|تعداد مشاهده مقاله||26,002,551|
|تعداد دریافت فایل اصل مقاله||10,693,345|
Modeling and Optimization of Different Acid Gas Enrichment Structures via Coupling of Response Surface Method with Genetic Algorithm
|Gas Processing Journal|
|دوره 9، شماره 1، شهریور 2021، صفحه 157-172 اصل مقاله (1.5 M)|
|نوع مقاله: Research Article|
|شناسه دیجیتال (DOI): 10.22108/gpj.2022.132984.1117|
|Maryam Pahlavan؛ Akbar Shahsavand* ؛ Mehdi Panahi|
|Chemical Engineering Department, Faculty of Engineering Ferdowsi University of Mashhad, Mashhad, Iran|
|Lean acid gases entering any sulfur recovery unit (SRU) can strongly damage the corresponding overall sulfur efficiency. The use of the acid gas enrichment (AGE) process is essential to increase recovery efficiency. Two novel scenarios are studied in the present work. The first low-pressure structure uses an enrichment tower with operating pressure above the atmosphere and lower than the regenerator pressure, while, the second high-pressure scenario limits the enrichment tower pressure between the amine flash drum and corresponding regenerator pressures. The combination of the Aspen-HYSYS process simulator and response surface method is successfully employed to generate training data and create reliable hyper-surfaces for mimicking the acid gas enrichment rate versus various operating parameters. An initial sensitivity analysis is recruited to pinpoint the most dominant input parameters. The optimization of fitted merit functions was carried out using our in-house genetic algorithm code. The corresponding enrichments for high and low-pressure scenarios were 83.63 and 70.53% respectively. These are more than 140% and 105% increases in H2S concentrations in comparison to the conventional design of the existing SRUs, which is based on around 34% H2S content in the acid gas feed. Economic and environmental evaluations of both scenarios revealed that the optimal low-pressure structure is much more favorable from the economic point of view, while the high-pressure design performs more environmentally friendly by reducing the SO2 emissions by at least 21.4%. To the best of our knowledge, the above complex and detailed study have never been performed previously for any AGE process.|
|Acid Gas Enrichment؛ SO2 emissions؛ Genetic Algorithm؛ Response Surface Method؛ Neural Network|
Most the natural gases produced from different reservoirs around the world are contaminated with various impurities, such as H2S, CO2, and water vapor. These impurities should be removed during various stages of gas processing, including sweetening and dehumidification processes. Acid gas components (CO2 and H2S) are traditionally removed from the sale gases using conventional amine treating processes. The acid gas produced in the gas treating unit (GTU) will have an unfavorably high CO2 to H2S ratio, when CO2 to H2S ratio is relatively high in a sour natural gas feed, leading to a poor quality feed of sulfur recovery unit (SRU). The acid gas feed stream typically requires at least 50 mole percent H2S, to achieve high temperature (T>926°C, 1700°F) combustion in the SRU Claus furnace. Otherwise, other Claus plant design options should be used instead of the straight thorough scheme such as split-flow design, and/or fire a supplemental fuel, recruit oxygen enrichment of the combustion air, or preheat the air and/or acid gas feed to the Claus furnace to maintain the highest attainable temperature (ZareNezhad & Hosseinpour, 2008) (Barvar, Safamirzaei, & Shariati, 2019). Acid Gas Enrichment (AGE) is widely used in the last three decades to process dilute acid gas streams. The process objective is to maximize CO2 slip and minimize the H2S leak into vent gas from the system, thereby producing a gas enriched in H2S to the greatest possible extent (Weiland, 2008).
AGE process is usually applied ahead of SRU to produce a richer acid gas stream. Its appropriate use can stabilize thermal section operation and achieve a higher temperature in the reaction furnace. Higher temperatures ensure partial or complete destruction of acid gas contaminants such as benzene, toluene, ethyl benzene, and xylene (BTEX components) (Garmroodi Asil, Shahsavand, & Mirzaei, 2017).
