Very Short-Term PV Power Prediction Using Machine Learning Models

Abstract

Due to the intermittency of solar photovoltaic (PV) power and fast fluctuations in the PV output power, very short-term PV power prediction is of paramount importance for efficient control of resources and units such as loads and energy storage systems and market regulation. As PV power is volatile and highly nonlinear, data-driven machine learning models are developed to predict PV power for a very short-term horizon. In this study, 10 previous samples (i.e., 50 minutes of data) are used as features to predict PV power for the current time and 5 next time periods (i.e., 25 minutes). Four machine learning techniques including Linear Regression (LR), Random Forest Regression (RFR), Multi-Layer Perceptron (MLP) neural network, and long short term memory (LSTM) are utilized in this study. Metrics including the coefficient of determination (R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) have been used to evaluate the performance of the developed machine learning models. Simulation results on a one-year dataset with a sampling resolution of five minutes indicate that the prediction accuracy of the proposed tuned machine learning methods is high and acceptable. The optimized RFR is found to be the best method in terms of computational performance and accuracy.