Renmin Zhujiang (Jan 2021)
Research on Annual Runoff Prediction Based on EMD-LSTM-ANFIS Model
Abstract
To improve the accuracy of runoff prediction,this paper proposes a runoff prediction model based on the combination of empirical mode decomposition (EMD),long short-term memory (LSTM) neural network,and adaptive neuro-fuzzy inference system (ANFIS),decomposes the original runoff sequence into multiple regular component sequences through EMD,and reconstructs the phase space of each component sequence by the autocorrelation function method (AFM) and the false nearest neighbor method (FNN) to determine the input and output vectors,establishes the EMD-LSTM-ANFIS prediction model,and constructs the EMD-LSTM,EMD-ANFIS,LSTM,ANFIS as comparison models,as well as predicts and compares the annual runoff of the Longtan Station in Yunnan Province by the five models.The results show that the average relative error of the EMD-LSTM-ANFIS model for the annual runoff prediction is 3.18%,which is reduced by 55.0%、65.2%、68.1%、78.4% compared with the EMD-LSTM,EMD-ANFIS,LSTM,and ANFIS models respectively,with higher prediction accuracy and stronger generalization ability.Therefore,the EMD-LSTM-ANFIS model is feasible and reliable for runoff prediction.
