Multichannel time-series modelling and prediction by wavelet networks

Multichannel time-series result from observations of a given engineering, biomedicai, econometric or environmental variable taken at different locations. Processing this type of signal presents problems associated with its extrapolation in given space ranges and its possible prediction. This paper presents a comparison of seasonal AR modelling of such signals and the application of wavelet networks to the system identification and prediction of a particular signal. The choice of wavelet functions and the optimization of their coefficients is discussed as well. Each method suggested in the paper is verified for simulated signals at first and then used for the analysis of real signals, including the observation of air pollution. All algorithms are written in the MATLAB environment.