Optimising Resource Allocation for Increased Network Reliability using Reinforcement Learning

Optimising Resource Allocation for Increased Network Reliability using Reinforcement Learning

Anup Kumar Yadav1, Kalpna Sachan2, Harimohan Soni3

1Mechanical Engineering Department, Bansal Institute of Science and Technology, Bhopal (M.P.), India.

2Mechanical Engineering Department, Bansal Institute of Science and Technology, Bhopal (M.P.), India.

3Mechanical Engineering Department, Bansal Institute of Science and Technology, Bhopal (M.P.), India.

---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract - Networks deliver essential services to society (e.g., electricity grid, telecommunications, water supply, transportation) yet are susceptible to interruption. Motivated by this, we examine a sequential choice problem whereby an initial network is enhanced over time (e.g., through the addition or augmentation of edge dependability), and incentives are accrued over time based on the network's all-terminal reliability. Actions within each time frame are constrained by the availability of resources, including time, financial capital, and labour. We employed a Deep Reinforcement Learning (DRL) methodology executed in OpenAI-Gym utilising Stable Baselines to address this issue. A Proximal Policy Optimisation (PPO) algorithm was employed to determine the edge requiring enhancement or a new edge to be included, contingent upon the network's present condition and the allocated budget. A reliability polynomial was utilised to compute the all-terminal dependability of the network. To comprehend the model's behaviour throughout diverse settings, we examined many network configurations with varying beginning link reliability, additional link reliability, node quantities, and budget frameworks. We finish with a review of the insights acquired from our series of constructed experiments.

Key Words:  Deep reinforcement learning, proximal policy optimisation.