Time series forecasting using ensemble of AR models with time-varying structure

In this paper an algorithm of nonstationary time series forecasting, using ensemble of autoregressive models with time-varying structure is described. The technique of recursive splitting is implemented in estimating of non-stationary autoregressive model. The procedure of individual models combining using fuzzy logic controller is defined. The feasibilities of the approach are demonstrated in time series forecasting on a set of time series with different time frequencies.