Research On Multipath Routing Algorithm Based On Reinforcement Learning

With the rapid development of smart cities,emerging applications such as real-time monitoring,remote control,and cloud computing have put forward higher requirements for Qo S indicators.This makes the network system as an infrastructure increasingly complex,and the network state is highly dynamic,which brings new challenges to the design of routing algorithms.Faced with the transmission pressure brought by massive traffic,the traditional single-access routing algorithm is limited by limited network resources and unbalanced resource utilization,and it is difficult to meet the demand.Therefore,multi-path routing algorithms that can fully utilize network resources have become the focus of current research.Due to the dynamic and changeable network state,the traditional multi-path routing algorithm relies on the modeling of the network environment,the parameter setting and optimization are too difficult,it is difficult to cope with the complex and changeable network environment,and it lacks the ability to dynamically adjust.Therefore,it is a major challenge to design the optimal multi-path routing according to the state characteristics of the network environment.Fortunately,with the rapid development of reinforcement learning and its successful application in many fields,reinforcement learning has become an important means to solve complex network control problems.In order to solve the current challenges of multi-path routing,reinforcement learning is introduced in this paper,relying on software-defined networks with the advantages of flexibility and programmability.Research on multi-path routing algorithms in network scenarios.First,the inter-domain network has a relatively fixed topology,and service traffic has obvious periodic changes.Therefore,this paper chooses Q-learning,which has the characteristics of fast convergence speed,low training cost,and flexible deployment as the basis,and proposes a collaborative algorithm for multi-path routing and sub-flow allocation based on Q-learning,so that it can calculate the optimal path in real time.Second,compared with the inter-domain network,the data center network topology is more complex,the link redundancy is higher,and the network status is various.This causes the state-action space of reinforcement learning to grow exponentially with the number of features,eventually falling into the curse of dimensionality.In this case,traditional table-based reinforcement learning algorithms such as Q-learning have poor convergence,and it is difficult to learn the optimal routing strategy.In this regard,this paper introduces deep reinforcement learning,and proposes a sub-flow adaptive multi-path routing algorithm based on deep reinforcement learning.The neural network learns the routing strategy in the way of function approximation to ensure the convergence of the algorithm.Third,in wireless sensor networks,the network topology will change due to node power failure.This will cause the previously trained routing model to fail in the face of the new network after topology changes.To this end,this paper combines the graph neural network with deep reinforcement learning and proposes the GRL-NET intelligent multi-path routing algorithm.The algorithm uses a graph neural network instead of a traditional neural network to build a deep reinforcement learning agent,which improves the algorithm’s adaptability in the face of topology changes.Finally,in view of the problems existing in the multi-path routing algorithm in the three scenarios,this paper designs experiments and analyzes the experimental results.The experimental results show that the algorithm proposed in this paper can effectively solve the multi-path routing problem faced by different network scenarios.