基于机器学习的数据流突变型服务功能链构建策略

在SDN/NFV协同的未来网络架构下，针对增强移动宽带场景中因数据流突变造成的服务功能链低可用问题进行了研究，并提出了一种基于启发式闭环反馈算法的服务功能链动态构建策略。该算法分服务功能链的部署模块和反馈调整模块两部分。首先，基于资源优化模型实现服务功能链的初始化部署，其中对资源优化模型的求解选择利用遗传算法；然后，通过引入机器学习算法随机森林回归对当前服务功能链可承载的数据流量大小进行实时预测，以实现相应的反馈调整。整个服务功能链的构建策略是一种基于遗传和随机森林回归预测的启发式闭环反馈算法设计。仿真结果表明，在应对突变数据流时与现有的遗传和禁忌搜索算法相比，所提算法的用户接受率提高了12%，对底层资源的占用降低了19%。

Construction strategy of data stream mutation service function Chain based on machine learning

In the enhanced mobile broadband scenario under SDN/NFV network architecture, this paper studied the problem of low availability of service function chain due to data stream mutation, and proposed a dynamic service function chain construction strategy based on heuristic closed-loop feedback algorithm. The algorithm had two parts: service function chain deployment module and feedback adjustment module. Firstly, it implemented the initial deployment of the service function chain based on the resource optimization model and used the genetic algorithm to solve the optimization model. Then, it used the random forest regression algorithm to predict the data traffic that could be carried by the current service function chain to achieve the corresponding feedback adjustment. Therefore, the whole service function chain construction strategy was a heuristic closed-loop feedback algorithm based on genetic algorithm and random forest regression. The simulation results show that compared with the existing genetic and tabu search algorithms, the proposed algorithm improves the user acceptance rate by 12% and the occupancy of the underlying resources by 19%.