Short-term wind forecasting via surface pressure measurements: stochastic modeling and sensor placement

Seyedalireza Abootorabi, Stefano Leonardi, Mario Rotea, Armin Zare · 2024

Read Short-term wind forecasting via surface pressure measurements: stochastic modeling and sensor placement by Seyedalireza Abootorabi, Stefano Leonardi, Mario Rotea, Armin Zare (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-14403.txt.

Introduction

Seyedalireza Abootorabi, Stefano Leonardi, Mario Rotea, and Armin Zare work on very short-term wind forecasts for turbines. Their 2024 paper aims to track fast changes in rough air flow. It draws on wind gauges mounted on turbines and on air pressure sensors at ground level.

Their method blends a simple stochastic model with Kalman filters that update as new data arrive. A projection step links ground pressure to conditions at hub height. The authors show how filter choice trades speed against accuracy near the lead turbines. They also give a method for picking a small set of pressure sensors that still meets a target level of accuracy.

Catalog record

CREATORSeyedalireza Abootorabi, Stefano Leonardi, Mario Rotea, Armin Zare
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-14403

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Short-term wind forecasting via surface pressure measurements: stochastic modeling and sensor placement Seyedalireza Abootorabi, Stefano Leonardi, Mario Rotea, Armin Zare Abstract We propose a short-term wind forecasting framework for predicting real-time variations in atmospheric turbulence based on nacelle-mounted anemometer and ground-level air-pressure measurements. Our approach combines linear stochastic estimation and Kalman filtering algorithms to assimilate and process real-time field measurements with the predictions of a stochastic reduced-order model that is confined to a two-dimensional plane at the hub height of turbines. We bridge the vertical gap between the computational plane of the model at hub height and the measurement plane on the ground using a projection technique that allows us to infer the pressure in one plane from the other. Depending on the quality of this inference, we show that customized variants of the extended and ensemble Kalman filters can be tuned to balance estimation quality and computational speed 1–1.5 diameters ahead and behind leading turbines. In particular, we show how synchronizing the sign of estimates with that of velocity fluctuations recorded at the nacelle can significantly improve the ability to follow temporal variations upwind of the leading turbine. We also propose a convex optimization-based framework for selecting a subset of pressure sensors that achieve a desired level of accuracy relative to the optimal Kalman filter that uses all sensing capabilities. Preprint: AIP/123-QED I Introduction Adjustments to turbine blade pitch, generator toque, and nacelle direction (yaw) have shown promise in increasing energy production, lowering operational costs, and stabilizing the power distribution grid [ 1 , 2 , 3 , 4 , 5 ] . However, in practice, the effectiveness of feedback control strategies can be challenged by atmospheric variability ranging from meandering wind patterns to gusts and squalls resulting in achieved gains that fall short of expectations set by ideal scenarios that entail steady winds. In fact, in the absence of effective forecasting tools, almost all modern-day plants rely on data collected at or just behind the wind turbine to adjust their settings, and can consequently lag optimal operation conditions. This motivates the development of short-term forecasting tools for estimating changes in the speed of wind seconds to minutes ahead of impact, thereby enabling model-based control systems to preemptively adapt to improve turbine efficiency and reduce structural loads and failures [ 6 , 7 , 8 ] . In order to estimate changes in the incoming wind velocity, forecasting tools rely on the ability of models in capturing the evolution of dominant coherent motions that affect the performance and structural durability of wind turbines. While high-fidelity models, such as those that are based on large-eddy simulations (LES), have been shown to successfully capture the vast range of spatio-temporal scales in wind farm turbulence, their computational cost precludes their utility in developing forecasting tools that can inform real-time turbine control. To address this challenge, various studies have sought lower-fidelity alternatives by constraining the computational domain of LES [ 9 , 10 , 11 ] , employing the Reynolds-averaged