Short-time data self-adaptive prediction method and correction method

Abstract

The invention relates to a short-time data self-adaptive prediction method and a correction method. The method comprises the following steps: a, performing first-order difference on a sequence in an observation time sequence window; b, performing polynomial fitting on the difference sequence, and calculating a prediction increment P1 based on polynomial fitting; calculating a prediction increment P2 based on time sequence analysis according to the influence factor of the first-order difference sequence; c, calculating a comprehensive prediction increment according to P1 and P2; d, summing the comprehensive prediction increment and the real value queue tail value to obtain a prediction value P; and e, carrying out adaptive adjustment on the time sequence window according to the relative error condition of the real value and the predicted value. According to the invention, long-term data trend and short-term data trend and fluctuation are considered at the same time, polynomial fitting and random sequence analysis are used for length complementation, the trade-off of the long-term data trend and the short-term data trend and fluctuation is changed in the mode that the real value sequence length of data is self-adjusted, a good effect can be achieved, and the invention is more robust to unsmooth and irregular data and data with random noise.

Description

technical field [0001] The invention relates to the technical field of data anomaly detection and correction of real-time response equipment, in particular to a short-term data self-adaptive prediction method and correction method. Background technique [0002] In real-time response equipment, data anomaly detection and correction are required to ensure the normal operation of the equipment and have a certain predictive function. Anomaly detection and prediction of time series is a regular demand in various industries and industries, and various methods and ideas emerge in endlessly. Classified according to different anomaly detection problems, including the following methods: [0003] Time series method: moving average, year-on-year and month-on-year, time series indicators; [0004] Anomaly detection (STL+GESD) statistical method: single feature conforms to Gaussian distribution, multiple irrelevant features conform to Gaussian distribution, multiple features are correla...

AI Technical Summary

Problems solved by technology

[0007] Although there are many existing methods, there is still a lack of methods with relatively low computational complexity, relatively robust effects, and relatively clear statistical significance, especially when the intrinsic nature or prior information of the data is not clear.

Images