Method and device for predicting change trend of high-dimensional recurrence concept drift flow data

Abstract

The embodiment of the present invention provides a method and device for predicting the change trend of high-dimensional reproduction concept drift flow data, which belongs to the technical field of flow data change trend prediction. The method includes: obtaining a real-time time-series data stream representing the state change of the target variable; taking the real-time time-series data stream as input, and outputting the final forecast time-series data stream representing the future state change of the target variable through the concept drift prediction model; The training set is obtained by offline training of the D-LSTM neural network, and online autoregressive prediction training of the D-LSTM neural network according to the real-time time series data stream. The present invention effectively improves the prediction accuracy of the change trend of the high-dimensional reappearance concept drift flow data by integrating the deep learning algorithm and the online prediction method, and solves the problem of the traditional static data change trend prediction method on the high-dimensional reappearance concept drift flow data change trend Predicting problems with limited generalization ability.

Description

technical field [0001] The invention relates to the technical field of flow data change trend prediction, in particular to a change trend prediction method for high-dimensional reproduction conceptual drift flow data and a change trend prediction device for high-dimensional reproduction conceptual drift flow data. Background technique [0002] With the advent of the era of big data, many data generated in the form of streams are flooding all aspects of life, such as data generated by sensors and other devices in factories, constantly changing social network data composed of users of social networking sites, and data in the stock futures market. Real-time transaction data, etc. Compared with static data, streaming data has many difficulties. For example, due to the long duration of streaming data, it is easy to change the distribution of data and cause concept drift, which affects the prediction accuracy of data, and reproduces concept drift. It is a type of concept drift, wh...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the embodiments of the present invention is to establish a concept drift prediction model by performing offline training and online autoregressive prediction training on the D-LSTM neural network, so as to solve the problem that the existing methods often use offline methods to obtain all training data and then implement learning and analysis of the prediction model. Training, leading to the problem that the accuracy of existing methods is gradually reduced when they are directly used to deal with streaming data with concept drift

Images