A Scenario Optimization Approach to the Prediction of Extreme Structural Responses to Environmental Loading

This paper proposes a data-driven approach to characterize the extreme response of a wind turbine subject to both uncertain properties and random loading. Whereas the loading is driven by environmental conditions measured in situ, the response is evaluated by computationally intensive simulations. The training data used to build the computational model might be given by the greatest simulated responses or by high quantiles of fitted worst-case distributions. Worst-case and chance-constrained optimization are used to learn functions that bound the data. We may choose to bound all the data in order to obtain a conservative prediction, or we can deliberately exclude a few outliers from the dataset in order to obtain a more informative but riskier prediction. In addition, we use scenario theory to rigorously bound the probability of future data exceeding the predicted extreme response.