Synthesis of a Two Cascade Neural Network for Time Series Forecasting

The paper suggests the method for the synthesis of a two-cascade neural network for time series forecasting. The first cascade contains the basic (popular) time series forecast models with the optimal parameters for the given forecast step that allow the implementation in neural basis. The first cascade includes the basic models from the given system of models that are important for the improvement of the forecast quality. Each model of the first cascade of the neural network determines one coordinate of the weight vector of the neural element of the second cascade that is found as a result of their joint learning.