Assessment of Online Weather Platform Data for PV Output Power Forecasting

Abstract

Forecasting is an essential element in supporting the integration of large-scale Grid-Connected Photovoltaic System (GCPV). However, meteorological data vital for accurate forecasting may not be available for a specific location of interest. This paper investigates the use of data obtained from online weather platforms in forecasting photovoltaic (PV) output power when site data is missing or unavailable. Three online platforms were compared and one is chosen as the basis of this paper. The analysis is benchmarked using real ground-based measurement installed at the Faculty of Electrical Engineering, Universiti Teknikal Malaysia Melaka (UTeM). For forecasting, the Linear Regression (LR) method is used because the performance is easier to interpret. Comparison has been made between the real, calculated, and forecasted PV output power. Root Mean Squared Error and coefficient of determination are the statistical metrics used to evaluate the forecasting performance. The results show that the chosen online weather platform has the capability to complement missing or unavailable local weather parameters. On top of that, forecasting using a simple machine learning algorithm provides marginal improvement compared to calculated values with RMSE of 5.95% and 7.62% respectively.