Non-stationary time sequence data prediction method and system, storage medium and computer device

Abstract

The invention relates to a non-stationary time sequence data prediction method, which comprises the following steps of: obtaining and decomposing original time sequence data to obtain three sub-sequences with different rule characteristics; selecting matched prediction models for the sub-sequences respectively, and analyzing and predicting the sub-sequences through the matched prediction models respectively to obtain prediction results of the sub-sequences; and integrating the prediction results of all the sub-sequences to obtain a prediction result of the original time sequence data. According to the method, the prediction result of the original sequence data is obtained by decomposing the original time sequence data and respectively predicting the subsequence and fusing the subsequence prediction result, the rule characteristics of each subsequence are comprehensively considered, the prediction accuracy is high, and processing and prediction of any non-stationary time sequence data can be realized. The invention further provides a non-stationary time sequence data prediction system, a storage medium and computer device.

Description

Technical field [0001] The present invention relates to the technical field of time series prediction, in particular to a method, system, storage medium and computer equipment for predicting non-stationary time series data. Background technique [0002] With the widespread application of sensor networks and handheld mobile devices and the rapid development of computer technology, people can obtain a large amount of time series data. These time series data change dynamically with time, and most of them are non-stationary and noisy, with a large degree of particularity and complexity. How to dig out hidden time series patterns from these time series data and analyze these patterns to extract valuable information and use it in practice is a great challenge to time series data mining and prediction technology. [0003] Time series forecasting technology has been widely used in dealing with uncertainty and further supporting decision-making, especially in fields involving time measurem...

AI Technical Summary

Problems solved by technology

[0004] Time series forecasting techniques in the prior art include: The first category is forecasting methods based on statistical probability. This method can only obtain better forecasting results under the assumption or condition that the time series has linearity and stationarity, but it is not suitable for nonlinear Data modeling is not ideal

The second category is the prediction method based on neural network, which can model complex nonlinear time series data, but limited by the amount of data, if the amount of data is not large enough, the modeling is still not accurate enough

In addition, if the time series data contains noise, it will affect the convergence of the model training and easily cause the model to overfit

Images