Abstract:
With traffic demand, the number of services carried in optical transport networks is increasing. However, due to some internal factors (e.g., device power failure, aging)...Show MoreMetadata
Abstract:
With traffic demand, the number of services carried in optical transport networks is increasing. However, due to some internal factors (e.g., device power failure, aging) and external factors (e.g., natural disasters), the link in the network may be damaged and some related services can be affected. Restoring these affected services efficiently is a challenging problem. In this paper, we convert the information of optical transport networks (with ODU-k switching capability) and the affected services into images (named a multi-modal method with image format). Then we propose a reinforcement-learning-based service restoration (RLSR) algorithm. In our experimental setup, RLSR uses advanced image recognition model MobileNetV2 and advantage actor-critic reinforcement learning algorithm. Numerical results show that the proposed method can achieve a higher ratio of service restoration than the benchmark and the restoration time is within the acceptable range (\sim0.086984s).
Date of Conference: 17-20 February 2020
Date Added to IEEE Xplore: 30 March 2020
ISBN Information:
Print on Demand(PoD) ISSN: 2325-2626