Abstract:

The theory of phase space reconstruction is introduced into the monthly runoff simulation. The C-C algorithm was used for phase space reconstruction, and the one-dimensional runoff time series were expanded to multi-dimensional. Furthermore, with the association of CV-SVM (cross validation-support vector machine) principle and methods, a runoff time series model was established. Meanwhile, traditional BP, double hidden layer BP and GA-BP runoff time series simulation model were constructed for comparison. The monthly runoff time series at Longtanzhai of Panlong River was taken as analysis example. The results showed that the model based on phase space reconstruction and CV-SVM can better handle the complex runoff series. During the simulation of test sample of 200 months, the average relative error e_MRE, the maximum relative error e_MaxRE, the determination coefficient DC, and the qualified rate QR was respectively 0.5717%, 5.5267%, 0.9999 and 100%, which demonstrated that the model is of high generalization ability and simulation precision. The simulation result is obviously superior to those of traditional BP, double hidden layer BP model, and is even better than GA-BP model. The results indicate that the model based on phase space reconstruction and CV-SVM model for runoff simulation is feasible.

Key words: phase space reconstruction, support vector machine, cross validation, chaos, runoff simulation