Time series data prediction method and system, medium and equipment

Abstract

The invention relates to a time series data prediction method and system, a medium and equipment, and the method comprises the steps: carrying out data augmentation of an original data set when the data size of the obtained original data set is smaller than a preset value, and obtaining a plurality of expansion sets meeting the data demands of building a prediction model; establishing a predictionmodel corresponding to each expansion set, and determining a covariance of a prediction error of each prediction model; predicting the tested data set by using all the prediction models to obtain a prediction result of each prediction model; and fusing the prediction results of all the prediction models in a covariance intersection mode to obtain a final prediction result. The data of the small-scale time sequence is augmented to obtain the plurality of expansion sets meeting the data requirement of establishing the prediction model, the corresponding prediction model is established for eachexpansion set, multi-model parallel prediction is carried out, and multi-model fusion is carried out based on error estimation, so that reliable prediction of the small-scale time sequence data is realized.

Description

technical field [0001] The present invention relates to the technical field of data processing, in particular to a time series prediction method, system, medium and equipment. Background technique [0002] Data has become an important resource for various artificial and natural systems in the modern information age. For various information management systems, on the one hand, it is necessary to pay attention to the real-time data generated by equipment and sensors to monitor the operating status of the controlled objects. Predictive awareness of data trends is required to make early decisions and take action to adjust system operations. Therefore, data prediction has received attention and research applications in unmanned system control, environmental monitoring, stock market and other fields. [0003] Big data in the Internet era has become the basis of data mining analysis, but in practical applications, time series data sets often cannot reach the expected scale. The s...

AI Technical Summary

Problems solved by technology

The above situation and the slow data collection in other information systems make the data set very small. At the same time, the time series is characterized by non-stationary, nonlinear, and complex noise.

[0004] For the currently commonly used deep learning networks, etc., small-scale data cannot meet the requirements of model training, specifically reflected in: due to the limited amount of information in the existing data, the models fitted by statistical methods such as ARIMA and ARCH cannot track new patterns ;The amount of data directly affects the effect of network training. If the amount of data is too small, it will not be possible to use machine learning methods to build models

Images