Intelligent decision-making agent with self-attention mechanism for electric substation grounding grids life cycle management

The electric substation grounding grid faces deterioration due to electrochemical reactions. The welding defects, acidic ambience, and inhomogeneity of conductor materials induce accelerated corrosion, and in severe cases, fractures may occur after a certain period of service time. However, periodic maintenance according to specifications cannot detect failure promptly, significantly increasing the risk to workers during electric substation operations. This study develops an intelligent life cycle maintenance framework through deep reinforcement learning and digital twins. Using practical inspection data and experimental results, reliable digital twins are established as the training environment for agents to determine the maintenance scheme. The intelligent agent dynamically optimises inspection intervals using historical resistance data and recognises corrosion trends, allowing for the replacement of steel sheets before failure happens. Monte Carlo simulations demonstrate that the agent can reduce personnel risk exposure by 25.7% while simultaneously saving 28.1% cost compared with the routine maintenance policy.