Unsteady-state Granger causality mining method for discrete time series data

Abstract

The invention discloses an unsteady-state Granger causality mining method for discrete time series data, and the method comprises the steps: firstly obtaining a space-time sequence data set, buildinga Hox model, initializing the Hox model, and labeling a time sequence with a category label; learning model parameters and Granger causality corresponding to each category of data through a Hawkes-EMalgorithm, and optimizing category classification of the spatio-temporal sequence data based on a greedy algorithm; calculating a final score of the hux model; repeating the steps S2 and S3 until thefinal score value meets a preset standard, wherein the category division situation of the spatio-temporal sequence data corresponding to the model parameters and the Granger causality obtained by mining the corresponding category are optimal solutions. According to the method, on the basis of an original Hawkes-EM algorithm, a greedy algorithm is combined to enable the original Hawkes-EM algorithmto become an unsteady-state Granger causality mining method of discrete time series data, so that data belonging to different categories in a section of discrete time series data is found out, and the Granger causality represented by the discrete time series data in the corresponding category is found out.

Description

technical field [0001] The invention relates to the technical field of data mining, in particular to a non-stationary Granger causality mining method for discrete time series data. Background technique [0002] With the progress of society and the development of science and technology, the things people need to know become more and more complex. The causal relationship within the system exists objectively. Causal discovery is to mine the causal relationship contained in the data, so as to help people understand the relationship between things. objective law, while Granger causality X t-1 →Y t Can be used to predict the occurrence of events. For example, in the design of TV program recommendation system, if there are two time series data X t-1 People watch sports and Y at time t-1 t People tune in for entertainment at t times our learning to x t-1 →Y t From this relationship, we can conclude that people tend to watch entertainment programs after watching sports program ...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the existing discrete time-series data based on the Hawkes Granger causality discovery method, the present invention believes that there is only one stable Granger causality under a piece of discrete time-series data, which cannot correctly reflect the causality of the time-series data , provides a non-stationary Granger causality mining method for discrete time series data

Images