Research On Resource Allocation Optimization Algorithm Of Network Based On Machine Learning

Effective and smart network management technology is a important way to improve the system performance for wireless network,especially in the condition with limited spectral resource.Therefore,this research proposes the resource allocation algorithms about the Long Term Evolution(LTE)downlink based on machine learning,which mainly deal with the problem that the actual transmission data amount is reduced due to the interaction of resource blocks in the existing scheduling algorithm,the scheduling problem under load change and the local optimal problem that may be caused by the optimal scheduling value strategy.The main contribution of this paper can be summarized as follows:(1)We propose a resource scheduling algorithm based on data variation transferred by resource blocks(RSA-DVTRB)aiming at the problem that the actual transmission data volume may be reduced due to the allocation of resource blocks.Firstly,we introduce the basic resource allocation process about LTE downlink and theoretically prove that the actual transmission data volume may be reduced due to the mutual influence on the channel quality of the resource blocks.Then we use the actual transmission data volume to be the new metric based on the consideration of complexity.Finally,we combine RSA-DVTRB and a fairness factor to construct a new algorithm named resource scheduling algorithm based on data transfer quantity and fairness(RSA-DTQF).The experimental results prove that RSA-DVTRB can achieve a higher grade than other algorithms which are good at throughput in the spectral efficiency and the RSA-DTQF strengthens the fairness of the RSA-DVTRB.(2)We propose an adaptive scheduling handoff algorithm based on load demand(ASHA-LD)based on the load demand aiming at the scheduling problem of load variation.Firstly,we analyse that different algorithms will generate different effect in network overhead and system performance.Then we train the model to verdict the condition of system load based on machine learning.Finally,we make handoff between proportion fair algorithm and RSA-DTQF based on the output of model.The experimental results prove that ASHA-LD can achieve approximate performance with RSA-DTQF based on the relatively small network overhead which means that the model can effectively judge the load condition of the system.(3)We propose a resource allocation algorithm based on reinforcement learning(RAOA-RL)in order to explore the more effective strategy.Firstly,we construct Markov decision process based on the characteristic of downlink resource allocation.Then we train the resource allocation RL model by using Deep Q-learning network(DQN).Finally,we also construct a secondary scheduling way and verify the performance of RL model by simulation.The experimental results prove that RL model can effectively improve the system performance and the secondary scheduling way can also change the optimized focus.