Adaptive Immune System reinforcement Learning-Based algorithm for real-time Cascading Failures prevention

Artificial intelligent algorithms have found a wide-range of applications in power systems, especially in solving long-existing problems immune to non-intelligent algorithms. Cascading Failures (CF), one of such problems, require load shedding as a current industrial solution. Load shedding results in losses to all power system stakeholders. This work proposes the use of an Artificial Immune System (AIS) algorithm to intelligently adjust the power output of the generators in the power system relative to one another in real time to prevent CF. AIS gives the artificial intelligent algorithm reinforcement learning capability by enabling it to pick the appropriate combination(s) for a particular system state; hence, the algorithm is called Immune System Reinforcement Learning-Based (ISRL-Based) algorithm. The algorithm was trained offline using both static and dynamic power equations and the effectiveness of both approaches was evaluated through statistical deviation. Analyses showed that using dynamic equations resulted in a more accurate solution than the static equations. CF was dynamically simulated on the IEEE 118-Bus system after an N-2 contingency, the results obtained agrees with the results from the analysis of the 2003 Northeast USA CF event. The effectiveness of the algorithm and online training were also experimentally validated after an N-1-1 contingency in a nine-bus system.