Up to now, different aspects of the acid gas enrichment process are studied in the limited available research articles. Two main approaches are traditionally used. In the first scheme, the research was usually focused on the selection of the optimal or at least more suitable solvent to be used in the separate AGE process, positioned between GTU and SRU, for selective absorption of H2S and rejecting the problematic CO2 from the acid gas stream.
Chilukuri and Bowerbank proposed an integrated lined-ups structure that created a better quality acid gas feed to the SRU. They claimed that such an integrated structure created new cost-effective modes of operation, reduced plant complexity, managed a wide range of feed gas contaminant uncertainties, and reached stringent emission requirements (Chilukuri & Bowerbank, 2016).
Dara et al studied the economic and environmental impact of a cooling system to cool down hot lean MDEA (40% by weight) solvent in an AGE unit located in the United Arab Emirates. It is found that reducing the lean solvent temperature increased the purity of both H2S and CO2 product streams (Dara et al., 2018).
ExxonMobil applied FLEXSORB™ SE solvent for AGE and tail gas clean-up. A novel compact and low-weight processing technology platform was used to perform selective H2S removal (Northrop, Seagraves, Ramkumar, & Cullinane, 2019). Al-Amri and Zahid proposed a new acid gas cleaning system that employed two GTUs instead of a conventional GTU+AGE system for treating high CO2 content natural gases while maintaining all process speciﬁcations such as sweet gas, acid gas, waste gas, and amine unit ﬂashed gas with allowable concentrations. In contrast to conventional systems, acid gas was produced with the required purity in the first GTU, while sweet gas was produced in the second GTU, with the required spec. They claimed that the proposed design required 22% less energy demand compared to the conventional design because of the reduced amine circulation rate, leading to lower re-boiler duties (Al-Amri & Zahid, 2020).
In the Second strategy, necessary modifications are usually applied to an existing GTU configuration while using the conventional solvent (Garmroodi Asil & Shahsavand, 2014a). Palmer proposed four different structures to increase H2S concentration in the acid gas stream as well as to produce a valuable CO2 by-product (Patent No. 20040226441, 2004) (Patent No. 7147691, 2006). Mak et al suggested a few configurations in a patent in which a portion of an isolated hydrogen sulfide stream is introduced into an absorber operating as a carbon dioxide rejecter (Patent No. 7635408, 2009).
Mahdipoor and Dehghani investigated both technical and economic aspects of two AGE configurations. In the first scheme, AGE off-gas was sent to the incinerator, while in the second scenario, it was used as the feed for the tail gas treating unit (TGTU). They reported that routing AGE off-gas to TGTU increases the size and cost of TGTU equipment while reducing the environmental pollutants (Mahdipoor & Dehghani Ashkezari, 2016).
Dai et al investigated a novel two-stage flash process of acid gas removal from the natural gas stream. The ﬁrst ﬂash drum is used to remove most of the light hydrocarbons, while the second one is recruited to remove most of the CO2 and a small fraction of the H2S. Afterward, the liquid absorbent is sent to the regeneration column to remove the absorbed H2S. They concluded that, compared with the traditional acid gas removal and traditional AGE processes, the two-stage ﬂash process of acid gas removal can successfully enrich acid gas while reducing the regenerator energy consumption (Dai, Peng, Qiu, & Liu, 2019).
Rameshni and Santo presented a patent for an integrated system that combined sour gas treating (for H2S Removal), separation of impurities (such as hydrocarbons, BTEX, and mercaptans), and partial acid gas enrichment. The new schemes comprised one or more absorbers coupled with the primary and secondary regenerators. The secondary regenerator function is to further enrich the H2S stream and to separate the hydrocarbons, mercaptans, and BTEX (Patent No. 20200039825, 2020).
