The integration of offshore wind power into modern electricity grids and markets is critically dependent on the accuracy of short-term power forecasts. High-quality forecasts support grid stability, facilitate market participation, and reduce balancing costs. However, generating reliable offshore wind power forecasts remains a significant challenge due to the inherently volatile nature of offshore wind regimes, which are characterized by rapid power ramps and complex meteorological patterns. Additionally, power curtailments caused by grid congestion further complicate the forecasting task by introducing inconsistencies between actual generation potential and observed outputs. This study presents the development and evaluation of advanced offshore wind power forecasting models using machine learning techniques, like gradient boosting methods (XGBoost) and neural network-based approaches (multi-layer perceptrons, MLPs) as well as classical time series methods such as Autoregressive Integrated Moving Average (ARIMA) to assess model generalizability and capture temporal dependencies.

A comprehensive analysis of input features, including numerical weather prediction (NWP) data, production values of past time steps and extrapolated curtailment signals from real wind speed measurements, is conducted to understand their impact on model performance.

The paper addresses methods to pre-process data affected by curtailment and explores techniques to improve the model’s ability to distinguish between physical variability (i.e. windspeed) and grid-induced limitations, such as grid congestions. Through a series of experiments using real-world offshore wind farm data, the study quantifies the forecasting skill of the proposed machine learning models under various scenarios, including high-ramp conditions and periods of curtailment.

Results demonstrate that, with appropriate feature engineering and data treatment, models like XGBoost and MLP can achieve high forecast accuracy, offering a promising approach for operational forecasting in offshore environments. These findings highlight the potential of machine learning, particularly ensemble and neural network models, to enhance the precision and reliability of offshore wind power forecasts—ultimately supporting the secure and efficient integration of offshore wind energy into the power system.

This work was carried out within the Femtec Innovation Lab, a program targeted towards female STEM students to connect them to partners from industry and science. Through interdisciplinary project work this fosters collaborative working with the main focus of increasing the participation and representation of women in STEM fields.