General Electric Company
MULTIVARIATE TIME-SERIES DATA SEARCH

Abstract:

The example embodiments are directed to a system and method which can perform a text-based search for a temporal data pattern in time-series data. The text-based search process is significantly faster than a distance measurement-based search performed based on temporal pattern comparisons. In one example, the method may include storing previously recorded temporal patterns of time-series data, determining a set of optimal bin boundaries based on the previously recorded temporal patterns, where the set of optimal bin boundaries divide the observed range of time-series data into a plurality of discrete bins each labeled with a respective symbol, transforming the previously recorded temporal patterns of time-series data into symbol strings based on the set of optimal bin boundaries, where a symbol string is based on data points in the plurality of discrete bins, and storing the symbol strings within a symbol storage.