Research On Machine Learning Based Traffic Prediction And Base Station Sleep Mode

With the rapid development of information technology,the demand for data services in wireless communication systems is increasing exponentially.In order to meet business needs,the number of network infrastructures such as macro base stations and small base stations has increased significantly,and the network structure has also developed into the heterogeneous network.A large number of low-power access nodes are deployed around macro base stations,which effectively improves network capacity.But at the same time,wireless communication systems are also facing more severe energy consumption issues.The wireless network traffic continuously changes with time and has an obvious "tidal effect".In the period of low load,all base stations are kept on,causing unnecessary energy consumption.Creating an energy-efficient wireless communication system and realizing green communication have become the key research directions of the communication industry today.This thesis is put forward at this technical background.With the goal of improving network energy efficiency,according to the characteristics of network traffic,researches on network traffic prediction technology and base station sleeping technology.Firstly,the characteristics of network traffic and network traffic prediction models are introduced in categories,and the network energy efficiency metrics are summarized according to the component level,equipment level and network level,and the base station sleeping technology is analyzed.Secondly,based on in-depth research on network traffic prediction technology,this thesis proposes a Time Varying Filtering Based Empirical Mode Decomposition(TVF-EMD)and LongShort-Term Memory(LSTM)combined network traffic prediction model.The TVF-EMD algorithm can simplify and decompose complex network traffic data into multiple intrinsic mode functions(IMF).Using the "memory characteristics" of the LSTM network and good prediction ability for long-range dependence problems,the LSTM model is constructed and predicted for each IMF component.And then the prediction results of each IMF component are superimposed and reconstructed to form network traffic prediction results.Simulation results show that compared with the traditional LSTM prediction model,the TVF-EMD-LSTM prediction model has smaller prediction error and better performance evaluation index performance.Finally,the network traffic prediction technology is combined with the base station sleep technology.Aiming at the time fluctuation of network traffic load and aiming at improving the energy efficiency of the network,a traffic load prediction based base station sleep mode algorithm is proposed.Through the joint analysis of network load prediction data and real-time network load conditions,sleeping and wake-up operations on micro base stations in the network can be performed.The algorithm includes three parts: sleep threshold preset,base station sleeping algorithm based on prediction and base station dynamic wake-up algorithm.In a heterogeneous cellular network scenario where macro base station and pico base stations are co-deployed,Classify the load level of the network in advance,and set the sleep ratio of the base station corresponding to each load level.At a sleep operation time point,the sleep operation is performed according to a preset threshold value based on the network load prediction,real-time load situation,and user switching situation.Until the next sleep operation time point,sleep settings are restarted.In the case of a sudden increase in the traffic load,in order to ensure the quality of service,the sleeping base station can be dynamically woken up in real time.Simulation results show that the proposed algorithm can reduce network energy consumption and improve network energy efficiency while maintaining the quality of service(QoS)and keeping interruption probability at alow level.