Modelling and analysis of real-world wind turbine power curves: Assessing deviations from nominal curve by neural networks. (September 2019)

• Record Type: Journal Article
• Authors: Ciulla, G.; D'Amico, A.; Di Dio, V.; Lo Brano, V.
• Is Part Of: Renewable energy. Volume 140(2019)
• Page Start: 477
• Page End: 492
• Publication Date: 2019-09
• Subjects: Wind energy, power curve -- Producibility estimates -- Aero-generator -- Anemometric campaign -- Artificial neural network; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2019.03.075
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: The power curve of a wind turbine describes the generated power versus instantaneous wind speed. Assessing wind turbine performance under laboratory ideal conditions will always tend to be optimistic and rarely reflects how the turbine actually behaves in a real situation. Occasionally, some aerogenerators produce significantly different from nominal power curve, causing economic losses to the promoters of the investment. Our research aims to model actual wind turbine power curve and its variation from nominal power curve. The study was carried out in three different phases starting from wind speed and related power production data of a Senvion MM92 aero-generator with a rated power of 2.05 MW. The first phase was focused on statistical analyses, using the most common and reliable probability density functions. The second phase was focused on the analysis and modelling of real power curves obtained on site during one year of operation by fitting processes on real production data. The third was focused on the development of a model based on the use of an Artificial Neural Networks that can predict the amount of delivered power. The actual power curve modelled with a multi-layered neural network was compared with nominal characteristics and the performances assessed by the turbine SCADA. For the studied device, deviations are below 1% for the producibility and below 0.5% for the actual power curves obtained with both methods. The model can be used for any wind turbine to verify real performances and to check fault conditions helping operators in understanding normal and abnormal behaviour. Highlights: A neural network model of wind turbine power curve is developed from real data. Accurate statistical analyses are carried out on wind and weather data. Neural network model is based on several, often neglected, parameters. Results demonstrate effectiveness of the proposed method.