基于SSA-LSTM模型的黄河水位预测研究

Research on the Prediction of the Yellow River Water Level Based on SSA⁃LSTM Model

黄河流域内的水资源分配和调度对于各个地区的经济发展和人民生活至关重要。为提高黄河水位预测精度,提出一种基于麻雀搜索算法(SSA)和长短期记忆(LSTM)网络融合的黄河水资源调度预测模型,即利用SSA算法优化LSTM模型的超参数后,对黄河水位进行预测。结果表明,SSA-LSTM模型的E_(MAP)(平均绝对百分比误差)、E_(RMS)(均方根误差)、E_(MA)(平均绝对误差)和R^(2)(拟合优度)分别为0.0063、0.0304、0.0247、0.9945。相较于层感知器(MLP)、LSTM对照模型,SSA-LSTM模型的E_(MAP)、E_(RMS)、E_(MA)明显减小,R^(2)有所提升。采用SSA算法自动进行参数选优的方式,可解决LSTM模型手动选择参数的难题。这种方法不仅大幅缩短模型训练时间,还能找到最优网络参数,从而发挥模型的最佳性能。利用SSA-LSTM模型预测黄河水位具有良好的准确性和鲁棒性,可以为黄河水资源调度提供依据。

The allocation and regulation of water resources in the Yellow River Basin is crucial for the economic development and people’s lives of various regions.In order to improve the prediction accuracy of the Yellow River water level,a water resources planning and prediction model of the Yellow River was proposed,which combined the sparse search algorithm(SSA)and the long short-term memory(LSTM)network.In other words,after optimizing the hyper⁃parameters of LSTM model by SSA algorithm,the Yellow River water level was predicted.The results show that the E_(MAP)(Mean Absolute Percentage Error),E_(RMS)(Root Mean Square Error),E_(MA)(Mean Absolute Error)and R^(2) of SSA⁃LSTM model are 0.0063,0.0304,0.0247 and 0.9945 respectively.Compared to Multi⁃Layer Perception(MLP)and LSTM control models,the E_(MAP),E_(RMS) and E_(MA) of the SSA⁃LSTM model are significantly reduced,while R^(2) improved.The difficult issue of manual parameter selection in LSTM model is solved by using SSA automatic parameter selection.This method can not only greatly reduce the training time of the model,but also find the optimal network parameters,so as to exercise the best performance of the model.The SSA⁃LSTM model has good accuracy and robustness in predicting the water level of the Yellow River,which can provide a basis for the regulation of water resources in the Yellow River.