Electrical power grids are an essential, yet extremely complicated piece of the infrastructure of city life. Various scholars used different reinforcement learning algorithm to train an agent that would make efficient decision. Yet, such agent had limitation as it took more than 4 7 years to train to have better performance compared to baseline model. For this we explored different methods, such as inference and behavior cloning, to shorten the training period. We were not able to shorten the training period with those methods. Therefore, we discussed about how we would use Model-Agnostic Meta-Learning (MAML) algorithm to shorten the training period.