In this paper we developed a new structure of wavelet network based on the traditional one. The developed wavelet network is called “polymorphic”. To determine the capability of the wavelet network to approximation of nonstationary time series a new metric was introduced. This metric is equivalent to signal-to-noise ratio. Polymorphic wavelet network has advantages of the time-series approximation capability and learning rate.
Keywords: нейронная сеть, вейвлет-сеть, аппроксимация нестационарных временных рядов, обучение нейронных сетей
