Abstract:

The power system operation becomes more demanding with the growing integration of renewable energy sources. A lot of small generation units (e.g. solar and wind power plants) work without operators on-site, i.e., only remotely monitored. Modern control equipment enables the safe operation of the plant while providing a huge amount of mainly structured, but also unstructured data for remote or local supervision. University North has the laboratory for renewable energy sources (RES) at the Department of electrical engineering that allows teaching staff and students to study renewable energy on real or near-real time data and also historical data. The data collected is analysed for the purpose of finding regularities and patterns, all for the purpose of detecting and predicting future events. Predicting future events and outcomes is essentially machine learning, an approach that provides the systems the ability to automatically learn and improve from experience without being explicitly programmed.

The purpose of the paper is to analyse and consequently use the historical data obtained from the RES to predict behaviour and the potential of renewable energy in the wider University North area. Different machine learning classification and regression methods, as well as validation methods, will be compared and implemented in Waikato Environment for Knowledge Analysis (WEKA). Special emphasis will be given to data visualization, a technique of creating visual representations that can significantly facilitate the processes of memory and analysis, and enable students to reach a conclusion or answer to a specific question as quickly as possible.

Keywords:

Classification, data visualization, machine learning, renewable energy sources, solar plant, validation methods.