Knowledge-Driven Submarine Optical Cable Routing Based on RL Framework With Embedded DNN

Authors:

Abstract
Submarine optical cable is a critical infrastructure for transoceanic communications, and route planning is an effective method to optimize its economy and survivability. Meanwhile, the current planning scheme fails to utilize the prior engineering knowledge that holds significant reference value. We propose a new paradigm for planning submarine cable routes based on engineering knowledge, in which a knowledge-driven deep neural network (KD-DNN) is used to extract engineering design knowledge from existing expert-designed submarine optical cable routes. The KD-DNN is then embedded into a reinforcement learning framework (Emb-DNN-RL) to design new submarine cable routes. The simulation results show that the differences in accumulated cost and length for the submarine cable route designed by Emb-DNN-RL are 6.75% and 6.21% of those of the route designed by expert, respectively. The Fréchet distance between the route designed by Emb-DNN-RL and the one designed by experts is 0.48 km. The submarine cable route designed by Emb-DNN-RL is highly similar to the one designed by expert. The deviation angle of candidate points, and the distance between the current location and the endpoint of submarine cable route are used as input data for KD-DNN to improve its prediction accuracy. The feasibility of transferring the engineering knowledge for similar scenarios is demonstrated.
Funding Information
  • National Key Research and Development Program of China (2022YFB2903300)
  • National Natural Science Foundation of China (62371064)
  • Basic and Frontier Exploration Project Independently Deployed by the Institute of Acoustics, Chinese Academy of Sciences (JCQY202412)

This publication has 9 references indexed in Scilit: