Posters

Abstract

In recent years, the industry has recognized the need to transition from a turbine-centric control approach to plant-level collective control. While it is important to ensure turbines perform according to their manufactured power curve, maximizing the production of individual turbines does not optimize wind plant production. Collective control is needed to achieve a paradigm shift in annual energy production (AEP). Shortcomings in plant operation due to turbine-centric control can reduce plant power output anywhere between 5 and 20% and globally cost the industry approximately $16B in annual revenue. Combating these shortcomings through strategies such as wake steering has historically been the predominant method for wake mitigation and employing collective control. However, WindESCo has shown that methods such as wake steering only addresses a single inefficiency in current plant operations. Other control strategies that leverage neighboring turbine information (e.g., collective yaw, predictive yaw, etc.) are required for a comprehensive solution to collective control and to achieve a significant return on investment. WindESCo recently commissioned its plant-level collective control system WindESCo Swarm at the Milford wind plant. Comprising 165 wind turbines and +300 MW of capacity, this commissioning is the largest known implementation of commercial wake steering and collective control technology to date. Although research has previously highlighted the AEP upside of collective control and wake steering, large-scale field testing of collective control has yet to demonstrate how attainable modeled gains are. This study will review the results of Phase I of the Swarm implementation at Milford that commenced in early 2022, including several of the the challenges encountered and innovations required to effectively implement collective control in the field. Historically, a static lookup table has been used to define the optimal control decisions for collective control. However, AEP gains realized through this approach have commonly fallen short of expectation. As part of Swarm, model-in-the-loop control as opposed to a static lookup table approach is used to ensure optimized control decisions across the range of wind, atmospheric, and operational turbine conditions observed onsite. However, incorporating a model in the control loop raises a range of other questions including, for example, how to effectively close the loop on the ambient wind conditions (i.e., wind speed, direction, turbulence), the relationship between yaw error and power decay (i.e., identifying the correct cosine exponent), etc. How closing the loop using various consensus-based strategies impacts the ability of the model to resolve wake and power dynamics will be demonstrated. Finally, production gains from Phase I of the Milford deployment including production improvements of 3.4% for a downstream waked turbine will be reviewed.