Impact of Forecast Uncertainty in O&M scheduling for OWF, FCM Machine Learning integration

The exploitation of offshore wind farm in energy production necessitates a reduction in cost associated to maintenance and operation that highly contribute to the total cost of the technology. Maintenance activity scheduled as preventive intervention, reduces the probabilities of failure for the turbines, decreasing economic losses due to production loss. Scheduling the operations with priority or urgency considerations, permit to exploit the activities on days with lower wind resources, maximizing the energy production. Contemplating the perturbations of the wind flow created from the wake effect by upstream turbines for specific wind direction along the farm, is possible to optimize the downtime energy losses. In fact, by a selectively deactivation of specific turbines for maintenance activities, the minimization of energy losses could be pursued, impacting on the overall energy production. For so, the knowledge of accurate weather conditions for the specific wind farm location becomes fundamental. Uncertainties in weather forecast might affect the decisions taken in maintenance scheduling, reaching suboptimal or non-optimal results. The present study evaluates the impact of forecast uncertainties in the optimization of scheduling preventive maintenance operations and quantifies their energetics consequences. The results obtained show lower optimizations than expected, due to inaccuracy of weather forecasted. An integration with a trained data-driven method with historical data is proposed to reduce forecast uncertainties in the estimation of wind directions. The performance of the Fuzzy C Means algorithm used to improve accuracy of the forecast varied across different months and wind speed scenarios analyzed. An improvement in the accuracy of the estimation of wind direction for maintenance purpose it has been reached obtain a small decrement of energy losses. The algorithm shows higher precision in annual energy production estimation. These findings underscore the potential of data-driven methods in mitigating weather forecast uncertainties for remote locations and the possibilities to apply them in maintenance scheduling activities for offshore wind farms.