As nations broaden their renewable energy portfolios, wind power is playing an increasingly important role in the energy landscape. Although onshore wind continues to dominate the wind energy mix in the EU, rising energy demand and ambitious climate targets have driven the development of offshore wind farms and spurred research into floating offshore wind turbines.

This thesis began with the goal of developing a data-driven modeling approach that maps site conditions to the load statistics on floating offshore wind turbines. The motivation for this research emerged from a need to accelerate the site-selection process, include more site variables, and achieve accurate load estimates while keeping the computational expense low.

Before installation, all modern wind turbines must be carefully assessed for structural loads to ensure safety and performance throughout their lifetimes. Floating turbines, in particular, introduce more complexity—more environmental variables to consider, and higher uncertainty in the dynamic response. This added complexity makes analyzing structural loads expensive and time-consuming, often more so than for fixed-bottom counterparts, and it makes understanding the behavior of floating wind turbines even more important. The process of fatigue damage calculation, in particular, is computationally intensive, often requiring thousands of costly simulations. Because it is not feasible to run simulations for every possible sea state, engineers typically reduce the problem by selecting a set of variables and binning or lumping sea states to limit the number of required simulations. However, for floating wind turbines, the choice of which variables to include and how to perform this binning remains an open question. This motivates the use of reliable data-driven surrogate models that can make quick estimates of loads on wind turbines while maintaining high accuracy.

Although most existing work relies on deterministic surrogate models, offshore wind environments exhibit strong stochasticity, making turbine loads inherently uncertain. This motivates the need for uncertainty quantification to characterize not only expected loads but also estimate their conditional variability. This dissertation, therefore, develops a probabilistic data-driven methodology that propagates environmental uncertainty to the 10-minute damage equivalent loads of onshore, offshore, and floating offshore wind turbines.

Several deterministic and probabilistic data-driven models are benchmarked on onshore and fixed-bottom offshore wind turbines. The evaluation consists not only of judging the accuracy of a model, but also of practical aspects such as the robustness of the model, sensitivity to hyperparameters, and ease of implementation. Compared to deterministic approaches, which require multiple seed repetitions prior to training, it is demonstrated that with probabilistic models, this step may not be necessary to achieve high accuracy predictions (𝑅2 > 0.95), thereby saving precious computational resources needed to generate the training database. Widely used Gaussian process regression is shown to accurately estimate the conditional mean of the response with a relatively small training dataset of Q102 - 103) samples. Wasserstein-conditional generative adversarial network is used as one of the probabilistic regression models. Despite learning the functional mapping with errors comparable to the best-performing models, it is found to be very complex to implement and requires extensive hyperparameter tuning. Simpler 4th-degree polynomials are shown to make good predictions in the onshore case, but are prone to overfitting and susceptible to the additional noise introduced by the hydrodynamic features. Overall, mixture density networks are shown to provide the best combination of consistency, accuracy, and practical ease of use.

Based on this analysis, the study extends the use of mixture density networks to a more sophisticated application of spar-type floating offshore wind turbine. It is shown to successfully capture the conditional response in terms of the normalized 2-Wasserstein distance despite the added complexity. The surrogate is further used to make probabilistic estimates of the lifetime damage equivalent loads on four potential floating wind turbine sites. Since the surrogate model is fast (order of milliseconds once trained), load predictions can be made over all sea states within seconds, without the need to lump or bin the sea states beforehand. The uncertainty in the aggregated lifetime fatigue loads due to stochastic inputs is extremely narrow, with variability on the order of only 0.1–0.5% of their mean values. This results from summing the 10-minute damage equivalent loads over a million occurrences, effectively nullifying the impact of the outliers. The use of a probabilistic surrogate that correctly captures the conditional distribution is still useful, as it minimizes the aggregation of error in the final response.

Through these analyses, it is demonstrated that surrogate models can be powerful tools for fatigue estimation in the site analysis process, especially for floating wind turbines, where the choice of variables and binning methods is still an open question. Additionally, using probabilistic surrogates like mixture density networks helps reduce bias in calculating the aggregate mean fatigue, as the conditional distributions are heteroscedastic and not always normally distributed.

Floating offshore wind turbines (FOWTs) experience complex hydrodynamic and aerodynamic loading influenced by substructure types and stochastic environmental conditions. Accurately estimating the lifetime fatigue loads requires the analysis of thousands of operational scenarios, leading to high computational costs. Moreover, choosing the right input features driving fatigue in floating wind systems and appropriately binning them still remains an open question. We present a fast probabilistic surrogate that maps the site conditions to the loads on the wind turbine. The probabilistic aspect allows the propagation and quantification of statistical uncertainties from the stochastic input quantities to the resulting loads. A fast surrogate eliminates the need to fit a distribution to the site conditions or bin the input data. Rather, all available metocean data can be directly used as input, which automatically accounts for the joint distribution in the calculations. The surrogate model in this study uses the mixture density network (MDN) to predict the conditional distribution of the 10 min damage equivalent loads (DELs) for a 6 MW spar-type floating wind turbine. The MDN achieves high accuracy (R2>0.99) in capturing DEL means while efficiently propagating the statistical uncertainties. Furthermore, the surrogate enables quick estimation of 25-year lifetime fatigue damage across a range of potential floating wind farm sites, demonstrating its capability to facilitate rapid decision-making during preliminary site analysis.

