A novel methodology based on artificial neural networks is proposed as an alternative to algebraic and numerical procedures to determine the I-V curve of a module under different conditions. Although there are methods that use neural networks for approximating the I-V curve, this is the first time that the measurement of the spectrum is incorporated as an input. In addition, a suitable selection of the training samples used to build the model is fundamental in order to get an accurate approximation. This is why a training sample selection based on a Kohonen self-organizing map is performed in this paper instead of a random selection. With the use of this preliminary step, the performance of the network trained with spectral information improves over the one without spectral information. Copyright © 2012 John Wiley & Sons, Ltd. A novel methodology based on artificial neural networks is proposed as an alternative to algebraic and numerical procedures to determine the I-V curve of a module under different conditions. It is the first time that the spectrum is incorporated as an input, and a selection of the training samples based on a Kohonen self-organizing map has been performed. With the use of this preliminary step, the performance of the network trained with spectral information improves over the previous models. Copyright © 2012 John Wiley & Sons, Ltd.
Photovoltaic module simulation by neural networks using solar spectral distribution
Piliougine M.;
2013
Abstract
A novel methodology based on artificial neural networks is proposed as an alternative to algebraic and numerical procedures to determine the I-V curve of a module under different conditions. Although there are methods that use neural networks for approximating the I-V curve, this is the first time that the measurement of the spectrum is incorporated as an input. In addition, a suitable selection of the training samples used to build the model is fundamental in order to get an accurate approximation. This is why a training sample selection based on a Kohonen self-organizing map is performed in this paper instead of a random selection. With the use of this preliminary step, the performance of the network trained with spectral information improves over the one without spectral information. Copyright © 2012 John Wiley & Sons, Ltd. A novel methodology based on artificial neural networks is proposed as an alternative to algebraic and numerical procedures to determine the I-V curve of a module under different conditions. It is the first time that the spectrum is incorporated as an input, and a selection of the training samples based on a Kohonen self-organizing map has been performed. With the use of this preliminary step, the performance of the network trained with spectral information improves over the previous models. Copyright © 2012 John Wiley & Sons, Ltd.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.