A Reinforcement Learning Framework for Intelligent Detection of Bad Data in Power System State Estimation

Résumé fourni par la source

The efficient and safe running of electrical grids depends on state estimation in power systems. Discovering bad data is key to keeping state estimates exact. Traditional approaches such as WLS and ANNs are not flexible, efficient and do not perform well on new data examples. As a result, it makes sense to explore different approaches. This paper offers a reinforcement learning (RL) framework for finding bad data in power systems. This research shows that RL is a powerful method for finding and separating corrupt measurements almost instantly. This research uses both Q-Learning and Deep Q-Networks to teach the agent as it interacts with the power system environment, gaining rewards every time it classifies things correctly. The IEEE 14-bus test system is applied to test the approach and bad data is added to emulate potential sensor failures and cyber issues. The results prove 100% accuracy, since no incorrect positive or negative outcomes are recorded. Despite the dynamic change in conditions, the RL agent is able to tell the difference between regular and manipulated measurements. Testing the model with different workloads and system setups proves that it works efficiently and is scalable. This research points out that RL is a capable and adaptive strategy for spotting bad data in power systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.