基于SSA-VMD-CNN-LSTM的水库月径流预测方法研究

Research on Monthly Runoff Prediction Method for Reservoirs Based on SSA-VMD-CNN-LSTM

摘要: 为提高月径流预测精度，降低由径流序列非线性、非平稳及高噪声引起的预测误差，本文提出一种基于变分模态分解与深度学习、优化搜索算法相结合(SSA-VMD-CNN-LSTM)的月径流预测方法。该方法首先利用VMD将原始径流序列分解为若干本征模态函数(IMF)，并将各子模态分量重构为二维特征矩阵；随后，将其输入经麻雀搜索算法(SSA)优化的卷积神经网络与长短时记忆网络(CNN-LSTM)进行预测；最终，通过叠加各子模态的预测结果，获得完整的月径流预测序列。研究结果表明，经SSA优化的VMD-CNN-LSTM模型在月径流预测中具有更高的精度，其预测效果优于LSTM、CNN-LSTM及VMD-CNN-LSTM模型。该模型可为水库的月径流预测提供有效参考。

Abstract: In order to improve the accuracy of monthly runoff prediction and reduce the error caused by the nonlinearity, non-stationarity and high noise of series, this paper proposes a monthly runoff prediction method based on variational mode decomposition and deep learning, optimized search algorithm (SSA-VMD-CNN-LSTM). The method first decomposes the original runoff series into several intrinsic mode functions (IMF) by VMD, and each sub-modal component is reconstructed into a two-dimensional feature. Subsequently, it is input into the convolutional neural network and long-short term memory network (CNN-LSTM) optimized by sparrow search algorithm (SSA). Finally, the complete monthly runoff prediction sequence is obtained by superimposing the prediction results of each sub-modal. The results of the study show that the SSA-MD-CNN-LSTM model has higher accuracy in monthly runoff prediction, and its prediction effect is better than LSTM, CNN-LSTM and VMD-CNN-LSTM models. The model can provide effective reference for the monthly runoff prediction of reservoirs.