New routing strategies are urgently needed to control exploding traffic and provide precise routing for the next-generation complex network. Traditional routing methods cannot satisfy its routing requirements of this complex traffic control situation. One of the important reasons is that it does not consider learning from traffic features. Using deep learning method to predict routing path is an emerging and promising solution, however, the existing deep learning-based methods in network traffic control field still have some disadvantages, such as low routing accuracy, high routing time complexity and need large data set to train complex deep learning system. To solve these problems, this paper proposes a block-based deep learning intelligent routing strategy (DLBR strategy), which divides the network into multiple sub-blocks according to a recursive partition method and uses three deep learning models to train and test them. Experiments show that the proposed network DLBR strategy has the ability to combine with different deep learning intelligent structures, and achieves higher accuracy and lower time complexity under the training of smaller training data.

迫切需要新的路由策略来控制爆炸的流量并为下一代复杂网络提供精确的路由。传统的路由方法无法满足这种复杂的流量控制情况的路由要求。重要原因之一是它不考虑从交通功能中学习。使用深度学习方法预测路由路径是一种新兴且有前途的解决方案，但是，现有的基于深度学习的方法在网络流量控制领域仍然存在一些缺点，例如路由精度低，路由时间复杂度高以及需要大数据集。训练复杂的深度学习系统。为了解决这些问题，本文提出了一种基于块的深度学习智能路由策略（DLBR策略），它根据递归分区方法将网络分为多个子块，并使用三种深度学习模型进行训练和测试。实验表明，所提出的网络DLBR策略具有与不同的深度学习智能结构相结合的能力，并且在较小的训练数据训练下可以实现较高的准确性和较低的时间复杂度。