River Water Turbidity Forecasting Based on Phase Space Reconstruction and Support Vector Regression

Due to the nonlinear and nonstationary of river water turbidity, a novel hybrid forecasting model based on phase space reconstruction and support vector regression (PSR-SVR) is proposed. Firstly, the embedding dimension is chosen by using the false nearest neighbor method, and the time delay is obtained by the average mutual information. The phase space is reconstructed from the time series with the embedding dimension and the time delay got. The reconstructed time array is used as the input signal of support vector regression network. Then the forecasting model is established. Utilizing the model to forecast the river water turbidity, and it shows the accuracy of this new forecasting model is superior to RBF and BP forecasting methods.