Short-term prediction of BP neural network based on difference method

Abstract

Aiming at the problem of short-term time series forecasting, a neural network based on the difference method(DMBP) is proposed. And then using sunspot data and Mackey-Glass chaotic time series data to test the performance of DMBP. In the experiment, DMBP, BP neural network algorithms, support vector regression machine (SVR),and autoregressive integrated moving average model (ARIMA) are compared in two cases with a prediction length of 2, 5. Experiment resaults show that the prediction accuracy of the DMBP algorithm is significantly improved compared to the BP neural network, and it is far better than SVR. It is equivalent to the ARIMA algorithm in short-term prediction, but the DMBP modeling process is simpler than ARIMA.