We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
RED NEURONAL PARA CONTROL DE COMBUSTIÓN EN UNIDADES GENERADORAS DE VAPOR.
- Authors
Meléan, Fernando; Ordoñez S., Bárbara A.
- Abstract
The objective of this investigation is to propose a neural network for combustion control in steam generating units. The research is supported by Lammers, H., Lammers, T. and Woodruff, E. (2004) in reference to combustion control systems, and by Aggarwal, C. (2018), Jurado, F. et al. (2012) and Berzal, F. (2018) when related to artificial neural networks. Type of research was considered projective, documentary and descriptive, with a non-experimental, field and transactional design. The analysis unit is made up of steam generating units used for thermal stimulation of extra-heavy crude oil wells, belonging to an oil services company that operates in the Lagunillas and Bachaquero fields in the Zulia region. Documentary reviews, unstructured interviews and analysis matrices were used as data collecting techniques. Development was comprised of five (05) design phases with a mixed methodology based on: (1) the objectives proposed for the investigation and (2) the methodology for the development of artificial neural network models proposed by Mariño, S. and Primorac, C. (2016). A new process diagram with control based on artificial neural networks was proposed. As a result of a comparative analysis, it was possible to present a multilayer artificial neural network model, with one (01) input layer, made up of 3 neurons; five (05) hidden layers of 10 neurons each implementing ReLU activation functions, and two (02) output layers with linear activation functions. Network training was carried out through the backpropagation algorithm implementing the Adam optimizer. The implementation tests performed on the modeled neural network validated its ability to predict the output parameters of interest for combustion control with very good precision.
- Publication
Revista Télématique, 2021, Vol 20, Issue 2, p43
- ISSN
1856-4194
- Publication type
Article