Key optimization problems on offshore wind farm based on machine learning

Abstract

Wind power is playing the leading role in the race to achieve the Paris Agreement goal of 1.5℃ global warming. Offshore wind power has a record year with over 21 GW new installations in 2021. This study mainly focuses on key optimization problems on offshore wind farms. Firstly, an integrated optimization framework for large-scale wind turbine blades is proposed to efficiently provide appropriate blade designs for the next-generation offshore wind turbines. Afterward, an ANN wake model is proposed and trained using a rigorously validated CFD simulation database. In the error analysis based on an independent testing dataset, the error percentage of wind velocity is smaller than 2%, while the absolute error of turbulence intensity is lower than 0.01. By introducing the superposition models, the power output of wind farms can be accurately predicted. Moreover, a data-driven layout optimization framework is proposed for large-scale offshore wind farms. A gradient-based local search algorithm is combined with a global search method, the multi-start method. A widely accepted analytical model is selected for comparison, which shows an error of 7~10% in overall power prediction. Consequently, the optimal solution of the analytical framework is about 4% worse than that of the ANN framework. The proposed ANN framework highly agrees with the CFD simulation. The power prediction error is lower than 1% in the validations. By adopting WFLO, the power production of the Horns Rev wind farm can increase by 28.77%, which can further improve by 118.63% when the number of wind turbines is doubled. Later the influence of wind conditions on the optimal potentials of wind farms is analyzed. Three indices are discussed: wind energy density, energy entropy, and wind direction. Linear regression and multiple regression on the optimal potential are conducted with wind energy density and energy entropy. The multiple regression considering both indices shows a higher correlation than linear regression with a single index. Meanwhile, the wind direction is found to have a limited impact on the performance of optimal layouts and mainly affects the initial layout performance. The realistic wind data is also investigated and classified into three categories using the hierarchical clustering method: European offshore area, Chinese offshore area, and other areas. With the realistic wind roses, Horns Rev wind farm and Lillgrund wind farm are studied using the layout optimization framework. The optimal potential is limited to 3.37% by different limiting factors. Thus, two suggestions for improving the offshore wind farm performance are proposed: changing the number of wind turbines and adjusting the tower heights of half of the wind turbines. The results show a decrease in the cost of energy of 12~25%. Overall, this study concentrates on improving offshore wind farm performance. The results of this study will be useful in offshore wind farm design. The discussions will also benefit when detailed standards are in the design stage.