Data-Driven LIDAR Feedforward Predictive Wind Turbine Control

Review of Data-Driven Models in Wind Energy: Demonstration of Blade Twist Optimization Based on Aerodynamic Loads

“…Through examination of the literature on data-driven models in wind turbine applications, several points are made evident. First, the various purposes of data-driven modeling efforts are generally focused in a handful of narrow topics: reduction of computational expense [33,35,52,92,94], improvements in accuracy or other performance over existing methods [34,[50][51][52][92][93][94], and leveraging the unique aspects of data driven models in novel approaches for existing problems that cannot be solved or are difficult to solve with extant methods, such as fault detection [32,92,94].…”

“…While a variety of performance monitoring applications are shown [31,[49][50][51][52][53][54][55][56][57], the authors highlight a comparative lack of controls-and optimization-oriented efforts compared to counterparts in condition monitoring applications like fault detection. Efforts such as [52][53][54][55][56] show promise in allowing predictive, systems-identifying data-driven models to augment performance of turbines through optimizing performance parameters, improving data efficiency to enable real-time applications, and reducing computational expense, yet few sources in the literature focused on tackling this in comparison to the now more well-tread areas of fault detection.…”

“…Performance monitoring is another ongoing avenue of research for data-driven modeling in the wind turbine space [31,[49][50][51][52][53][54][55][56][57][58][59][60][61][62][63]. For the purposes of this paper, performance modelling is classified here as any task where data-driven models are utilized to observe, predict, control, and optimize the performance of individual turbines, or turbine farms, referencing tasks as such as in [31].…”

“…A more modern effort for the active control of wind turbine parameters, as compared to [49] as aforementioned, is discussed in a paper by R. Dinkla et al This effort uses a model predictive control method, specifically, subspace predictive control, to contribute to a closed loop controller for the blade pitch actuation. The goal of this effort is to improve performance of the blade pitch controller by reducing the total amount of pitch actuation, by incorporating detected future wind disturbances instead of traditional methods, with the added benefit of improving data efficiency of the controller through less requisite data retention [52].…”

“…Similar controls-oriented efforts can be seen in work by Yang Liu et al [53], whom also designed a data-driven pitch controller to reduce computational expense of the system model, as well as in work by Jingjie Xie et al [54], who use a NN-based adaptive controls algorithm to manage pitch and torque control. The efforts mentioned in [52][53][54] take advantage of the reduction in computational expense that data-driven methods provide, as does the work by M. Nouri Manzar and A. Khaki-Sedigh [55]. Their effort, too, involves the use of a data-driven model to enable active control of pitch and torque control schemes, through reduction of computational expense for the controller to operate in real-time.…”