Abstract:

With the emergence of new services and the continuous evolution of IP network technology, cloud-network convergence has entered a new stage, showing the development characteristics of digitization, intelligence and servitization. Intelligence needs to be combined with relevant artificial intelligence technologies.Deep learning and deep reinforcement learning are commonly used artificial intelligence algorithms. With the development of graph neural network and other technologies, the ability of deep learning to represent graph information and the ability of deep reinforcement learning to deal with optimization problems have been improved. IP networks can be represented abstractly by using graph structures, and related prediction and optimization problems can be processed and solved by using deep learning and deep reinforcement learning algorithms. Therefore, this paper describes the related algorithms and applications of deep learning and deep reinforcement learning in three scenarios including traffic prediction, network planning and traffic engineering, and analyzes the possible problems and challenges that may occur in practice.

Key words: deep learning, deep reinforcement learning, traffic prediction, network planning, traffic engineering, cloud-network convergence