A Rare Subsequence Mining Method for Multidimensional Time Series Data Applied to Air Pollution Control

Abstract

The invention discloses a rare subsequence mining method of multi-dimensional time sequence data. The method comprises: firstly, line segment fitting is carried out on one-dimensional time sequence data; generating a line segment mode sequence, and combining line segment modes within a certain time span into vectors; obtaining vectors of all dimensions, calculating the similarity among all the vectors, utilizing the similarity to respectively cluster each dimension, obtaining a one-dimensional rare sequence set, then utilizing an improved frequent pattern tree algorithm to construct an association mode for all the one-dimensional rare sequences, and finally outputting association rules meeting related conditions.

Description

technical field [0001] The invention belongs to the technical field of data mining, and in particular relates to a rare subsequence mining method for multi-dimensional time series data applied to air pollution control. Background technique [0002] Multidimensional time series data exists widely in various fields. In finance, data such as stocks, futures, exchange rates, and interest rates are multidimensional time-series data. Compared with ordinary weather monitoring data, rare severe weather is more meaningful and valuable for research. Normalized air quality data usually appear frequently, while uncommon weather phenomena, such as severe pollution weather, are relatively rare. However, these abnormal weather does not appear randomly, and there are also some common laws. Rare subsequence analysis of air quality index data can reveal the law of rare weather and the relationship between different indicators, thus providing data support for air pollution control. Therefor...

AI Technical Summary

Problems solved by technology

However, rare patterns in multidimensional air pollution data involve complex reactions and balances, and rare patterns between each dimension may take time to form

Since the set time window size of this algorithm is only used in the first construction, the time parameters of subsequent constructions must be the same, resulting in too strict time constraints, and it is difficult to find high-level rare pattern association rules

[0008] There are many multivariate time series data in real life, but because of their different application fields, their characteristics are very different

It is not possible to find a universally applicable method

At present, the academic community has also carried out a lot of research on association rule mining of multivariate time series, but all of them are aimed at data in specific fields, and these existing methods have certain limitations.

Images