Power prediction of offshore wind farms based on a fusion-driven of physical knowledge and operational data

Power prediction is a fundamental research topic in wind industry. Offshore wind power prediction mostly relies on either data-driven or physics-based approaches. Few approaches combine physical knowledge and operational data. Nevertheless, there is significant potential for complementarity between these two approaches. In this study, a physics-based Gaussian wake model for wind farms is first constructed, and parameters of the empirical wake model are optimally identified by Particle Swarm Optimization algorithm based on the actual operational data. A purely data-driven power prediction method is constructed through K-means clustering and parallel weighting Long Short-Term Memory with empirical mode decomposition. Based on these methods, an innovative fusion approach combining the physics-based wake model with the data-driven method is constructed using symbolic regression. Taking the real measured data from an offshore wind farm in Jiangsu, China, as a case study, the results show that the accuracy of the proposed approach is 21.67 % higher than that of the data-driven approach and 35.17 % higher than that of the physics-based approach. These results confirm the superiority of the physics-data fusion approach for wind farm power prediction. • A physical wake model for wind farm power prediction based on parameter optimization utilizes the Particle Swarm Optimization algorithm with actual operational data. • The measured data are processed by exploiting the local temporal regularity, and the proposed clustering weight network achieves the power prediction of wind farms in a data-driven manner. • A power prediction of wind farms method based on a fusion-driven of physical-knowledge and operational-data is innovatively constructed through symbolic regression methods.