Research On Handover Optimization Based On Machine Learning

With the widespread use of various mobile terminals and Internet of things(IoT)terminals and other user equipments(UEs),as well as the continuous expansion and popularization of various Internet data services,mobile data traffic grows exponentially.Ultra-dense network(UDN)is one of the effective technical that meets the requirement of high experience rate,high peak rate,and extremely high traffic of UEs in the future.How to face the mobility management challenge in UDN and enhance the quality of service(QoS)of UE are the key challenges that need to be solved urgently at present.Consequently,to reduce Handover(HO)latency,ensure business continuity and enhance experience of UE in UDN scenarios,this paper studies and designs effetely HO algorithm in mobility management.First,from the perspective of architecture,this paper introduces the network architecture model of wireless network based on software defined network(SDN).In UDN,the concept of SDN is introduced,because the traditional network architecture of Internet protocol cannot solve HO redundancy in UDN(that is,the HO delay caused by excessive HO signaling is too long,which affects the QoS of UE),thus the interaction of HO signaling is reducedSecond,in the SDN-based network architecture,a HO assistance strategy based on machine learning is proposed to optimize the HO signaling process,which predicts the probability of HO success between base station(BS)in different frequency bands based on extreme gradient boosting algorithm.The results show that the strategy based on SDN proposed in this paper can improve the HO success rate between two BSs with different frequencies,enhance the QoS of UE,promote the load balance between networks,and effectively improve network capacity.Finally,considering the deficiency that the hysteresis(Hys)and time-to-trigger(TTT)parameters are fixed in the HO process,this paper proposes an intelligent adaptive adjustment algorithm for HO parameters.After selecting a suitable target BS according to the BS selection algorithm,deep reinforcement learning is applied to the traditional HO decision-making to dynamically adjust Hys and TTT.This means that the signal levels from the serving and target BS are both placed in the context of dynamically adjusted Hys and TTT,which are the outputs of the SDN controller based on deep reinforcement learning.The simulation results show that the proposed algorithm can obtain the best HO trigger point,to minimize HO failures and ping-pong HOs while ensuring throughput.