE分别减少了39.87％、17.86％、27.61％、20.48％、13.58％;RMSE 分别减少了39.08％、29.10％、31.86％、15.11％、22.66％.因此,本文提出的模型有效加强了LSTM模型的预测精度,为月径流预测提供了一种新的耦合模型.
Abstract
The medium-and long-term runoff time series are strongly nonlinear and non-stationary,so it is difficult to conduct model prediction accurately.Therefore,combining symplectic geometric mode decomposition(SGMD),sample entropy(SE),African vulture optimization algorithm(AVOA),and long-and short-term memory neural network(LSTM),this paper proposes a new hybrid model,SGMD-SE-AVOA-LSTM.This paper first uses SGMD and SE to preprocess data.After that,the paper makes use of AVOA to optimize the hyperparameters of LSTM.Finally,by superimposing the predicted results of each sequence,the paper obtains the predicted value of the monthly runoff.The monthly runoff data of Yingluoxia Hydrological Station in the Heihe River Basin and Hongjiadu Hydropower Station in the Wujiang River Basin are selected for example validation and compared with the LSTM model,AVOA-LSTM model,EEMD-LSTM model,SGMD-SE-LSTM model,and EEMD-AVOA-LSTM model.The results show that the NSE and R of the SGMD-SE-AVOA-LSTM model at Yingluoxia Hydrological Station are as high as 0.8961 and 0.9498,respectively.Compared with the comparison model,the MAE decreases by 45.26％,18.95％,26.33％,20.09％,and 14.07％,respectively;the RMSE decreases by 40.06％,25.74％,31.24％,19.24％,and 21.65％,respectively;and the NSE and R of the SGMD-SE-AVOA-LSTM model at Hongjiadu Hydropower Station are as high as 0.7949 and 0.8935,respectively.Compared with the comparison model,the MAE decreases by 39.87％,17.86％,27.61％,20.48％,and 13.58％,respectively;the RMSE decreases by 39.08％,29.10％,31.86,15.11％,and 22.66％,respectively.Therefore,the model proposed in this article effectively enhances the prediction accuracy of the LSTM model and provides a new hybrid model for monthly runoff prediction.
NETL
NSTL
万方数据