Wind-Farm Power Tracking Via Preview-Based Robust Reinforcement Learning

This article aims to address the wind-farm power tracking problem, which requires farm's total generation track time-varying references and, therefore, allows wind farm participate in ancillary services such as frequency regulation. A novel preview-based robust deep reinforcement learning (PR-DRL) method is proposed handle tasks are subject uncertain environmental conditions and strong aerodynamic interactions among turbines. To our knowledge, this for first time that a data-driven model-free solution developed tracking. Particularly, reference signals treated preview information embedded system specially designed augmented states. The control problem then transformed into zero-sum game quantify influence of unknown future signals. Built upon $H_\infty$ theory, PR-DRL can successfully approximate resulting game's achieve Time-series measurements long short-term memory networks employed DRL structure non-Markovian property induced by time-delayed feature interactions. Tests based on dynamic simulator demonstrate effectiveness strategy.