Monthly Streamflow Prediction Based on Random Forest Algorithm and Phase Space Reconstruction Theory

Abstract

In order to find an effective time series forecasting method, a random forest prediction model based on phase space reconstruction theory is proposed in this paper. The Cao method and mutual information function method are used to reconstruct the phase space, and the parameters of the random forest (RF) model are discussed. The feasibility of the model is verified by analyzing the monthly runoff data of Pingshan hydrological station in Jinsha River Basin. Compared with BP neural network (BPnet) model and traditional support vector machine (SVM) model which optimizes parameters by grid algorithm, random forest model has higher prediction accuracy and less calculation in dealing with complex nonlinear hydrological time series as shown by the result.