A time series prediction algorithm based on linear fuzzy information granule
[Objective]Our research aims to tackle the prevalent challenge of time series prediction in diverse fields such as economics,finance,environment,and ecology.With the continuous growth in the scale of time series data due to advancements in computer and IoT technologies,predicting these large-scale sequences has become increasingly complex.Primarily,we attempt to explore innovative approaches for constructing fuzzy information granules that accurately reflect both dataset size and trend information.In this study,we provide a robust solution to inherent difficulties in forecasting large-scale time series by enhancing the efficiency of prediction algorithms through the strategic design of these granules.[Methods]The research employs a multi-faceted methodology.Initially,the study establishes comprehensive distance definitions for various fuzzy information granules,including interval,triangular,and Gaussian types,based on the fuzzy extension principles.Subsequently,a novel class of fuzzy information granules is introduced,and the central line segment and dispersion of the dataset are considered.These granules effectively capture the development trends and dispersion characteristics of time series within a specified timeframe.In the study,we further present a functional expression and geometric interpretation for the distance of Gaussian fuzzy information granules.To operationalize these granules for time series prediction,we design a fuzzy inference prediction system,and leverage historical data as well as rules extracted from Gaussian fuzzy information granule distances.[Results]The functional expression for Gaussian fuzzy information granule distance constitutes a concise mathematical representation,allowing for a reasonable interpretation as the amalgamation of disparities in central lines and deviation degrees.Then,the fuzzy inference prediction system,in which Gaussian fuzzy information granule distances is utilized,successfully extracts effective rules from extensive historical data,facilitating long-term predictions for time series.Results emphasize the superiority of the proposed approach in terms of root-mean-square error and mean-absolute-percentage error,thus highlighting its potential for improving the accuracy of long-term time series predictions.Comparative analyses against various numerical prediction algorithms and alternative fuzzy information granule inference methods,including autoregressive models,autoregressive neural networks,and regression vector machines,consistently demonstrate enhanced outcomes achieved by combining linear Gaussian fuzzy information granules with the fuzzy inference system.[Conclusions]Our study provides a comprehensive exploration of challenges inherent in time series prediction and proposes a methodology to address these challenges effectively.The designed fuzzy information granules,informed by meticulous distance definitions and consideration of dataset characteristics,offer results for accurate and efficient long-term time series predictions.Satisfactory results in time series prediction suggest that,by skillfully designing information granules,we can accurately capture key features in the dataset,thereby enhancing the efficiency of other data mining tasks.This outcome includes improvements such as fast computational speed and accurate results.
linear fuzzy information granulefuzzy inference systemtime series prediction
万方数据
维普