The use of load surrogates in offshore wind turbine site assessment has gained attention as a way to speed up the lengthy and costly siting process. We propose a novel probabilistic approach using mixture density networks to map 10 min average site conditions to the corresponding load statistics. The probabilistic framework allows for the modeling of the uncertainty in the loads as a response to the stochastic inflow conditions. We train the data-driven model on the OpenFAST simulations of the IEA 10 MW reference wind turbine (IEA-10MW-RWT) and compare the predictions to the widely used Gaussian process regression. We show that mixture density networks can recover the accurate mean response in all load channels with values for the coefficient of determination (R2) greater than 0.95 on the test dataset. Mixture density networks completely outperform Gaussian process regression in predicting the quantiles, showing an excellent agreement with the reference. We compare onshore and offshore sites for training to conclude the need for a more extensive training dataset in offshore cases due to the larger feature space and more noise in the data.

Floating offshore wind turbines can extract energy from deep offshore locations,
typically unfit for fixed bottom designs. The complex interaction between the structural
behavior of the floating offshore wind turbine and the stochastic site conditions, however,
is an active area of research. Characterizing the relationship between the environmental
conditions and loads may help design reduced-order models, surrogate models, and physicsbased engineering models for floating wind turbines. This study uses data from the TetraSpar prototype equipped with a 3.6 MW Siemens Gamesa wind turbine. One-to-one simulations performed using an aero-servo-hydro-elastic software are included for comparison. Various tools, including linear correlation, mutual information, feature ordering using conditional independence, and sensitivity analysis using a data-driven variogram fit, are used for the assessment. This study is also helpful in validating the engineering model for future global sensitivity analysis using elementary effects or Sobol indices that require a rigid sampling of features and can, therefore, only be calculated with simulation tools. We find a good agreement between the experiments and simulations. The 10-min. damage equivalent loads on the tower show a correlation, particularly with the wind speed statistics and the significant wave height.

This paper presents a surrogate-assisted optimisation approach to speed up the substructure analysis in the preliminary design phase. The approach consists of replacing the radiation-diffraction analysis in a frequency domain analysis model for floating wind turbines with a data-driven surrogate model predicting the hydrodynamic coefficients for parameterised substructure geometries. This procedure is compared with the reference approach of estimating the hydrodynamic coefficients via radiation-diffraction analysis. A representative use case of assessing the trade-off between minimising the capital cost and reducing the wave-induced nacelle acceleration standard deviation for a semi-submersible substructure is presented. The accuracy of the surrogate model is found to increase significantly up to training datasets consisting of 400 designs and less noticeably afterwards. For a dataset consisting of 400 designs, the mean error on the prediction of the hydrodynamic coefficients and the error at one standard deviation from the mean are generally below 7% and 10%, respectively. For the same dataset size, the mean error on the most probable maximum wave-induced pitch over a 3h storm period is below 17%, while the error at one standard deviation from the mean is lower than 27%. The same values for the most probable maximum nacelle acceleration are under 7% and 12%, respectively. The surrogate model can capture the trade-off between the two objective functions, and the optimal designs identified with the surrogate model generally follow the same trend as those obtained with the reference model. However, relying on the surrogate model for performing the analysis of the substructure introduces local minima in the objective function that cause a discrepancy between the optimal designs identified with the surrogate model and those identified with the reference model.

Long Short-Term Memory Recurrent Neural Networks (LSTM) are used to build surrogate models to forecast time-series blade loads for both fixed and floating offshore wind turbines. In this paper, we train surrogate models on datasets generated with OpenFAST on the IEA-15MW-RWT under a range of metocean conditions. The aim of the surrogate models is to generate load forecasts inexpensively and accurately such that they can be used in a model predictive controller. Two cases are investigated with different model inputs: one with only measurements available to typical PI controllers and another one with additional wave elevation and deflection measurements (alongside the endogenous variable). The model performances are evaluated and compared. It was found that for the fixed turbine, the models predicted all three blade loads to a high degree of accuracy. The floating turbine surrogate models performed relatively worse, but edgewise and pitching moments are still reasonably accurate. The surrogate model forecasts the flapwise moment to a satisfactory accuracy only in 58% out of 400 test cases. The addition of wave elevation and blade deflection features did not significantly improve the prediction performance of the surrogate, demonstrating that just the information used by current PI controllers may be sufficient for forecasting blade loads.

