基于GCN-LSTM融合模型的自适应智能路由算法

Adaptive intelligent routing algorithm based on GCN-LSTM fusion model

海量的电力终端设备接入使得现有路由算法难以满足业务需求,因此,文中提出一种基于图卷积神经网络(Graph Convolutional Network,GCN)和长短期记忆网络(Long Short-Term Memory,LSTM)的自适应智能路由算法。首先,通过GCN-LSTM提取链路的状态特征和网络流量的时空特征,对链路的平均时延进行预测;其次,通过全连接层建立预测结果与最优路径的映射关系;最后,通过深度强化学习(Deep Reinforcement Learning,DRL)框架来训练融合模型。实验结果表明,文中所提算法能够自适应动态的网络变化,相比于常用的智能路由算法,具有更低的平均时延和较强的泛化性。

Existing routing algorithms struggle to suit business needs owing to the accessibility of enormous power terminal equipment.Therefore,this paper proposes an Adaptive Intelligent Routing Algorithm based on Graph Convolutional Neural Network(GCN)and Long Short-Term Memory(LSTM).Firstly,the state characteristics of links and the spatiotemporal characteristics of network traffic are obtained through the GCN-LSTM,and the average delay of links is predicted.Besides,the mapping relationship between the prediction results and the optimal path is established via the fully connected layer.Finally,the fusion model is trained by the Deep Reinforcement Learning(DRL)framework.The experiment results suggest that the algorithm proposed by this paper can adapt to dynamic network changes.Compared with the commonly used intelligent routing algorithms,the lower average delay and the stronger generalization are achieved.