Time-series similarity measurement method based on segmented statistical approximate representation

Abstract

The invention discloses a time-series similarity measurement method based on segmented statistical approximate representation. The method comprises the steps of feature extraction and dynamic pattern matching. First, a time series is segmented into sub series, the various statistical features of the sub series are sequentially extracted, and local pattern feature vectors are constructed; then the distance between the local pattern feature vectors is calculated by the weighted Euclidean distance, local pattern matching is achieved, the matched local pattern is used as the sub program of a dynamic programming algorithm, and global pattern matching is achieved. The method is superior to other measurement methods by a large degree on the aspects of measurement precision and calculation efficiency, and plays an important role in daily activities and industrial production of people, for example, financial transactions, traffic control, air quality and temperature monitoring, industrial flow monitoring, medical diagnosis and the like. Large scale sampling data or high-speed dynamic data flow is subjected to similarity-based search, classification, clustering, prediction, abnormal detection, on-line pattern recognition and the like.

Description

technical field [0001] The invention relates to the fields of database, data mining, machine learning, information retrieval and the like, and in particular to a time series similarity measurement method based on segmental statistical approximate representation. Background technique [0002] Time series widely exist in people's daily life and industrial production, such as real-time transaction data of funds or stocks, daily sales data in the retail market, sensor monitoring data in the process industry, astronomical observation data, aerospace radar, satellite monitoring data, real-time Weather temperature and air quality index, etc. The industry has proposed many time series analysis methods so far, including similarity query methods, classification methods, clustering methods, forecasting methods, anomaly detection methods, etc. Among them, many methods need to judge the similarity of time series, such as kNN classifier, k-means clustering method, etc. Therefore, the tim...

AI Technical Summary

Problems solved by technology

Such methods include segmented aggregation approximation, segmented linear approximation, symbolic aggregation approximation, singular value decomposition, principal component analysis, etc. The first three methods need to segment the original time series first, and then process each sub-segment separately (Segmented aggregation approximation is to calculate the average value of each segment, segmented linear approximation is to perform line segment fitting on each segment, and symbolic aggregation approximation is to discretize the average value of each segment into symbols based on segmental aggregation approximation), because The extracted features are relatively simple, which makes it less capable of expressing time series fluctuation patterns

Singular value decomposition and principal component analysis are implemented by performing a unified characteristic matrix decomposition on all time series. The typical defects of these two types of methods are high computational complexity, and the decomposition process can only be completed in memory, and the scalability of data scale is very low.

Images