As can be seen from the above literature review, research on acid gas enrichment was somehow limited. Furthermore, most of the research is focused on adding a separate enrichment unit placed between GTU and SRU unit, by using its solvent. Evidently, such an independent scheme provides more operational flexibility while leading to excessively high investment costs. When capital investments are limited, the second choice seems more realistic. In this scheme, the conventional GTU process is usually equipped with an enrichment tower to enhance acid gas enrichment. Evidently, such a design is less flexible from an operational view but severely reduces the required capital investments. Research on this approach is quite limited (Patent No. 20040226441, 2004) (Patent No. 7147691, 2006) (Patent No. 7635408, 2009). Due to shortcomings in capital investments in the Iranian natural gas refineries, the second approach will be investigated in the present work.
Garmroodi Asil and Shahsavand simulated three different enrichment schemes for Khangiran natural gas refinery acid gas enrichment. In the first scheme, a fraction of the acid gas from the top of the regenerator is recycled back to the main GTU contactor. In the other two, a fraction of the acid gas stream is recycled back to an added enrichment tower (ET), which was placed between the amine flash drum and the regenerator. In the second scheme, the ET pressure was lower than the regenerator pressure, while in the third scheme the ET pressure was kept between the flash drum and regenerator pressures. They concluded that the SRU feed stream can be enriched from its original value of nearly 34 mol% H2S to about 54 and 70 mol%, by resorting to the 2nd and 3rd schemes, respectively (Garmroodi Asil & Shahsavand, 2014b). Due to the importance of the acid gas enrichment process and the great impact of this process on economic and environmental issues of sulfur recovery units and also due to the relative lack of study in this field, the purpose of this article is to thoroughly investigate structures 2 and 3 and find the optimal structure and optimum values for operational variables.
In the first step, the previous structures were slightly modified and the response surface method (RSM) along with artificial neural networks (ANNs) are used to model the entire process and provide the necessary cost function. Then genetic algorithm (GA) optimization technique is used to pinpoint the optimal operation conditions. Furthermore, the optimal selected scenario was studied from both economic and environmental viewpoints. To the best of our knowledge, such a complex approach has not been addressed, previously.
The present article compares two different schemes of AGE processes for the Khangiran natural gas refinery. An enrichment tower is used between the GTU amine flash drum and the corresponding regenerator column, in either of both designs. The pressure of the enrichment tower is kept below the regenerator pressure in the first scheme (low-pressure scenario), while in the second arrangement it is bounded between the amine flash drum and the regenerator pressures (high-pressure scenario).
To investigate the effect of more operational parameters, the enrichment structures of this study are slightly different from the previous ones (Patent No. 20040226441, 2004) (Patent No. 7147691, 2006) (Garmroodi Asil & Shahsavand, 2014b). For example, two heat exchangers are added to both rich and lean amine streams entering the ET, to investigate those streams' temperatures. Furthermore, a flow splitter has been installed on the rich amine stream to investigate the effect of the rich amine split ratio sent to ET.
In the present work, modeling and optimization of the acid gas enrichment process have been performed by resorting to RSM, ANN, and GA techniques, while in the previous studies, the effect of the operational parameters has been studied by simple trend analysis (i.e. plotting a few diagrams without applying any modeling or optimization method).
Figure (1-a) depicts the original acid gas sweetening unit of the Khangiran natural gas refinery in the absence of any acid gas enrichment. Figures (1-b) and (1-c) show our in-house version of the Palmer design for low and high-pressure AGE schemes, respectively (Patent No. 20040226441, 2004) (Patent No. 7147691, 2006) (Garmroodi Asil & Shahsavand, 2014b). Additional but different unit operations are required for each AGE design. For example, a pump is required to boost the pressure of the liquid stream departing the enrichment tower and receiving by the regeneration column, in the first scheme, while the use of a suitable compressor is essential in the second design to compress the acid gas stream leaving the acid gas splitter and entering the enrichment tower.
