24-Hour ahead PV power forecasting based on the univariate hybrid machine learning model

Abstract

The forecasting of photovoltaic power is very crucial for the integration of photovoltaic plants into the grid. This paper developed a hybrid model that combines an Artificial Neural Network with Wavelet Transform for direct PV power forecasting. Firstly, the wavelet transformation is employed to decompose, usually dishonest PV power time series data into a well-formed constituent data set series. Secondly, these constitutive data series are utilized for training the Artificial Neural Network using a back-propagation algorithm to forecast photovoltaic power's future values. The data used for this study is collected from a 100 kW rooftop photovoltaic power plant, which is connected to the grid located at Ghaziabad, India. The average root mean square error, mean absolute percentage error, and symmetric mean absolute percentage error of the proposed model are 0.035, 6.75%, and 9.95%, respectively. The forecasting results show that the proposed model significantly improves forecasting performance with less computational cost.