Department - Author 1
Electrical Engineering Department
Degree Name - Author 1
BS in Electrical Engineering
Date
3-2010
Primary Advisor
Xiao-Hua (Helen) Yu
Abstract/Summary
This project involved designing and programming an artificial neural network, testing its function, and testing the resulting function against existing wind speed measurement data, in order to determine the ability of an artificial neural network to learn the relationship between measured data and future wind speed. The artificial neural network includes gradient momentum, batch training, incremental training, and a function to test the results of the trained neural network against an additional set of data without back propagation. It was first found that the artificial neural network was able perform unsupervised learning, and learn the model for a XOR gate. At a learning rate of .1 and a gradient momentum of .05, the network was able to learn a XOR gate function to an absolute error of .025 in 450 iterations of batch training and testing found an absolute error of .019.
When testing the ability of the neural network to learn in a wind speed prediction application, it was found that an artificial neural network can produce a good prediction of the wind speed. The prediction of the wind speed in the next 1 minute interval was found during the testing portion of the run to reach an average absolute difference of .032. For the use of logarithmic scaling function, an average absolute difference of .0336 was obtained.
Comparing this to a non-linear regression technique for determining the coefficients in a wind speed model, the absolute error of .0234 from the traditional method was less than the ANN results.
URL: https://digitalcommons.calpoly.edu/eesp/24