Compared to our previous work (Garmroodi Asil & Shahsavand, 2014b) several modifications are used in the present study to enhance the overall enrichment efficiencies in both schemes. For example, heat exchangers are added to adjust the temperatures of lean amine and rich amine feed streams entering the enrichment tower of each scheme. Furthermore, the rich amine splitter is used in both designs to investigate the effect of the rich amine flow rate entering the enrichment tower. Table 1 provides the sour feed gas specifications entering each absorber of the four GTUs. Each GTU includes two parallel absorbers and two strippers.
(b) Low-pressure AGE scheme, (c) High-pressure AGE scheme
Table 1: Khangiran natural gas sweetening sour gas feed specifications entering each absorber of the four parallel gas treating units (GTUs) (Garmroodi Asil & Shahsavand, 2014b)
3.1. Main operational input and output parameters
For a successful investigation of the entire process, the most important operational parameters must be recognized. Our primary investigations revealed that the following six parameters are anticipated to have relatively strong effects on the overall performance of the AGE schemes.
Two obvious output parameters can be recognized as a) H2S mole percent in acid gas leaving the entire enrichment process and entering the sulfur recovery unit (designated as the first response: R1) and b) H2S mole percent in off gas leaving the top of the enrichment tower (response 2: R2). Evidently, R1 indicates the enrichment rate and R2 provides the environmental constraint, which will be used in the optimization algorithm.
Table 2: Operational inputs variables and the corresponding bounds
3.2. Sensitivity analysis
The entire processes depicted in Figures (1b) and (1c) are simulated using Aspen-HYSYS (V11) process simulator. The acid gas - chemical solvents property package has been recruited for this purpose. Five equi-spaced points were considered for each of the six input variables and the corresponding responses R1 and R2 were computed via simulation for that input variable at the corresponding value, while other input variables are kept fixed at their average values, as depicted in Table 2.
Figure 2 shows sensitivity analysis of two AGE process (denoted as HP and LP) responses R1 and R2 versus the scaled input variables across their entire corresponding input domains. All inputs are scaled between (-1) and (+1) for ease of illustration. Table 3 provides the conclusions obtained from Figure 2 about the degree of importance of various input parameters for both AGE scenarios.
Figure 2: Sensitivity analysis of responses R1 and R2 versus various input variables.
Top) High pressure (HP) and Bottom) Low pressure (LP)
Table 3: Importance of various input parameters based on sensitivity analysis of Fig. 2.
Due to the complexity of the AGE process, two different empirical modelings via response surface method (RSM) and MATLAB software neural network toolbox arebeing used to model the entire process shown in Figures (1b) and (1c). Initially, RSM which is a subdivision of Design-Expert software (220.127.116.11) is used to specify the desired operating points in the input domain of Table 2, for both low and high pressure enrichment tower scenarios. Tables A1 and A2 of the appendix section provide the entire 78 operating points and the corresponding values of responses R1 and R2 for both low and high-pressure AGE designs, as predicted via the Aspen-HYSYS simulator for the processes depicted in Figures (1b) and (1c).
After necessary preprocessing and internal optimization, two parallel multiple input single output (MISO) data sets of Tables A1 and A2 are used by the RSM technique to ultimately provide equations (a) to (d) for the functionality of R1 and R2 at high and low-pressure scenarios, respectively. As can be seen, the lean amine temperature (LAT) is omitted from equation (a), which is in accordance with the sensitivity analysis of Table 3.
Figure 3 illustrates the relatively successful recall performances of the RSM fitted hyper-surfaces for R1 and R2 at both high and low pressures using equations (a) to (d). Since the values of response R2 drastically vary in the entire domain of inputs as depicted in Tables A2 and A3, therefore the logarithmic plot is used to emphasize the lower values of R2<0.2. Due to environmental considerations, the amount of H2S in the off-gas leaving the top of the enrichment tower should be less than 2000 ppm (Al-Amri & Zahid, 2020).
