Artigo Acesso aberto Revisado por pares

Machine learning for solar trackers

2019; American Institute of Physics; Volume: 2126; Linguagem: Inglês

10.1063/1.5117524

ISSN

1935-0465

Autores

Jose A. Carballo, Javier Bonilla, Manuel Berenguel, Jesús Fernández‐Reche, Gines Martinez Garcia,

Tópico(s)

Infrared Target Detection Methodologies

Resumo

A new approach for solar tracking, based on deep learning techniques, is being studied and tested using Tensorflow, an open source machine learning framework. Tensorflow makes the implementation more flexible and increases the development capabilities. Tensorflow facilitates the neural network implementation on several kinds of devices (embedded and mobile devices, mini computers, etc.). Furthermore, Tensorflow supports different types of neural networks which can be tuned and retrained for particular purposes. The presented results are promising, since the retrained networks correctly identify the Sun and the target, with this information the system can be controlled to properly track the Sun's apparent trajectory without further information.

Referência(s)
Altmetric
PlumX