Artificial neural network optimisation for monthly average daily global solar radiation prediction. (15th July 2016)

• Record Type: Journal Article
• Authors: Alsina, Emanuel Federico; Bortolini, Marco; Gamberi, Mauro; Regattieri, Alberto
• Is Part Of: Energy conversion and management. Volume 120(2016)
• Page Start: 320
• Page End: 329
• Publication Date: 2016-07-15
• Subjects: Artificial neural networks -- Solar prediction -- Monthly average daily global solar radiation; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2016.04.101
• Languages: English
• ISSNs: 0196-8904

Abstract

Graphical abstract: Highlights: Prediction of the monthly average daily global solar radiation over Italy. Multi-location Artificial Neural Network (ANN) model: 45 locations considered. Optimal ANN configuration with 7 input climatologic/geographical parameters. Statistical indicators: MAPE, NRMSE, MPBE. Abstract: The availability of reliable climatologic data is essential for multiple purposes in a wide set of anthropic activities and operative sectors. Frequently direct measures present spatial and temporal lacks so that predictive approaches become of interest. This paper focuses on the prediction of the Monthly Average Daily Global Solar Radiation (MADGSR) over Italy using Artificial Neural Networks (ANNs). Data from 45 locations compose the multi-location ANN training and testing sets. For each location, 13 input parameters are considered, including the geographical coordinates and the monthly values for the most frequently adopted climatologic parameters. A subset of 17 locations is used for ANN training, while the testing step is against data from the remaining 28 locations. Furthermore, the Automatic Relevance Determination method (ARD) is used to point out the most relevant input for the accurate MADGSR prediction. The ANN best configuration includes 7 parameters, only, i.e. Top of Atmosphere (TOA) radiation, day length, number of rainy days and average rainfall, latitude and altitude. The correlation performances, expressed through statistical indicators as the Mean Absolute Percentage Error (MAPE), range between 1.67% and 4.25%, depending on the number and type of the chosen input, representing a good solution compared to the current standards.