Project Abstract

A microgrid is a small power system. It has an interconnected cyber layer mainly dealing with data transmission and decision-making based on data gathered by advanced metering infrastructure (AMI), including smart meters, sensors, or communication channels. Further, the microgrid is a complex network, and any failure or anomaly in a part of the system may lead to huge damage to the overall system in a short period. Hence, early detection of cyber-attacks is critical for a timely and effective response. In this context, the framework of the quickest change detection algorithm is required to protect the microgrid.

In this research work, a real-time cyber-attack detection algorithm is proposed for microgrids using the framework of model-free Reinforcement learning Algorithm (RLA) for partially observable Markov decision process (POMDP). For the problem of online attack, in the pre-change state, the system will be operated under normal conditions, and using the system model, the pre-change measurement probability density function (pdf) can be specified highly accurately. On the other hand, the post-change measurement PDF can take different unknown forms depending on the attacker’s strategy. RLA is known to be effective in controlling uncertain environments. Hence, the POMDP problem can be effectively solved using RL.

In the proposed work, mathematical modelling with state variables and an RL algorithm will be developed to detect the wireless sensor-based microgrid systems’ cyber-attacks.