Photovoltaic Performance Assessment in Different Weather Conditions Utilizing an Artificial Neural Network Ensemble

The adoption of Renewable Energy Sources (RESs), particularly solar Photovoltaic (PV) systems, is essential for sustainable energy systems. To enhance PV system performance, precise and efficient simulation tools are crucial. Conventional simulation methods often involve complex computations and lengthy processes. This research study presents an alternative approach to solar energy modeling by exploiting Artificial Neural Networks (ANNs) in order to improve the accuracy and reliability of simulations of photovoltaic systems. The investigation is carried out in two phases. First, six separate ANNs were constructed each dedicated to predicting the hourly electrical output for one of six different PV modules. and total horizontal solar radiation of 24 locations were entered as features, while the corresponding hourly output of a 5-parameter photovoltaic model served as output. ANNs were trained with varying numbers of hidden layer neurons, and the optimal number of neurons was determined based on common accuracy metrics. In the second, each ANN was subjected to validation considering 24 other locations, generating the corresponding hourly electric power forecasts. A comparative analysis was then conducted both in terms of accuracy metrics evaluated on an hourly basis and errors on the yearly energy produced. Optimal ANNs demonstrate high R-square values along with low root mean square error and mean absolute error values. Across different global locations and PV modules, accuracy diminishes but remains relatively elevated in regions with extreme temperature or solar radiation values.