Nowadays, more and more electric power services are carried on the power information communication network (PICN) including power grid production and scheduling, communication, and environment sensing, in the form of data, voice and video. To improve the resource utilization efficiency, it is necessary to carry out traffic prediction approach in PICN. However, the accessing businesses have diversified characteristics, which are reflected to different types of traffic flow in PICN. Moreover, the traditional PICN is a distributed network and cannot be controlled flexibly, which leads to the poor accuracy of traffic prediction algorithm. To address these problems, we combine the Software Defined Networking (SDN) architecture and Radial Basis Function neural network (RBFNN) for traffic intelligent prediction in PICN. The SDN controller can acquire global knowledge of PICN in each time slot to guide the data sampling process. Further, the complex nonlinear relationships of large-scale network traffics are analyzed by RBFNN model to realize high-precision traffic identification. The proposed scheme is evaluated based on by POX and Mininet platforms. Simulation results show that the proposed SDN-based intelligent prediction scheme can accurately forecast the change trend of each traffic flow and has better performance and lower prediction error than current schemes.

如今，越来越多的电力服务以数据、语音和视频的形式承载在电力信息通信网络（PICN）上，包括电网生产调度、通信、环境感知等。为了提高资源利用效率，有必要在PICN中进行流量预测方法。但是接入业务具有多样化的特点，这体现在PICN中不同类型的业务流上。而且，传统的PICN是分布式网络，无法灵活控制，导致流量预测算法精度较差。为了解决这些问题，我们结合软件定义网络 (SDN) 架构和径向基函数神经网络 (RBFNN) 在 PICN 中进行交通智能预测。SDN控制器可以在每个时隙获取PICN的全局知识来指导数据采样过程。进一步利用RBFNN模型分析大规模网络流量的复杂非线性关系，实现高精度流量识别。所提出的方案是基于 POX 和 Mininet 平台进行评估的。仿真结果表明，所提出的基于SDN的智能预测方案能够准确预测各个交通流的变化趋势，比现有方案具有更好的性能和更低的预测误差。