Leak Localization Using Graph Based Reinforcement Leaning for Subterranean Electrical Cables

It is challenging to create an RL-based model for subterranean electrical cables, reinforcement learning, and sequential localization of multiple underground transmission leaks is a new area of research that hasn't gotten much attention. The fact that different-sized breaches can occur in the same network and that they don't occur frequently would be one of the key causes. It is critical to identify the locations of these leaks and find an optimal graph according to the magnitude of the leaking. Using accuracy as the performance metric for these models could be misleading because it does not consider the quantity of false positives and as well as false negatives hence reinforcement learning which leans by trial and error would be the recommended solution. We have created a reinforcement learning, double deep Q learning with encoder decoder, algorithm to address this issue. The dataset is a part of an unrestricted piping system. This data has been modified and a volume feature that tracks the amount of wasted dielectric fluid has been introduced.