基于OS-MBRL的网络切片资源动态分配方法研究

Research on dynamic allocation of network slicing resources based on OS-MBRL

随着网络用户业务需求的增长,如何实现网络切片动态和准确的资源分配是当下网络必须解决的问题。考虑传统无模型强化学习方法需要较长的模型训练时间,提出了一种基于OS-MBRL(model based RL supported by online SVM)的网络资源动态分配方法。该方法利用在线支持向量机算法构建了一个系统模型,保证在分配较少资源的情况下产生较低的服务等级协议(service level agreement,SLA)违规次数。仿真实验结果表明,与归一化优势函数(normalized advantage function,NAF)算法、深度Q网络(deep Q-network,DQN)算法和双延迟深度确定性策略梯度(twin delayed deep deterministic dolicy gradient,TD3)算法相比,该方法能够最高减少80%的SLA违规次数,同时降低9%的资源分配。

With the growth of business needs of network users,how to achieve dynamic and accurate resource allocation of network slicing is a problem that must be solved in the current network.Considering that traditional modelless reinforcement learning methods require a longer model training time,a dynamic resource allocation method based on OS-MBRL was proposed.The online support vector machines algorithm was utilized to construct a system model that could handle dynamically changing data streams and continuously update the model to adapt to new data,ensuring a lower number of SLA violations when allocating fewer resources.Simulation experiment results show that compared with NAF algorithm,DQN algorithm,and TD3 algorithm,the proposed method can reduce SLA violations by up to 80%and resource allocation by 9%.