Traffic matrix recovery prediction method and system based on automatic encoder

Abstract

The invention belongs to the technical field of traffic matrix prediction, and particularly relates to a traffic matrix recovery prediction method, system and platform based on an automatic encoder. The method comprises the following steps: acquiring a traffic matrix data set; sequentially dividing into three subsets according to the acquisition time of the traffic matrix; a sliding window method is used for generating data samples for model training, meanwhile, a corresponding mask mark matrix is randomly generated for each traffic matrix according to a specific proportion, whether data values are missing or not is marked, and a data sample set capable of being trained is obtained; and training the constructed automatic encoder network model by using the obtained data sample set, and adjusting model parameters according to errors between a recovered traffic data value and a missing true value and between predicted traffic matrix data and a predicted target value so as to obtain a final recovery and prediction model of the traffic matrix. The traffic matrix acquisition cost can be reduced, and the traffic matrix prediction performance can be improved.

Description

technical field [0001] The invention belongs to the technical field of traffic matrix prediction, and in particular relates to an automatic encoder-based traffic matrix recovery prediction method, system and platform. Background technique [0002] The traffic matrix represents the traffic value of all nodes in the network in pairs to form a stream from the source end to the destination end within a certain time interval. Accurately predicting the future traffic matrix plays an important role in network management, especially in traffic engineering, traffic scheduling or rerouting, capacity planning, network design, resource planning, network anomaly detection, and quality of service. By collecting the traffic matrix and accurately predicting the future traffic matrix, the network manager or operator can predict the possible traffic demand of the network in advance and make corresponding decisions in advance, thereby saving network costs and optimizing network resources. How...

AI Technical Summary

Problems solved by technology

However, such a method is difficult to truly fill in missing flow values, and the prediction performance of the model is limited.

[0004]Therefore, in view of the technical problems of the above traditional time series forecasting technology with limited performance and difficulty in simulating the nonlinear characteristics of network traffic, it is urgent to design and develop an autoencoder based The traffic matrix recovery prediction method, system and platform of

Images