Computing Day-Ahead Dispatch Plans for Active Distribution Grids Using a Reinforcement Learning Based Algorithm

The worldwide aspiration for a sustainable energy future has led to an increasing deployment of variable and intermittent renewable sources (RESs). As result, predicting planning the operation power grids become more complex. Batteries can play critical role this problem as they absorb uncertainties introduced by RESs. In paper, we solve computing dispatch plan distribution grid with RESs batteries novel approach based on Reinforcement Learning (RL). Although RL is not inherently suited problems that require open loop policies, have developed iterative algorithm calls trained agent at each iteration compute plan. Since feedback given cannot be directly observed because computed ahead operation, it estimated. Compared conventional scenario-based optimization, our RL-based exploit significantly prior information uncertainty computes plans faster. Our evaluation comparative results demonstrate accuracy well adaptability input data diverge from training data.