Navier–Stokes (RANS) in conjunction with simple turbulence models [ 12 , 13 , 14 ] , developing reduced-order models based on data-driven methods such as proper orthogonal decomposition [ 15 , 16 ] and dynamic mode decomposition [ 17 ] , and more recently, capturing the dynamics of velocity fluctuations around the time-averaged flow using the stochastically forced linearized Navier-Stokes (NS) equations [ 18 , 19 ] . Together with data-assimilation strategies such as Kalman filtering reduced-order models can adapt to atmospheric variations that affect the velocity field impinging a wind farm. Kalman filtering algorithms were initially used to improve meteorological models that provide long-term forecasts, i.e., predictions of daily-averaged wind speeds over wind plants and the resulting energy production [ 20 , 21 ] . However, the focus of more recent studies has shifted toward short-term forecasting of the direction and speed of wind to enable real-time turbine control (see, e.g., Ref. [ 22 ]). I.1 Real-time flow estimation for wind farms Wind velocity measurements from light detection and ranging (LiDAR) scanners, nacelle-mounted anemometers, and meteorological towers have provided invaluable insights into the dynamics of wind turbine wakes (see, e.g., Refs. [ 23 , 24 ]) in addition to opportunities for estimation and control. The early work of Schlipf et al. demonstrated the utility of LiDAR-based wind measurements in the context of preview control [ 25 ] and nonlinear model predictive control [ 26 ] . Iungo et al. [ 17 ] proposed Kalman filtering as a means to dynamically update predictions of wind turbine wakes obtained by a DMD-based reduced-order model that is trained using LES data. In this work, the authors used LES data throughout the entire 3D domain to correct model-based predictions at each time step. Sinner et al. [ 27 ] also adopted Kalman filtering to capture velocity variations at turbines’ hub height based on noisy anemometer readings. In contrast to prior efforts, they proposed the use of a polynomial-based class of data-driven prior models that are trained to match the direction and magnitude of velocity at the hub-height of turbines across a wind farm. Doekemeijer et al. [ 28 ] used ensemble and approximate Kalman filters to assimilate longitudinal and lateral wind speed measurements collected from a coarse-grained mesh with the predictions of a 2D LES-based model to estimate the mean centerline velocity impacting wind turbines. Similar Kalman filters were later shown to estimate longitudinal and lateral velocity fields using only power production measurements [ 29 ] . This estimation framework was further extended to incorporate both turbine power and velocity measurements in the context of extended, ensemble, and unscented Kalman filters that predict the velocity field at hub height [ 22 ] . The authors demonstrated improved estimation accuracy for scenarios in which a row of measurement sensors (which could be potentially provided by a LiDAR scanner) are placed downstream of wind turbines. The dynamic wake meandering model has also been used as a prior for Kalman filtering. In particular, Kalman filtering algorithms were used to assimilate power production [ 30 ] and LiDAR [ 31 ] measurements with the predictions of this model. While most studies assume precalculated power coefficients, Lio et al. [ 32 ] estimated this parameter by incorporating measurements of the rotational speed of turbine rotors into their Kalman filtering algorithm. I.2 Motivation, contribution, and preview of results In this paper, we propose an estimation framework that utilizes an alternative, significantly cheaper, sensing technology relative to Doppler LiDAR. Our framework relies on the sequential self-correcting property of the Kalman filter in assimilating measurements from ground-level air-pressure sensors and nacelle-mounted anemometers with model-based predictions of hub-height velocity fluctuations. To facilitate real-time estimation, we limit the dimensional complexity of our models by confining their computational domains to the hub height of wind turbines. We bridge the vertical gap between the domain of our 2D models of hub-height velocity and the measurement plane on the ground by leveraging a data-driven projection scheme for inferring ground-level pressure from hub-height pressure. Our contributions can be listed as follows: • Stochastic dynamical modeling of the fluctuating velocity field at the hub height of a wind farm; • Devising customized linear and nonlinear Kalman filtering algorithms for assimilating ground-level

Read the complete text