A Large-Scale Wireless Cell Long-Term Daily-Granularity Forecasting Method

Abstract

Optimizing and managing wireless communication network, including improving the utilization of network resources, energy efficiency, automatically carrying out wireless network planning and network construction, is very important to the communication service providers (CSPs). Key performance indicators (KPIs) forecasting for wireless cells, especially the long-term forecasting task, plays a key role in wireless network planning and construction. A new adaptive combination forecasting method is proposed in this paper. The adaptive combination forecasting method has been verified by a real large-scale wireless network dataset which contains thousands of wireless cells and corresponding daily KPIs. After a series steps such as dataset analysis, and Auto-encoder algorithm, K-means algorithm and time series forecasting algorithms, we can obtain the prediction model, then compare its symmetric mean absolute percentage error (SAMPE) value with Holt exponential smoothing, Comb method and Theta method.Experimental results have demonstrated that the proposed method has a better performance, especially in the medium and long term forecasting scenario in terms of symmetric mean absolute percentage error (SMAPE) when compared with some existing methods. It proved that our method can be more suitable for complex wireless communication network environment.