Posters

Abstract

During the development of offshore wind farms, designers, contractors and operators require long-term time series of ocean parameters (waves, water level, current speed, and direction, etc.) at several stages during the process. Present industry practices rely on costly accurate high resolution local model data for detailed design, but more approximate data from e.g. existing databases in early-stage project development and for areas outside the wind farm important to e.g. cable routes and planning of installation. Both accessibility speed of data (months or hours) and precision are paramount to the optimization of wind farm layouts, construction and operation for ensuring not only the structural integrity of the turbines but also the overall efficiency and reliability of the energy generation system. To address the challenge of the missing local data and modeling of complex physical processes which require significant computational time, we propose a methodology and model framework that can satisfy mostly comparable requirements as the classical numerical modelling simulations but in a more computationally efficient and time effective manner. The MIKE Metocean Simulator is a hybrid model, which combines state-of-the-art machine learning algorithms and numerical models that tries to predict the behavior of the sea parameters in a specific point of interest with a significant reduction of the computational time and simultaneously providing minimal accuracy degradation. The method is already available for wave modelling (MIKE Metocean Simulator (mikepoweredbydhi.com) [https://www.mikepoweredbydhi.com/products/mike-cloud/mike-metocean-simulator/]) and we here present a novel extension to hydrodynamic modelling. The model forcing state vector of the hybrid model is similar to that of a numerical hydrodynamic model and thus entails the water level, horizontal and vertical components of current speed in each boundary together with the atmospheric pressure and wind components at 10m. The model outputs are the water level and depth averaged ocean current speed and direction at a specific point. More specifically, the dimensionality reduction method, Principal Component Analysis, will be used to reduce the forcing state vector space facilitating the required hybrid modelling selection and interpolation process. The event selection methodology includes the extraction of a subset of timesteps with representative sea states. For this purpose, the Maximum Dissimilarity Algorithm is used which can capture the most dissimilar cases and ensure the selection of the most extreme events. The modelling of the time series of the sea state at the output locations is a reconstruction based on the statistical interpolation techniques. Results using both the Radial Basis Function and Gaussian Process Regression are presented. Finally, the validation of the interpolated times series with simulated and real data is performed. A variety of statistical tools is used as validation metrics while graphical methods like scatter plots are shown. The methodology has been tested in a case study on the eastern coast of Denmark, and the results confirm that the hydrodynamic flow-module of MIKE Metocean Simulator can reconstruct the 40-year time series of the desired parameters with a high accuracy at a point of approximately 250 times faster than a classical hydrodynamic simulation.