Context
Wind generation assets require continuous performance monitoring to ensure optimal output and detect deviations early.
Challenge
Identifying underperformance or inefficiencies at the turbine level is difficult without continuous comparison against expected performance benchmarks.
AI Solution
The solution creates turbine-specific reference power curves for each turbine in the fleet. The AI model compares how much power a wind turbine should produce versus what it actually produces. It helps spot performance issues early so turbines can run more efficiently and reliably. The solution uses machine learning models to predict turbine-specific reference power curves, starting with linear and polynomial regression, exploring tree ensemble models and artificial neural networks (ANNs), with Gradient Boosting selected as the best-performing model.
Impact & evidence
The solution has significantly improved operational clarity by replacing ambiguity in power curve reviews with clear, colour-coded insights, enabling the wind operations team to quickly identify and verify turbine performance issues. This has accelerated issue resolution through immediate engagement with the OEM, helping avoid potential revenue losses. A concrete example is the identification and correction of an incorrect noise mode implementation, which resulted in an estimated revenue increase of €300k–400k annually. While still in its early stages, the tool also shows strong potential to further enhance performance monitoring by enabling rapid assessment of the impact of site changes.
Implementation / Adoption
To implement this solution, other organisations would need access to long-term historical SCADA data and the capability to build machine-learning models that generate turbine-specific reference power curves. The approach also requires a phased rollout, starting with pilot sites, followed by iterative model refinement and scaling across assets through regularly refreshed dashboards. In addition, early stakeholder engagement and integration into wind operations are needed to support adoption and day-to-day use. The setup should also allow for onboarding new assets and adding forecasting capabilities over time.
Data
The solution uses 10-15 years of historical wind Supervisory Control and Data Acquisition (SCADA) data.
Future enhancements
Planned future enhancements include integration of new assets and forecasting.
“The reference power curve and associated dashboard is a game changer in how the Wind Operation team tracks and verifies the performance of the turbines. Where previously there was ambiguity in the power curve reviews, now there is clarity. Any issues can be picked out quickly due to clever colour coding, which allows immediate engagement with the OEM to resolve the issue and avoid lost revenue.
In its first month of operation, the new tool highlighted the incorrect implementation of a noise mode on 4 turbines on one of our wind farms. When pointed out to the OEM the fix was instant and has resulted in an estimated increased revenue of 300k–400k per annum.
The tool is only in its infancy, but in time we expect to gain massive value from being able to quickly verify any changes on site such as services, software upgrade, tree felling and blade works, have had a positive or negative impact on the performance of the turbine against its now known reference.” – Sean Og Gargan – Wind Operations Manager (ESB Generation Trading)