Estimation of maximum disturbing load in distribution grids using multi-agent learning

Interaction among disturbing loads in the same distribution system could cause critical harmonic levels. In this paper we propose a multi-agent methodology to analyze that interaction, identifying the maximum allowable disturbing load in every node of a distribution system. Nodes are considered as agents while states, actions, and profits are defined using harmonic distortion indexes. A Q-learning algorithm is implemented to optimize load connection strategies for every agent and avoid critical scenarios. Finally, several load scenarios are simulated and their impact is assessed in terms of TDD and THDv harmonic distortion indexes.