Machine learning techniques (MLTs) can create accurate predictions of solar outputs that are used in photovoltaic system performance analysis. The issue with MLT application is the requirement for large amounts of historical data for training the prediction models that is not always available. Since the photovoltaic system behaviour is non‐linear due to the unpredictable nature of the weather conditions throughout the year, MLT training requires annual historical data to create the prediction model.
The photovoltaic system production meters only store up to 3 months of historical power values. This information served as motivation to research different types of MLTs, in search of one that would accurately predict the daily solar energy values of a photovoltaic system based on the available 3‐month historical data. The aim of this work is to implement a photovoltaic system performance analyser that estimates daily solar alternating current energy outputs for any rooftop photovoltaic system, based on daily solar irradiation values, without being influenced by the seasons of the year nor the photovoltaic system installation location. Therefore, 5 MLTs were studied and compared, which include the regression tree, artificial neural network, multigene genetic programming, Gaussian process, and the support vector machine for regression. Results show that the regression tree MLT provides acceptable results to be used in all locations and all seasons of the year, while the support vector machine for regression is best for spring and summer training dataset months, and the Gaussian process is best for the autumn and winter training dataset months.