Abstract
This article reviews the applications of artificial neural networks (ANNs) in greenhouse technology, and also presents how this type of model can be developed in the coming years by adapting to new technologies such as the internet of things (IoT) and machine learning (ML). Almost all the analyzed works use the feedforward architecture, while the recurrent and hybrid networks are little exploited in the various tasks of the greenhouses. Throughout the document, different network training techniques are presented, where the feasibility of using optimization models for the learning process is exposed. The advantages and disadvantages of neural networks (NNs) are observed in the different applications in greenhouses, from microclimate prediction, energy expenditure, to more specific tasks such as the control of carbon dioxide. The most important findings in this work can be used as guidelines for developers of smart protected agriculture technology, in which systems involve technologies 4.0.
Highlights
Greenhouses are systems that protect crops from factors that can cause them damage
Another way to delimit their number is to apply the sensitivity analysis to the different input variables to determine which ones are more relevant to the variables whose behavior is to be determined, just as did Seginer et al [125], who found that solar radiation and outdoor air temperature are the factors that have the greatest impact on the temperature and internal humidity of the greenhouse
The use of the combination of artificial neural networks (ANNs) with mathematical models has been little explored, as can be seen in Yousefi et al [218] and Linker et al [205], the approach can be considered from two perspectives: First, using techniques such as fuzzy logic for optimization in the random choice of the initial parameters and second, to use the physical models for the generation of synthetic data that help the network in the learning process, minimizing the errors due to the lack of information that a base can present of data in situ
Summary
Greenhouses are systems that protect crops from factors that can cause them damage. They consist of a closed structure with a cover of translucent material. For the environmental control of greenhouses, conventional proportional, integral and derivative controllers (PID) are mainly developed due to their flexibility, architecture and good performance [38] Another topic of interest derived from the production of greenhouse crops is energetic consumption, in which solar energy is presented as a viable substitute for traditional sources (fuel and electricity). Presents trends for future research in the development of this type of model will improve its application and integration with the 4.0 technologies that are currently applied in smart agriculture (SA) but are little used in greenhouse production such as the internet of things (IoT), machine learning (ML), image analysis, big data, among others. Section in 6 addresses challenges in the development of NNs in other studies and ANNs in greenhouse applications.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call