Ziqi Zheng , USA

In this paper, two types of algorithms are proposed to improve the power grid distribution in which one uses static methods (Dijkstra’s, Ford-Fulkerson) to consider the capacity/loss from plants and transmission lines, and another uses probability/reinforcement learning based methods (Markov Decision Processes, and Q-Learning) to take into consideration the uncertainties, such as fuel shortages and wind variability for the goal of optimizing its energy flow. We took the dataset for Cuba’s power plants as a case study to test the effectiveness of these algorithms for power distribution. Our results show a major potential improvement of 22-68% in energy generation from using these two kinds of algorithms (static/reinforcement learning) compared to the current country’s available operating power.

Energy Optimization, Graph Theory, Reinforcement Learning, Markov Decision Processes, Ford-Fulkerson, Dijkstra's, Q-Learning