Real time photovoltaic power forecasting and modelling using machine learning techniques

Abstract

Photovoltaic (PV) system installations have increased in recent years partly due to growing energy needs from a rising population. Such PV systems producing electricity contribute in promoting green energy. However, solar energy is highly intermittent and uncontrollable due to its high spatial and temporal variations of atmospheric conditions. With such variability, PV power forecasting is therefore crucial for full integration of solar energy into the grid. In this study, Support Vector Regression (SVR) and Random Forest Regression (RFR) models were built and used to forecast real-time PV power output of a 1.5kW solar PV system installed at the Department of Physics, University of Nairobi in Kenya. SVR model outperforms RFR model with root mean square (RMSE) of 43.16 adjusted R2 of 0.97 and mean absolute error (MAE) of 32.57 on the validation. Dataset compared to RMSE of 86, adjusted R2 of 0.90, MAE of 69 were obtained for RFR model. A real time power forecast application based on the SVR model was successfully built using the Shiny application in R software. This shows that SVR model is more robust than RFR and has capabilities of reducing errors during computations.