Fuzzy time series granulation methods for data analysis

The growing dimensionality of data, driven by the multitude of heterogeneous time series, requires the development of efficient methods for their processing and compression. This article presents an approach to data compression where the data are represented as time series, using granulation with fuzzy logic methods. The study analyzes average daily temperature data in the Republic of Tatarstan collected from 1881 to 2024. Data granulation enabled a significant compression of the data volume. Fuzzy summarization was applied to transform the original numerical data into information granules, facilitating the automatic generation of granular descriptions of time series behavioral patterns. The summarization of time series states was carried out using fuzzy logic methods, including a rule set, membership functions for each season, interval-based linguistic variables, and a defuzzification software module. The implementation of the proposed approach demonstrated a reduction in data volume from 52,534 to 7,504 points, achieving a compression ratio of approximately 85%. The developed methods are applicable for analyzing large datasets across various domains.