دانلود رایگان ترجمه مقاله برآورد خلوص تری اتیلن گلیکول (TEG) در واحدهای کم آبی گاز طبیعی (نشریه الزویر ۲۰۱۴)

تخمین خلوص تری اتیلن گلیکول یا TEG در واحدهای دهیدراسیون گاز طبیعی با استفاده از شبکه عصبی فازی

Estimation of triethylene glycol (TEG) purity in natural gas dehydration units using fuzzy neural network

Abstract

Natural gas usually contains a large amount of water and is fully saturated during production operations. In natural gas dehydration units’ water vapor is removed from natural gas streams to meet sales specifications or other downstream gas processing requirements. Many methods and principles have been developed in the natural gas dehydration process for gaining high level of triethylene glycol (TEG) purity. Among them, reducing the pressure in the reboiler at a constant temperature results in higher glycol purity. The main objective of this communication is the development of an intelligent model based on the well-proven standard feed-forward back-propagation neural network for accurate prediction of TEG purity based on operating conditions of reboiler. Capability of the presented neural-based model in estimating the TEG purity is evaluated by employing several statistical parameters. It was found that the proposed smart technique reproduces the reported data in the literature with average absolute deviation percent being around 0.30%.

۱ Introduction

Generally, large amount of water is accompanied by natural gas (NG) in the reservoir. Because of this, the produced NG is completely saturated or at the water dew point. With the aim of meeting sales specifications or other downstream gas processes like gas liquid recovery, gas dehydration operation is employed in NG industry to remove the water vapor (Bahadori, 2009a; Bahadori, 2009b; Bahadori et al., 2008). From economic and safety points of view, the water moisture content of NG must be maintained below a certain threshold (Bahadori, 2007; Bahadori and Vuthaluru, 2009a,b; Gharagheizi et al., 2013; Ghiasi, 2012; Ghiasi and Mohammadi, 2013; Masoudi et al., 2005; Mohammadi et al., 2005). In cases where dew point depressions should be of the order of 15
to 49
C, glycols are commonly used (Lubenau and Mothes, 2009). Amongst different glycols such as diethylene glycol (DEG), triethylene glycol (TEG), and tetraethylene glycol (TREG), that are used as liquid desiccants, the most common choice for NG dehydration is TEG (Piemonte et al., 2012). Operation and maintain of liquid desiccant dehydration equipment is simple (Gironi et al., 2010; Nivargi et al., 2005). This type of dehydration could be easily automated for unattended operation; glycol dehydration at a remote production well is such an example (Bahadori, 2009b; GPSA, 2004).

Fig. 1 shows the gas stripping section in NG dehydration unit. It is well-known that pressure reduction in the reboiler (reconcentrator) at a constant temperature contributes to higher glycol purity. Operating range of most reconcentrators is between 1.7 and 5.2 kPa of pressure (Stewart and Arnold, 2011; Wichert and Wichert, 2004). On standard atmospheric reconcentrators, pressures more than 7 kPa could lead to glycol loss from the still column and reduction of both lean glycol concentration and dehydration efficiency (Bahadori et al., 2014). Furthermore, pressures more than 7 kPa are commonly associated with excess water in the glycol. Consequently, a vapor velocity exiting the still great enough to sweep glycol out will be created (Stewart and Arnold, 2011). On the other hand, pressures less than atmospheric are responsible for increase in the concentration of lean glycol. This is a consequence of decrease in boiling temperature of rich glycol/water mixture (Stewart and Arnold, 2011).

As a result of aforementioned issues, it is crucial to calculate TEG purity as a function of reconcentrator pressure and temperature. The main objective of the presented study is evolving a neural-based model for accurate prediction of TEG purity on the foundation of multilayer perceptron (MLP) artificial neural network (ANN) as a well-proven algorithm of machine learning approach. To the best of author’s knowledge, there is no published work on the subject of TEG purity modeling versus reboiler temperature at various levels of pressure in gas dehydration systems by means of MLP neural networks. Overview of ANNs, computational procedure to develop a proper model, and results of the presented model constitute the rest of the manuscript. The last part concludes this communication.