基于差分序列多尺度深度学习的径流量预测方法OACSTPCD

Accurate forecasting of runoff is the main basis for preventing flooding accidents.Due to the strong non-stationarity of runoff data,runoff information is difficult to be fully mined by a single method,which easily leads to low prediction accuracy.In this paper,a runoff prediction method based on multi-scale deep learning of differential sequence is proposed.Firstly,the first-order differential series of runoff is calculated to provide modeling samples for model building.Secondly,aiming at the volatility of differential series,the variational mode decomposition(VMD)method is used to transform the difference series,and the multi-scale LSTM method is used to estimate the components of the transformed difference series.Finally,the runoff prediction is obtained by combining the original runoff tail data and the differential series prediction results.The results show that the proposed modeling method based on differential series can realize adaptive error correction,while the multi-scale deep learning method solves the data fluctuation characteristics,and the overall prediction performance is superior.