Day-Ahead Prediction of PV Power Output: A One-Year Case Study at Changwon in South Korea

Abstract

This paper is a one-year case study of day-ahead prediction of PV output at Changwon in South Korea. We are focused on day-ahead hourly PV power forecasting and long-term experiments in this paper. We introduce three machine learning based forecasting methods that predict hourly PV power for the next day at midnight, and show performance of them for a 51 kW PV system located at Changwon for a year. Our methods learn relationship of historical meteorological factors, and then predict 24 h PV power considering the trained relationship and weather forecasts from weather forecasting organizations. We show monthly performance of all the proposed methods and a persistence model for a year. Since South Korea is located in a temperate zone with four distinct seasons, and has complex climate characteristics, it is difficult to show actual performance of PV forecasting methods by short-term experimental results. We believe that long term experimental results in this paper are valuable data for the next studies.