A Framework For Critical Infrastructure Monitoring Based On Deep Reinforcement Learning Approach

Abstract

Critical Infrastructure (CI) are nowadays linked with IOT devices that communicate data through networks to achieve significant collaboration. With the progress in internet connectivity, IOT has disrupt numerous aspects of CI comprising communication systems, power plants, power grid, gas pipeline, and transportation systems. As a disruptive paradigm, the IOT and Cloud computing utilizing Smart IOT devices equipped with numerous sensors and actuating capabilities play significant roles when deployed in CI surroundings with the aim of monitoring vital observable figures consisting of flow rate, temperature, pressure, and lighting situations. Over the years, oil pipeline infrastructure have been the main economic means for conveying refined oil to assembly and distribution outlets. Though damages to the pipelines in this area by exclusion have influence the normal transport of refined oil to the outlets across the country like Nigeria which has influence the stream of income and damages to the environment. Reinforcement Learning (RL) approach for infrastructure reliability monitoring have receive numerous consideration by researchers denoting that RL centered policy reveals superior operation than regular traditional control systems strategies. Many of the studies utilised mainly algorithms for environment with discrete action and observation spaces unlike others with infinite state space. This study proposed a framework for critical infrastructure monitoring based on Deep Reinforcement Learning (DRL) for oil pipeline network and also developed a pipeline network monitoring (PNM) architecture with expression of the environment dynamics as Markov Decision Process. The sample observation space data and strategy for evaluation of the framework was also presented.