Data Analysis for Solar Energy Generation in a University Microgrid

This paper presents a data acquisition process for solar energy generation and then analyzes the dynamics of its data stream, mainly employing open software solutions such as Python, MySQL, and R. For the sequence of hourly power generations during the period from January 2016 to March 2017, a variety of queries are issued to obtain the number of valid reports as well as the average, maximum, and total amount of electricity generation in 7 solar panels. The query result on all-time, monthly, and daily basis has found that the panel-by panel difference is not so significant in a university-scale microgrid, the maximum gap being 7.1% even in the exceptional case. In addition, for the time series of daily energy generations, we develop a neural network-based trace and prediction model. Due to the time lagging effect in forecasting, the average prediction error for the next hours or days reaches 27.6%. The data stream is still being accumulated and the accuracy will be enhanced by more intensive machine learning.