Comparison of the Wind Power Output of a Small Scale Turbine using Statistical and Feed-Forward Neural Network Techniques

Zaccheus O. Olaofe and Komla A. Folly


Site Power Curve, Average Power Output, Feed-Forward Neural Network (FNN), Root Mean Square Error (RMSE)


This paper presents two techniques for estimating the wind power output at Napler wind site: the use of statistical technique, where the power output of the turbine is estimated based on the developed site power curve; and the use of feed forward neural network (FNN) technique, where the power output of the turbines are estimated based on the artificial intelligence (AI). The FNN utilizes the supervised learning technique where the weather data and the target (desired output) are used to train the network. The site’s wind power potential is evaluated using a 10-minute weather data obtained along the coast of South Africa, recorded for the period of 21 months on a 20m hub height. A small scale 40kW wind turbine is sized for the available wind resources at this height. Results show that the statistical analysis of the wind resources based on the site power curve is an accurate technique for wind energy estimation. In addition, the FNN provides accurate prediction of the wind power output of the 40kW turbine as the wind fluctuates, and the predictions are compared with the wind power output obtained from the statistical technique. The time series wind power estimates using the feed-forward neural network (FNN) technique shows a strong agreement with the estimation obtained using the statistical technique (site power curve). The available energy generation based on the available 21 months wind data at 20m height are estimated at 88.62MWh, 70.95MWh, 70.91MWh using the turbine power curve, site power curve & FNN, respectively. Furthermore, a 2 days power generation is predicted in advance.

Important Links:

Go Back