Reliable prediction of aviation’s environmental impact, including the effect of nitrogen oxides on ozone, is vital for effective mitigation against its contribution to global warming. Estimating this climate impact however, in terms of the short-term ozone instantaneous radiative forcing, requires computationally-expensive chemistry-climate model simulations that limit practical applications such as climate-optimised planning. Existing surrogates neglect the large uncertainties in their predictions due to unknown environmental conditions and missing features. Relative to these surrogates, we propose a high-accuracy probabilistic surrogate that not only provides mean predictions but also quantifies heteroscedastic uncertainties in climate impact estimates. Our model is trained on one of the most comprehensive chemistry-climate model datasets for aviation-induced nitrogen oxide impacts on ozone. Leveraging feature selection techniques, we identify essential predictors that are readily available from weather forecasts to facilitate the implementation therein. We show that our surrogate model is more accurate than homoscedastic models and easily outperforms existing linear surrogates. We then predict the climate impact of a frequently-flown flight in the European Union, and discuss limitations of our approach.

While efforts have been made to curb CO2 emissions from aviation, the more uncertain non-CO2 effects that contribute about two-thirds to the warming in terms of radiative forcing (RF), still require attention. The most important non-CO2 effects include persistent line-shaped contrails, contrail-induced cirrus clouds and nitrogen oxide (NOx) emissions that alter the ozone (O3) and methane (CH4) concentrations, both of which are greenhouse gases, and the emission of water vapour (H2O). The climate impact of these non-CO2 effects depends on emission location and prevailing weather situation; thus, it can potentially be reduced by advantageous re-routing of flights using Climate Change Functions (CCFs), which are a measure for the climate effect of a locally confined aviation emission. CCFs are calculated using a modelling chain starting from the instantaneous RF (iRF) measured at the tropopause that results from aviation emissions. However, the iRF is a product of computationally intensive chemistry-climate model (EMAC) simulations and is currently restricted to a limited number of days and only to the North Atlantic Flight Corridor. This makes it impossible to run EMAC on an operational basis for global flight planning. A step in this direction lead to a surrogate model called algorithmic Climate Change Functions (aCCFs), derived by regressing CCFs (training data) against 2 or 3 local atmospheric variables at the time of emission (features) with simple regression techniques and are applicable only in parts of the Northern hemisphere. It was found that in the specific case of O3 aCCFs, which provide a reasonable first estimate for the short-term impact of aviation NOx on O3 warming using temperature and geopotential as features, can be vastly improved [1]. There is aleatoric uncertainty in the full-order model (EMAC), stemming from unknown sources (missing features) and randomness in the known features, which can introduce heteroscedasticity in the data. Deterministic surrogates (e.g. aCCFs) only predict point estimates of the conditional average, thereby providing an incomplete picture of the stochastic response. Thus, the goal of this research is to build a new surrogate model for iRF, which is achieved by : 1. Expanding the geographical coverage of iRF (training data) by running EMAC simulations in more regions (North & South America, Eurasia, Africa and Australasia) at multiple cruise flight altitudes, 2. Following an objective approach to selecting atmospheric variables (feature selection) and considering the importance of local as well as non-local effects, 3. Regressing the iRF against selected atmospheric variables using supervised machine learning techniques such as homoscedastic and heteroscedastic Gaussian process regression. We present a new surrogate model that predicts iRF of aviation NOx-O3 effects on a regular basis with confidence levels, which not only improves our scientific understanding of NOx-O3 effects, but also increases the potential of global climate-optimised flight planning.

Heteroscedastic Gaussian process regression, based on the concept of chained Gaussian processes, is used to build surrogates to predict site-specific loads on an offshore wind turbine. Stochasticity in the inflow turbulence and irregular waves results in load responses that are best represented as random variables rather than deterministic values. Moreover, the effect of these stochastic sources on the loads depends strongly on the mean environmental conditions - for instance, at low mean wind speeds, inflow turbulence produces much less variability in loads than at high wind speeds. Statistically, this is known as heteroscedasticity. Deterministic and most stochastic surrogates do not account for the heteroscedastic noise, giving an incomplete and potentially misleading picture of the structural response. In this paper, we draw on the recent advancements in statistical inference to train a heteroscedastic surrogate model on a noisy database to predict the conditional pdf of the response. The model is informed via 10-minute load statistics of the IEA-10MW-RWT subject to both aero- and hydrodynamic loads, simulated with OpenFAST. Its performance is assessed against the standard Gaussian process regression. The predicted mean is similar in both models, but the heteroscedastic surrogate approximates the large-scale variance of the responses significantly better.