Posters

Abstract

One of the main criticisms about wind energy development is the threat it poses to aerial biodiversity. The most common solutions to reduce the risk of bird mortality in a wind farm are sound deterrence emitted by the turbine and shutting down the turbines to avoid collision with the blades. Both solutions involve the use of real-time detection of birds approaching the turbines. This detection is performed by a specific technology using cameras placed on the mast of the wind turbine and a machine learning algorithm. Developed specifically for this application, this neural network aims to detect any flying object and to classify it as a "large bird" or "noise". The innovation of the system lies in the classification method: it analyses the moving objects detected by camera on an image stacking. The silhouette and the trajectory of an object are some elements which allow its classification. This classification allows calculation of a collision risk level by integrating the time of presence of the bird in the field of vision of the camera. Depending on this level, the system will emit a dissuasive sound or an order to stop the wind turbine concerned. The data collected from the system installed on six wind turbines of two wind farms (20,5 MW) in a geographical area with high environmental value and an important activity of several protected species allows the continuous improvement of the classification. Such technology demonstrates satisfactory results with less than 10% of false-positives and less than 5% of false-negatives.