Abstract

Nowadays, due to the increasing integration of solar energy into electrical systems, it is necessary to improve methodologies used to analyze this type of technology. In this regard, a topic that has had permanent attention is solar irradiation forecast and photovoltaic (PV) plant energy production estimation. This study analyzes the performance of three methodologies commonly used in PV plant energy production estimation: Multiple Linear Regression (MLR), Artificial Neural Networks (ANNs), and Support Vector Machines (SVMs). Calculations lead to the conclusion that SVMs usually produce lower training errors. However, the CPU used by this type of algorithm is significantly higher than those required by ANN and MLR. Considering this, an MLR-based algorithm was developed to deliver training errors similar to those delivered by ANNs and SVMs, but with significantly less processing time and use of computing resources. The algorithm proposed (F-MLRsolar2) is at least 1000000 times faster than SVM and 5000 times faster than ANN. Additionally, the proposed algorithm allows obtaining a simple equation that can be used analytically.