About One Method of Learning Sample Generation and Normalization for Time Series Extrapolation Problem in The Absence of a Priori Information about Size of Changing of Its Values in Future

摘要

Methods of learning sample generation and normalization for the case of absence of a priori information about size of changing of time series values on extrapolation interval is proposed, that is time series can arbitrarily increase and decrease. Each example of learning sample is normalized only based on values of input signal of neural network (NN). Therefore in the result of normalization similar in shape input signals are transformed to close vectors, being inputs of NN.