A Hybrid Linguistic Time Series Forecasting Model combined with Particle Swarm Optimization

Abstract

Linguistic time series forecasting model (LTS-FM) which is proposed by Hieu et al. by utilizing hedge algebras theory is an efficient forecasting model. Instead of partitioning the universe of discourse (UD) of the linguistic variable into subintervals and assigning fuzzy sets to them, it establishes a formalism to convert historical numeric time series data into linguistic one based on numerical semantics of words which are transformed from the semantically quantifying mapping (SQM) values of the corresponding words. Therefore, a LTS-FM is established in such a way that it handles directly words of linguistic variable and their qualitative semantics. However, the fuzziness parameter values of the LTS-FM which determine the SQM values of words are currently specified by human experts, so the forecasted results may not be optimal. This paper proposes a hybrid LTS-FM in which particle swarm optimization is utilized to optimize the fuzziness parameter values. A new formula of computing crisp forecasted values is also proposed. The experimental studies carried out over two practical forecasting problems of the enrollments of University of Alabama and killed in car road accident in Belgium show that the proposed forecasting model obtains better forecasted results.