Figure 3: Recall performances of response surface method (RSM) for R1 and ln (R2) corresponding to high and low pressure (HP and LP) scenarios
In the second attempt, data sets of Tables A1 and A2 are used to train four separate artificial neural networks (NN) via the neural network fitting toolbox of MATLAB software. After several trials, the best architecture was selected as a single hidden layer feed-forward network with 10 Sigmoid neurons. The back propagation method coupled with the Levenberg-Marquardt optimization technique is used for the training process. Figure 4 depicts similar recall performances of the above 4 mentioned neural networks for responses R1 and R2 at both high and low pressures.
Figure 4: Recall performances of Neural network (NN) model for R1 and ln(R2) corresponding to high and low pressure (HP and LP) scenarios.
In practice, the recall performances can be mischievous. To ensure that the above-fitted hyper-surfaces perform adequately in the generalization phase, the corresponding performance should be evaluated over the entire domain of the inputs. Figures 5 and 6 compare various three-dimensional generalizations performances of RSM and MATLAB-NN fitted surfaces for both responses R1 and R2 at high and low-pressure scenarios. Table 3 is used to select the most important input parameters as the abscissa of all 3D plots. The other input parameters are kept fixed at the corresponding average values.
As can be seen in Figures 3 and 4, both RSM and MATLAB-NN models perform quite similarly in their recall performances. However, they perform differently in their generalization performances, especially when mimicking the response R2, as depicted in Figures 5 and 6. For both R1 and R2 responses, the RSM predictions are smoother compared to the relatively oscillatory behavior of MATLAB-NN.
Since, the differences in the generalization performances of the above models, as depicted in Figure 6, are not very extensive, therefore, both models will be used in the optimization process to pinpoint the best optimal operating conditions to achieve maximum enrichment while satisfying the constraint R2<0.2.
Figure 5: Comparison of 3D generalization performances of MATLAB-NN and RSM models for responses R1 and R2 at low pressure ET, over the entire domain of the corresponding inputs.
Figure 6: Comparison of 3D generalization performances of MATLAB-NN and RSM models for responses R1 and R2 at high-pressure ET, over the entire domain of the corresponding inputs.
The fitted hyper-surfaces of the above section are used as the merit functions of the optimization process to find the optimal operating conditions. The genetic algorithm (GA) optimization method, which is based on the survival of the fitness, is used. It uses three main operations of selection, crossover, and mutation to produce new stronger generations from the old population. An in-house computer code was developed based on our previously published article (Lotfikatooli & Shahsavand, 2017a). Although GA is a powerful goal-oriented optimization method, however, it usually suffers from lack of a reliable termination criteria. Traditionally, some pre-speciﬁed maximum number of generations is usually used as the termination criterion. In many practical applications, the stopping criteria can signiﬁcantly inﬂuence both the ﬁnal optimal solution and the overall execution time of the entire optimization process. A modiﬁed version of GA is presented and successfully applied in our previous works (Lotfikatooli & Shahsavand, 2017a) (Lotfikatooli & Shahsavand, 2017b), which uses a novel termination criterion parameter named ANDI (approximate number of decisive iterations). The Constrained Genetic Algorithm Optimization (CGAO) coupled with the previously devised ANDI termination criterion which is used in the present work is slightly different from the main algorithm used in our earlier article (Lotfikatooli & Shahsavand, 2017b). It is modified to accommodate for the constraint(s) and solve the constrained optimization problems. Figure A1, in the appendix depicts the complete flowchart for the above-constrained optimization process, which recruits ANDI for its termination criteria.
Table 4 presents the reported optimization results employing our in-house GA when RSM and MATLAB-NN fitted hyper-surfaces are used as the objective function for R1 and constraint for R2 at both high and low-pressure scenarios. To ensure the optimal performances of the acid gas enrichment at the above optimal points, all R1 and R2 values are re-computed and checked via the Aspen-HYSYS simulations.
Table 4: Comparison of GA optimization results obtained for both low and high-pressure scenarios via RSM and NN models with Aspen-HYSYS simulations, using the conventional constraint on R2.
