Statistics of Reinforcement learning versus model predictive control: a comparison on a power system problem