Game Theoretic Wind Farm Control Based on Level-k Cognitive Modeling.

This paper presents a control strategy applicable to systems operating in dynamically coupled environments for which only low-fidelity models describing the environmental dynamics and system rewards are available. The control problem is posed as a two player game wherein the plant and the environment are considered to be agents that seek to maximize their individual reward. Under this paradigm, the control decisions of the plant are then identified using a level-k cognition modeling approach, where historical sensor data is leveraged to augment predictions of the reward maximizing control input. We demonstrate the proposed framework via simulation of a multi-turbine wind farm wherein the objective is to maximize energy production when only low-fidelity models of the wake effect induced by the interactions between upstream turbines and the wind environment are accessible.