Exploring Gaussian processes for short-term forecasting in offshore energy systems

Abstract

Offshore renewable energy (ORE) systems are expected to play a pivotal role in addressing global climate challenges by harnessing renewable energy sources from the ocean. This study explores short-term forecasting (predicting a few seconds into the future), to enhance the performance and reliability of ORE. Forecasting wave excitation force and wave height is essential for the management of operations, ensuring efficient energy extraction and safeguarding against potential risks. Thus, in this study Gaussian processes (GPs) are studied as a powerful forecasting tool, utilising experimental data from both lab-scale WEC systems and real-world wave measurements. The presented study evaluates various GP kernel functions, such as Squared Exponential, Periodic, Matérn and Gaussian Mixture, under diverse sea state conditions and forecasting horizons, ranging from fractions of the typical sea state period to twice that period. Results demonstrate the higher predictive accuracy and reliability of GP compared to the more common autoregressive methods, highlighting its ability to effectively model complex data patterns and uncertainties inherent to offshore environments. This comprehensive analysis underscores the capability of GPs to enhance operational decision-making in offshore energy, contributing to improved performance, efficiency and safety of ORE technologies.