Table 4 clearly illustrates that for all cases, the model predictions at the optimal points for R1 are relatively close to the values obtained from Aspen-HYSYS simulations. On the other hand, both RSM and MATLAB-NN considerably over-predict the values of R2, when compared to the more reliable values generated via Aspen-HYSYS simulations at the prevailing optimal operating conditions.
Based on the optimal results presented in Table 4, it is evident that the MATLAB-NN provides better enrichment in both low and high-pressure scenarios. RSM fails because it has less flexibility while searching in the R2 range.
Since both models largely over-predict the values of R2, therefore it was decided to increase the constraint value (R2) from the actual value of 0.2 mole% (2000 ppm) to a proper value for each case. Table 5 illustrates the new optimal results when the appropriate constraint value was selected via a trial and error procedure. In each case, the Aspen-HYSYS simulation result was computed and checked to ensure that the environmental constraint (R2<0.2) is satisfied. As can be seen, the RSM model provides higher acid gas enrichment values at the corresponding optimal points for both low and high pressures. Both values are considerably better than the optimal enrichment rates obtained from Table 4.
Table 5: Similar comparisons as to Table 4 but employing a much higher constraint on R2, due to over-predictions of both RSM and NN models for R2 values.
Table 6, summarizes the improvement achieved in the present work via different approaches compared to our previous study (Garmroodi Asil & Shahsavand, 2014b) and the actual design enrichment data collected from the existing AGE unit, based on the H2S mole percent (@ 34 mol%) in the acid gas entering the sulfur recovery unit in the absence of AGE unit. The above improvements are due to both structural modification (such as more flexible lean and rich amine temperatures) and using more advanced modeling and optimization techniques (RSM, Matlab-NN, and GA).
Table 6: Improvements achieved in the present study compared to our previous work (Garmroodi Asil & Shahsavand, 2014b) and actual design data based on original acid gas H2S mole%, in the absence of AGE unit.
Figure 1 illustrates that the backbone of both high and low-pressure structures are essentially the same. However, some different equipment is used in each scenario. For example, the expensive compressor in high-pressure scenario is substituted with a more affordable pump in a low-pressure scenario. The cost estimates for different equipment are performed separately to compare them from an economic standpoint, using the cost estimation correlations of Table 7, as presented in reference to the year 2000 (Seider, Seader, & Lewin, 2018). The required specifications of the corresponding equipment involved in economic evaluations (such as height and diameter of ET, powers of the pump, and compressor) have been obtained by resorting to the Aspen-HYSYS process simulator and reported in Table 8. Table 9 presents the overall cost estimation of different equipment used in high and low-pressure enrichment structures, corrected for June 2021. The Chemical Engineering Plant Cost Index (CEPCI) in June 2021 was around 717.6, which was 21.4% greater than the corresponding value in June 2020 (591.1). As can be seen, the equipment cost for those parts which are different in the two scenarios is around 7 times more for the high-pressure structure, compared to the low-pressure design, most of which is due to the cost of acid gas compressor.
Table 7- Cost estimation formulas used in the economic evaluation of various scenarios (Seider et al., 2018)
Table 8- Parameters used in economic calculations (Seider et al., 2018) (Walas, 1998)
Table 9: Equipment costs (unit:1000$)
Since the low-pressure scenario pump requires much less electrical power consumption than the compressor of the high-pressure structure, therefore the low-pressure design would be more realistic from a merely economic standpoint, regarding both capital investment and operating costs.
To investigate the effect of acid gas enrichment scenarios on the overall performances of four existing Khangiran sulfur recovery units, the entire combination of a typical gas treating unit (GTU) and corresponding sulfur recovery unit (SRU) is simulated using the SULSIM package of the Aspen-HYSYS process simulator for both enrichment scenarios. Figure 7 shows the corresponding simulated flow diagrams.
