Prevision model for energy production in solar concentrator using Artificial Neural Network

During the last few decades, technology and energy demands are playing a greater role in modern society. So far, most of the energy production comes from oil, natural gas, and coal, all of which are depleting in nature. Renewable energy is crucial for the future of both humanity and the world’s ecosystems. Amongst these, solar energy (SE) is the one that has been the most widely adopted and received the most financial investment. SE is produced either by Photovoltaic (PV) or Solar Concentrator Power (CSP); the first system is the most prevalent, thanks to a well-established technology and a drastic reduction of costs allowing mass production. CSP, however, requires greater funding and area. Despite these factors, CSP is getting more and more attention because of Thermal Energy Storage (TES). This reservoir allows CSP to work during the night and during days where there is little sunlight. There are two main ways to increase the efficiency of these systems: by developing and implementing new materials and/or by studying the operating conditions for optimizing and predicting performances. In recent years, Artificial Neural Network and Artificial Intelligence have become essential for this purpose. Their capability of identifying “hidden” correlations between Inputs and Outputs, outperforms the former and more complex methods for solving these analyses. The aim of this thesis is to find an ANN coupled with a Solar Parabolic Dish. The Net belongs to the feedforward class (Multilayer perceptron) and exploits atmospheric conditions as Inputs, and the temperature of CS as Output in order to predict energy production. In this work, learning algorithms: Levenberg-Marquardt, Bayesian regularization, Resilient Backpropagation and Scaled Conjugate Gradient are analyzed. The selected evaluation criteria are: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Mean Bias Error (MBE) and R-squared (R2). The reliability of each model can be found in the Conclusion section.