UC Santa Cruz

This paper proposes a model for making power purchasing and routing decisions at the prosumer level and presents forecasting techniques that can be used in the optimization of that model.

To implement this model, the author has built and installed tracking solar panels, solar radiation sensors, power measurement devices, and automated data collections systems at the University of California Santa Cruz (UCSC) and the National Aeronautics and Space Administration (NASA) Renewable Energy Lab (RE Lab) in Mountain View, CA. Further, the author has programmed data collection and database tools, Artifical Neural Networks, and data analysis tools to capitalize on the data collected at the RE Lab. This paper reviews the use of Artificial Neural Networks (ANNs), a powerful forecasting method that has proven to be among the most accurate of the machine learning algorithms, for the prediction of power production and consumption. This paper includes analysis of ANNs for the prediction of solar and wind power production and individual power consumption. Finally, a new method of feature selection is proposed and found to improve short term solar power predictions.