Predicting and modelling of nonstationary temporal signals with fractal characteristics

This paper presents a scheme for predicting and modelling of nonstationary signals possessing fractal characteristics, using a resource-allocating network (RAN). One significant feature of a RAN is its ability to allocate resources corresponding to the complexity of nonstationary signals, thus tracking and matching the complexity of nonstationary signals can be achieved. The experimental results of predicting chaotic time series and short-term power load have shown RAN is suitable for modelling and predicting such nonstationary signals with the fundamental advantage of complexity matching and tracking capability.