Research on the method of nonlinear combining forecasts based on fuzzy-neural systems

In this paper, a new nonlinear combination forecasting method based on a fuzzy neural network is presented to overcome some limitation that linear combination forecasting may meet. Furthermore, a gradient descent-based backpropagation algorithm is employed to adjust the parameters of the fuzzy neural network. Theoretical analysis and forecasting examples all show that the new technique has reinforcement learning properties and universal capabilities. With respect to combined modeling and forecasting of non-stationary time series in nonlinear systems, which has some uncertainties, the method is more accurate and reasonable than other existing combining methods which are based on linear combination of forecasts.