A FWO doctoral fellowship has recently been awarded to Lorenzo Schena

Lorenzo Schena in front of the Vki Low Speed Wind tunnel

We are pleased to announce that a Doctoral (PhD) Grant Strategic Basic Research has been awarded for 2 years by the FWO to Lorenzo Schena. Lorenzo is investigating how to blend machine learning technique and robust classical control approaches, to retain the best of the two worlds: the adaptive capabilities of the former and physical roots of the latter. He works in collaboration with Prof. Miguel Mendez from VKI, Dr. Wim Munters from VKI and Prof. Jan Helsen from VUB.

The Doctoral (PhD) Grant Strategic Basic Research awarded by FWO (Fonds Wetenschappelijk Onderzoek - Vlaanderen) allows young researchers to develop into strategically thinking and innovation-oriented scientists. Strategic basic research in the context of a PhD grant stands for challenging and innovative research (at PhD level), which, if successful, may in the longer term lead to innovative applications with economic added value (for specific companies, for a collective of companies, or a sector, or in line with the VRWI transition areas.

Abstract: Wind energy is an environmentally friendly and cost-effective solution to the ever-increasing energy demand. As modern wind turbines continue to increase in size, operational and maintenance costs, the closed-loop control of their performances becomes crucial to their economic viability. Traditional control systems acting on blade pitch and yaw angles or on the generator torque are based on Proportional Integral Derivative (PID) controllers. These are calibrated around different design conditions using linearized models of the wind turbine dynamics. The linear framework simplifies the control design and is robust close to design conditions. However, it provides sub-optimal performances in transient conditions and blade overloading when approaching the power-limited regime.

This work proposes a novel control technique: Reinforced model predictive control (RPC), a unique blending of model-free and model-based machine learning techniques. The technique will be applied in the wind-turbine control framework at three different scales: the single wind-turbine, two-turbines system, and, finally, the wind-farm. In all these test cases, the performances of the proposed controller are going to be benchmarked against state-of-the-art solutions, both numerically and experimentally. Emphasis is placed on the validation of the new tool proposed, being the experimental tests of machine learning-based methods, or more generally of new proposed control strategies for wind turbines, lacking in the literature.