Application of support vector machine based on phase-space reconstruction to medium-term and long-term runoff forecast

Effective medium-term and long-term runoff forecast is of great significance to the successful water resources planning and management, reservoirs and hydropower stations operation. The combined method, called support vector machine (SVM) based on the phase-space reconstruction, is developed according to the limitation of the past chaotic forecasting methods. Firstly, the existence of chaos in runoff time series is determined. Secondly, phase-space reconstruction of the runoff series is conducted and SVM, based on a principle that aims at minimizing the structural risk, is employed for forecasting, and the radial basis kernel is used as kernel function. Mutative scale chaos optimization algorithm is employed to search optimal parameters of the SVM. Comparison of the proposed method and artificial neural network indicates that the former is superior to the latter both in the forecasting accuracy and generalization ability.