Ocean Wave Prediction for Arrays of Wave Energy Converters
(2026)Department of Automatic Control
- Abstract
- Short-term wave prediction is a key enabling technology for advanced control of wave energy converters (WECs), where accurate knowledge of the incident wave field over one to a few wave periods can improve power capture, reduce structural loads, and support operational decision-making. Many existing prediction methods rely on measurements at the prediction point or dedicated up-wave sensors, which are costly and impractical for dense WEC arrays. This thesis investigates whether a WEC array can instead function as a distributed sensing system for short-term wave prediction using only measurements available within the array.
Using numerical simulations representative of the Aguçadoura test site, several array-level wave prediction models... (More) - Short-term wave prediction is a key enabling technology for advanced control of wave energy converters (WECs), where accurate knowledge of the incident wave field over one to a few wave periods can improve power capture, reduce structural loads, and support operational decision-making. Many existing prediction methods rely on measurements at the prediction point or dedicated up-wave sensors, which are costly and impractical for dense WEC arrays. This thesis investigates whether a WEC array can instead function as a distributed sensing system for short-term wave prediction using only measurements available within the array.
Using numerical simulations representative of the Aguçadoura test site, several array-level wave prediction models are developed and compared. The primary focus is on covariance-based DMS prediction models, which model wave elevation as a stationary Gaussian process and exploit spatial and temporl correlations between multiple devices. Unfiltered and filtered Direct Multi-Step (DMS) variants are evaluated alongside purely data-driven artificial neural network (ANN) prediction models and a hybrid model combining linear DMS prediction with a neural-network-based residual correction. Performance is assessed using metrics consistent with industrial practice for one-, two-, and three-buoy configurations.
The results show that array-level prediction consistently outperforms single-point forecasting under simulated linear wave conditions, both at the array-average level and for individual devices. The unfiltered covariance-based DMS prediction model provides robust and computationally efficient performance across most operational sea states. Accuracy is largely insensitive to wave height, mean direction, and directional spreading when averaged across the array, but degrades with increasing wave period.
Under simulated weakly nonlinear wave conditions, the DMS prediction model exhibits systematic performance degradation. In sea states with strong nonlinear effects, accuracy reductions of up to approximately 40% are observed, indicating that nonlinear wave dynamics can substantially limit purely linear statistical models. The online-trained ANN prediction model provides selective improvements in some energetic sea states, particularly for the threebuoy configuration, but does not consistently outperform the DMS prediction model and does not fully recover nonlinear performance losses. The hybrid prediction model does not yield measurable gains in the tested configurations, although its design space has not been exhaustively explored.
Overall, the results demonstrate that array-level short-term wave prediction is feasible without dedicated up-wave sensors and that covariance-based DMS prediction models provide a strong baseline for practical deployment. However, in simulated nonlinear sea states with strong nonlinear effects, performance degradation becomes substantial, and online-trained nonlinear prediction model extensions offer limited mitigation, suggesting that more dataintensive or pretrained approaches may be required for robust performance under highly nonlinear wave conditions. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9225598
- author
- Järudd, Hugo
- supervisor
- organization
- year
- 2026
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6306
- other publication id
- 0280-5316
- language
- English
- id
- 9225598
- date added to LUP
- 2026-05-11 09:32:17
- date last changed
- 2026-05-11 09:32:17
@misc{9225598,
abstract = {{Short-term wave prediction is a key enabling technology for advanced control of wave energy converters (WECs), where accurate knowledge of the incident wave field over one to a few wave periods can improve power capture, reduce structural loads, and support operational decision-making. Many existing prediction methods rely on measurements at the prediction point or dedicated up-wave sensors, which are costly and impractical for dense WEC arrays. This thesis investigates whether a WEC array can instead function as a distributed sensing system for short-term wave prediction using only measurements available within the array.
Using numerical simulations representative of the Aguçadoura test site, several array-level wave prediction models are developed and compared. The primary focus is on covariance-based DMS prediction models, which model wave elevation as a stationary Gaussian process and exploit spatial and temporl correlations between multiple devices. Unfiltered and filtered Direct Multi-Step (DMS) variants are evaluated alongside purely data-driven artificial neural network (ANN) prediction models and a hybrid model combining linear DMS prediction with a neural-network-based residual correction. Performance is assessed using metrics consistent with industrial practice for one-, two-, and three-buoy configurations.
The results show that array-level prediction consistently outperforms single-point forecasting under simulated linear wave conditions, both at the array-average level and for individual devices. The unfiltered covariance-based DMS prediction model provides robust and computationally efficient performance across most operational sea states. Accuracy is largely insensitive to wave height, mean direction, and directional spreading when averaged across the array, but degrades with increasing wave period.
Under simulated weakly nonlinear wave conditions, the DMS prediction model exhibits systematic performance degradation. In sea states with strong nonlinear effects, accuracy reductions of up to approximately 40% are observed, indicating that nonlinear wave dynamics can substantially limit purely linear statistical models. The online-trained ANN prediction model provides selective improvements in some energetic sea states, particularly for the threebuoy configuration, but does not consistently outperform the DMS prediction model and does not fully recover nonlinear performance losses. The hybrid prediction model does not yield measurable gains in the tested configurations, although its design space has not been exhaustively explored.
Overall, the results demonstrate that array-level short-term wave prediction is feasible without dedicated up-wave sensors and that covariance-based DMS prediction models provide a strong baseline for practical deployment. However, in simulated nonlinear sea states with strong nonlinear effects, performance degradation becomes substantial, and online-trained nonlinear prediction model extensions offer limited mitigation, suggesting that more dataintensive or pretrained approaches may be required for robust performance under highly nonlinear wave conditions.}},
author = {{Järudd, Hugo}},
language = {{eng}},
note = {{Student Paper}},
title = {{Ocean Wave Prediction for Arrays of Wave Energy Converters}},
year = {{2026}},
}