April, Tuesday 10th
14:30 (room 2014, 'Digiteo Shannon' 660 building) (see location
Berna Bakir Batu
Title: A Reinforcement Learning Approach for Simulating Cascading Failures in Power Grids
Understanding and modelling cascading effects of line overloads on the transmission lines that resulting in blackouts in a power grid are challenging due to unavailability of sufficient historical data and limitations of simulations as a result of a combinatorial increase of possible cascading failure scenarios as the number of failures increases. Although power grids are usually designed to meet N-1 criteria and large blackouts occur rarely, they are still more likely than expected and have high economic and social impacts when they occurred. Whether small or large cascades, revealing such triggering patterns can be used for, for example, analyzing the vulnerability of power grids and for improving their robustness. In this study, potentials of reinforcement learning approach for developing a simulation model for cascading failures in power grids were examined. As failure type, line overloads and their propagations through the network were considered. The state features were defined based on line capacity usages and value function was fitted based on randomly simulated states up to N-5 contingency. Finally, cascades were simulated using approximated value function and results were discussed.
Contact: guillaume.charpiat at inria.fr
All TAU seminars: here