Presentations

Abstract

IAV has developed a comprehensive Model Predictive Turbine Control (MPTC) framework that optimizes both individual wind turbine and wind farm operations under structural, operational, and economic constraints. At turbine level, control is formulated as a constrained optimal control problem solved in a receding horizon framework using a dynamic turbine model, onboard state estimation, and short-term inflow preview information. By explicitly considering actuator limits, safety constraints, and competing objectives such as power tracking and load mitigation, MPTC enables preventive control actions that improve load management while maintaining high operational efficiency. A key performance driver of predictive control is the quality of inflow preview information, for example obtained from lidar-based measurements. The benefits of MPTC increase with rotor size, structural flexibility, and turbulence intensity, where anticipatory actuation provides significant potential for fatigue load reduction. The control formulation offers physically interpretable tuning parameters, enabling systematic adaptation to different turbine platforms and operational strategies. The MPTC approach has been validated in collaboration with a Chinese wind turbine OEM using a staged verification process. Extensive Software-in-the-Loop (SiL) and Hardware-in-the-Loop (HiL) testing covered normal operation as well as safety-critical scenarios. The tested control system demonstrated reliable execution of turbine start-up and shut-down procedures and correct handling of grid-related safety functions such as low-voltage ride-through and grid frequency adjustment. Following laboratory validation, MPTC was deployed in a field test on a 6 MW-class wind turbine in Inner Mongolia. The turbine accumulated more than 1,100 hours of fault-free operation under realistic environmental and grid conditions. Field measurements confirmed stable closed-loop behavior and validated the load reduction potential predicted during SiL studies, demonstrating robustness with respect to model uncertainties, measurement noise, and inflow variability. Building on the operational flexibility enabled at the turbine level, a coordinated wind farm control concept is proposed. This involves a decentralized coordination framework based on model-free reinforcement learning to adapt turbine-specific MPTC weighting parameters, actively influencing power and load distribution across the farm. The learning agent uses observations derived from onboard estimators, including incremental fatigue damage metrics, actuator usage indicators, and representative inflow characteristics. A tailored reward function balances competing objectives such as energy production, component lifetime consumption, and operational wear. The combined architecture of constrained model predictive control at the turbine level and learning-based coordination at the farm level enables scalable, adaptive, and economically optimized wind farm operation. By supporting long-term operational optimization, the proposed approach aims to reduce total cost of ownership and increase the lifetime profitability of wind assets, representing a significant shift from reactive turbine control to predictive, model-based operation.   Keywords: LiDARs, single WTG optimization, optimal use of forecasts in applications, real-world case studies.