The Modeling of Interval-Valued Time Series Using Possibility Measure-Based Encoding-Decoding Mechanism

Abstract

Interval-valued time series (ITS) is a collection of interval-valued data whose entires are ordered by time. The modeling of ITS is an ongoing issue pursued by many researchers. There are diverse ITS models showing better performance. This paper proposes a new ITS model using possibility measure-based encoding-decoding mechanism involved in fuzzy theory. The proposed model consists of four modules, say, linguistic variable generation module, encoding module, inference module and decoding module. The linguistic variable generation module can provide a series of linguistic variables expressed in fuzzy sets used to described dynamic characteristics of ITS. The encoding module encodes ITS into some embedding vectors with semantics with the aid of possibility measure and linguistic variables formed by linguistic variable generation module. The inference module uses artificial neural network to capture relationship implied in those embedding vectors with semantic. The decoding module decodes for the outputs of the inference module to produce the output of linguistic and interval formats by using the possibility measure-based encoding-decoding mechanism. In comparison with existing ITS models, the proposed model can not only produce the output of linguistic format, but also exhibit better numeric performance.