Reinforcement learning resource allocation optimisation for network reliability

The power grid, telecommunications, water supply, and transportation are all examples of networks that provide vital societal functions but are vulnerable to disruption. Based on this, we take a look at a sequential choice problem where an initial network is improved over time (for example, by adding or enhancing edge dependability) and incentives are added depending on the network’s all-terminal reliability. Time, money, and labour are all limited resources that determine what can be done within each time frame. To fix this, we used a Deep Reinforcement Learning (DRL) approach in OpenAI-Gym with Stable Baselines. Based on the current state of the network and the budget, a Proximal Policy Optimisation (PPO) method was used to identify the edge that needed improvement or to incorporate a new edge. The network’s all-terminal dependability was calculated using a reliability polynomial. We evaluated numerous network configurations with different starting link reliability, additional link reliability, node amounts, and budget frameworks to understand the model’s behaviour in different conditions. Last but not least, we go over the key points from our set of controlled tests.