Search documents

Browse topics

Document details

A More Accurate Machine Learning Photovoltaic System Performance Analyser by Using Fuzzy Logic
S. Rodrigues, J.P. Carvalho, H. Geirinhas Ramos, F. Morgado-Dias
PV System Performance, Artificial Intelligence, Machine Learning, Takagi-Sugeno Fuzzy Logic Inference System
PV Systems - Performance, Applications and Integration
Subtopic: Operation, Performance and Maintenance of PV Systems
Event: 35th European Photovoltaic Solar Energy Conference and Exhibition
Session: 6BO.8.3
1594 - 1604
ISBN: 3-936338-50-7
Paper DOI: 10.4229/35thEUPVSEC20182018-6BO.8.3
0,00 EUR
Document(s): paper


Machine Learning Techniques (MLTs) deal well with the non-linear behaviour of the Photovoltaic systems and therefore are ideal to be used in Photovoltaic System Performance and Fault Analysis. The annual historical data of a roof-top Photovoltaic system located in Portugal in years 2015 and 2016 were used to train and test five different MLT prediction models such as the Regression Trees, Artificial Neural Networks, Multi-Gene Genetic Programming, Gaussian Process and Support Vector Machines for Regression. The aim of this work is to determine which MLT provides the most accurate prediction output results of the daily AC solar energy based on the solar irradiation inputs. An experimental setup including three experiments and 12 different scenarios was designed to achieve the aim of this work. Each scenario is associated to a season of the year in order to break down the complexity of the prediction model training. The first and second experiments simply compare various MLTs, but the third experiment describes a novel methodology that provides very accurate MLT estimate results, by means of a Hybrid Machine Learning and Computational Intelligence System that combines a Takagi-Sugeno type Fuzzy Logic Inference System with the various MLTs.