Mobile Network Traffic Prediction Based On Cross-Patch Feature Fusion

• Published in: 2024 4th International Conference on Neural Networks, Information and Communication (NNICE) pp. 1006 - 1010
• Main Authors: Zhou, Jiangxin, Luo, An, Zhou, Ning
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.01.2024
• Subjects: Analytical models; Computational modeling; Mobile Network Big Data; Network Traffic Forecast; Predictive models; Quality of service; Telecommunication traffic; Time series analysis; Time Series Forecasting; Training
• DOI: 10.1109/NNICE61279.2024.10498199

Time Series Prediction (TSP) has been effectively employed across a broad spectrum of fields since its inception and has also found success within the realm of mobile network traffic analysis. Mobile operators typically rely on the forecasts of mobile network traffic to undertake network planning and resource allocation, ultimately enhancing the quality of service provided to users. Conventional network traffic prediction methods often consider the traffic data of a single time point as a basic element and compute the dot product with other points to create an attention matrix. However, such methods are prone to the influence of random data fluctuations and usually result in extended training times. To overcome these limitations, this article introduces a novel Patch-based neural network model for predicting mobile network traffic. Traffic data is decomposed into trend and seasonal series by employing multiple sliding windows instead of one. Furthermore, the decomposed series are divided into multiple patches using different size scales, with each patch forming the base unit for computing the attention matrix, thereby reducing the impact of random data.In our experimental analysis, we conducted a comparative analysis of the predictive performance between the model proposed in this paper and the current state-of-the-art models. The results indicate that, compared to various existing Time Series Prediction (TSP) models, our proposed model exhibits superior accuracy in predicting mobile network traffic. This enhanced predictive capability highlights the potential for its effective application in mobile network traffic analysis.