Research On Dynamic Resource Allocation Based On Deep Reinforcement Learning In Optical Network

With the continuous development of new services,optical network is facing huge challenges.Compared with traditional services,new services are dynamic and unpredictable,and they put forward more stringent requirements for delay and bandwidth.Due to the limitation of bandwidth,traditional optical transmission technology can’t meet the increasing business needs.Abundant dynamics also make the allocation of dynamic resources in optical network more complicated.As an emerging information technology in intelligent manufacturing,digital twin(DT)is specialized in achieving low-cost and high-efficiency physical equipment dynamic modelling and controlling.Meanwhile,deep reinforcement learning(DRL)is proved to be capable of sensing complex environment states and learning the best policies through real-time interactions with the environment,which can be used as the enabling technology for digital twin.This paper analyzed the possible application directions of digital twin technology in optical network,described a digital twin architecture in optical network,and pointed out the advantages of combining deep reinforcement learning with digital twin.is based on the digital twin technology enabled by deep reinforcement learning,through the configuration of programmable optical transceiver(POT),the dynamic resource allocation in optical network is realized.The main innovations of the paper are as follows:First,in the access network,in order to ensure the delay performance,services usually occupy a large bandwidth,which causes bandwidth waste.For the access network environment,a DRL-enable DT POT configuration scheme in the optical access network is designed.Based on the DRL algorithm,the learning promotion model is implemented.The scheme can adapt to changes in network conditions and physical conditions,and optimize network delay and bandwidth occupancy.The simulation results show that the spectrum occupation is reduced by 19.4%at the cost of 0.7%delay increase.Finally,the adaptive configuration of the programmable optical transceiver in the access network is realized.Second,in the transmission network,we are eager to improve the bandwidth utilization of the transmission network and improve the throughput of the whole network.Based on the DT enabled by DRL,a POT configuration scheme in the optical transport network is proposed and implemented.Experimental results show that compared with the traditional scheme based on static control model and neural network,the proposed scheme can reduce the blocking rate by 7.0%on average and increase the throughput by 15.5% on average.