Interval-valued data forecasting using dual-parametric neural networks

Abstract

This study is focused on the approach to modelling and forecasting interval-valued data using a dual-parametric neural network (DPNN). This concept was proposed as a subclass of interval neural networks that contains two types of parameters: real and interval ones. This approach makes it possible to get guaranteed inclusion of an exact (single value) solution into interval calculation results. In this paper we intend to give a theoretical overview of previous research on the subject and description of the new developed methods and algorithms for learning DPNN.  The experiments demonstrate that interval calculation results obtained by using the proposed approach include of exact solution at least in 60% of cases.