Comparison of Data-Driven Modeling Approaches for Control Optimization of Floating Offshore Wind Turbines

Athul Krishna Sundarrajan, Daniel R. Herber · 2025

Read Comparison of Data-Driven Modeling Approaches for Control Optimization of Floating Offshore Wind Turbines by Athul Krishna Sundarrajan, Daniel R. Herber (2025) 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-2505-14515.txt.

Introduction

Athul Sundarrajan and Daniel Herber race rival shortcut models for floating wind turbines in a 2025 paper. A turbine on a bobbing platform is a control designer's puzzle, and optimizers need hundreds of test runs. Full simulations are too slow, so faithful stand-ins decide the search.

The authors build stand-ins three ways: classic system identification, deep learning, and derivative surrogates that mimic how the state changes rather than the state itself. Their new variant shapes the surrogate with a parameter-varying linear scheme. It struck the best balance of speed and accuracy, running about fifty times faster than the full model. The comparison guides engineers choosing design tools.

Catalog record

CREATORAthul Krishna Sundarrajan, Daniel R. Herber
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-14515

Browse the full library · About the Tower of Babel

Full text

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

Comparison of Data-Driven Modeling Approaches for Control Optimization of Floating Offshore Wind Turbines Athul Krishna Sundarrajan, Daniel R. Herber Abstract Models that balance accuracy against computational costs are advantageous when designing wind turbines with optimization studies, as several hundred predictive function evaluations might be necessary to identify the optimal solution. We explore different approaches to construct low-fidelity models that can be used to approximate dynamic quantities and be used as surrogates for design optimization studies and other use cases. In particular, low-fidelity modeling approaches using classical systems identification and deep learning approaches are considered against derivative function surrogate models (DFSMs), or approximate models of the state derivative function. This work proposes a novel method that utilizes a linear parameter varying (LPV) modeling scheme to construct the DFSM. We compare the trade-offs between these different models and explore the efficacy of the proposed DFSM approach in approximating wind turbine performance and design optimization studies for controllers. Results show that the proposed DFSM approach balances computational time and model accuracy better than the system identification and deep learning based models. Additionally, the DFSM provides nearly a fifty times speed-up compared to the high-fidelity model, while balancing accuracy. \confshortname IDETC/CIE2025 \conffullname Proceedings of the ASME 2025 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference \confdate 17-20 \confmonth August \confyear 2025 \confcity Anaheim \confcountry CA \papernum DETC2025-167963 Keywords: surrogate models; dynamic systems; system identification; optimization; floating offshore wind turbines 1 INTRODUCTION Models that accurately capture the dynamics are needed to identify and understand system-level optimal designs. As a system’s complexity and the fidelity of the underlying analyses increase, such models become more computationally expensive to evaluate [ 1 , 2 ] . For example, in highly-coupled multidisciplinary systems like floating offshore wind turbines (FOWTs) and marine turbines, the effect of the different subsystems (like the rotor, support structure (tower), floating platform, etc.) on each other must be fully considered to get an accurate response of the system [ 3 ] . Tools like OpenFAST [ 4 ] and WEC-Sim [ 5 ] can accurately capture the dynamics of renewable energy systems like wind turbines, marine turbines, and wave energy converters. A consequence of capturing the detailed dynamic response of these systems is the high computational cost associated with simulating them. This computational expense, consequently, can make the direct use of these high-fidelity modeling tools in the following applications inefficient. Estimating system performance: To evaluate the relevant performance metrics for systems like wind and marine turbines, the dynamic model must be simulated for several hundred design load cases (DLCs) corresponding to different environmental conditions and operating scenarios as specified by the IEC design standards [ 6 ] . Depending on the system, the simulation time for a single DLC can be several minutes to hours. Therefore, a significant computational cost in terms of CPU hours is incurred to evaluate the system’s performance by simulating the system model over all relevant DLCs. Use in design optimization studies: To identify the optimal physical design and/or control law, evaluating the dynamic system several hundred or more times is necessary. Additionally, the software architecture of these system models might be such that it is impossible to link all the necessary variables of interest directly to an optimizer. State variables are an example, as many simulation tools consider them internal, input-dependent quantities. Engineers are also interested in understanding the impact of changing the system’s physical parameters on its dynamics. To solve these problems efficiently, engineers often use gradient-based solvers. Many simulation tools do not provide the accurate gradient information that these gradient-based solvers require. One recourse is to use finite-difference approaches to estimate the gradients, but these gradient estimates for simulations can be noisy or insensitive to small perturbations in the design variables and can be expensive to evaluate. For these reasons, engineers are limited when setting up design optimization studies and resort to using gradient-free solvers like genetic algorithms (GA), which require several hundred function evaluations, or methods like COBYLA, which do not scale well with the number of design variables [ 7 ] . Therefore, computationally inexpensive models that can capture some key dynamic quantities like the system states and metrics with sufficient accuracy could be helpful [ 8 , 1 , 9 ] . These approximations are generally referred to in the literature as low-fidelity models because of their limited scope in capturing the system response. The necessity of low-fidelity models for many applications with computationally expensive high-fidelity models has been outlined in detail in design optimization literature. The scope of this article is limited to developing low-fidelity models for the open-source wind and marine turbine simulation tool OpenFAST. Broadly, there are two different ways to construct low-fidelity models for wind turbines, namely physics-based and data-driven. The focus of this article will be on constructing data-driven low-fidelity models of wind turbines that can address the two drawbacks of using high-fidelity models outlined above. Section 2 outlines the problem statement and the requirements of the low-fidelity model that is developed to address the drawbacks outlined above for OpenFAST. A more detailed outline for the rest of the article is presented at the end of Sec. 2 . 2 BACKGROUND Figure 1 : A high-level workflow diagram illustrating the different components associated with simulating and/or performing design optimization studies for a wind turbine system. A high-level workflow diagram illustrating the different steps and software components associated with simulating a wind or marine turbine model using OpenFAST is shown in Fig. 1 . At the heart of this process are OpenFAST [ 10 ] and the open-source controller ROSCO introduced in Ref. [ 11 ] . To understand Fig. 1 and the motivation behind this article, it is necessary to understand how these core components are set up and operate. 2.1 OpenFAST OpenFAST is an open-source modeling tool that is extensively used to simulate the dynamic response of fixed or floating wind turbine systems to various wind and wave excitations. Although primarily used for wind turbines, OpenFAST has been recently extended to model the response of marine turbine systems as well [ 12 , 13 ] . Given a description of the different components, such as the rotor, generator, tower, platform mooring lines, etc., and their properties, OpenFAST can predict the aerodynamic and hydrodynamic loads, generator performance, and the displacement, velocities, and reaction loads of the turbine components, for the exciting wind and wave elevation signals. OpenFAST can be broken down into a set of interconnected modules that contain the relevant physics needed to model the different aerodynamic, hydrodynamic, generator, and structural reactions [ 10 ] . OpenFAST formulates the dynamics of the turbine as a set of differential-algebraic equations (DAEs) and solves these DAEs to predict the dynamics. Because OpenFAST needs to resolve the underlying physics of the different modules individually and account for the coupling between modules, it can be computationally expensive. Figure 2 : Generator torque and blade pitch vs. wind speed. 2.2 Control of Wind Turbines Wind turbines must operate in stochastic offshore environments where they can be subject to extreme forces. A controller is essential to mitigate the loads the turbines are exposed to and ensure power production [ 14 ,

Read the complete text