During these simulations, the main semphasis was on maximizing the hydrogen sulfide concentration in the acid gas stream entering SRU, which leads to maximum elemental sulfur production rate, especially in the reaction furnace of the thermal section. Enriching the acid gas entering the sulfur recovery unit has several other benefits as well, such as reducing the Benzene-Toluene-Ethyl Benzene-Xylene (BTEX) content of the acid gas stream, due to flash vaporization of BTEX components in the enrichment tower, especially when dealing with the low-pressure scenario. Furthermore, the enriched acid gas creates higher combustion chamber temperatures leading to better destruction of BTEX components and providing much-improved catalyst performance due to less BTEX deposition on the catalysts of various SRU catalytic beds.
The current Aspen-HYSYS simulator performs very adequately in the simulation of both GTU and AGE processes, while it has some difficulties in simulating the SRU reaction furnace, due to complete combustion of BTEX components, even at relatively low temperatures of around 800°C. In light of the above discussion, it is anticipated that the actual improvement due to the use of the AGE process would be practically higher than the predictions provided via the Aspen-HYSYS simulator.
The sour gas feed data depicted in Table 1 was used to simulate the original GTU. The rich amine stream collected from the amine flash drum is then used as the feed for both high and low-pressure scenarios of Figures (7a) and (7b). Table 10 provides the acid gas stream specifications outgoing from the three scenarios.
Table 10: Acid gas specifications entering typical Khangiran SRU for various scenarios.
Figure 7: Aspen-HYSYS simulation schemes for: a) Low pressure enrichment scenario, b) High pressure enrichment structure and c) Typical sulfur recovery unit
Table 11 compares the overall environmental benefits of various optimal enrichment scenarios with the original design conditions, due to less sulfur dioxide emission into the adjacent atmosphere. In the original design, the furnace temperature is usually lower than the threshold required for complete combustion of BTEX components, therefore both conditions are considered. When using AGE in either of the two optimal scenarios, the furnace temperature is sufficiently high to completely burn BTEX. Furthermore, small differences in the moles of H2S entering the SRUs for various scenarios are due to differences in the lean loading amine streams, collected from the bottoms of the corresponding regenerators.
Table 11: Comparison of sulfur recovery and sulfur dioxide emissions of different scenarios for a typical SRU of Khangiran natural gas refinery.
*Assuming no BTEX combustion (actual data)
** Simulation result (actual data (Dahiya & Myllyvirta, 2019))
*** Based on the entire capacities of four SRUs.
As can be seen, the more expensive high-pressure scenario can at least reduce the sulfur dioxide discharge by about 21%, while the much less low-pressure scenario can successfully reduce the sulfur dioxide emission for the entire refinery by 17.8%. A more detailed study considering other benefits of the acid gas enrichment on the overall performance of the sulfur recovery unit will result in a much higher environmental protection efficiency, due to less discharge of BTEX components into the surrounding atmosphere, among many others (Ibrahim, Jagannath, & Raj, 2020).
A relatively complex approach is used in the present work for the first time by coupling various powerful tools such as Aspen-HYSYS simulation software, response surface method, artificial neural network, and genetic algorithm optimization technique to select the optimal acid gas enrichment scenario for an Iranian natural gas refinery. Initial sensitivity analysis of the two proposed scenarios for the acid gas enrichment of the Khangiran sour gas treating units revealed that, while the acid gas split ratio is the most influential operating parameter, however, the enrichment tower pressure and the corresponding inlet streams temperatures and flow rates can also appreciably affect the overall H2S enrichment rate.
It was established that the recall performances of both RSM and NN methods are almost identical, while the RSM method provides a more realistic performance on generalization. This issue was validated by resorting to optimization of the corresponding hyper-surfaces using an in-house genetic algorithm (GA) code. The maximum enrichment rates of 83.63 and 70.53 were obtained from a sour feed gas containing 3.4% H2S, by optimizing the RSM model cost function for high and low pressure scenarios. The original design of the existing SRUs receives the acid gas from GTU with 34 mole percent H2S content. Therefore, the optimal enrichment processes show more than 140% and 105% increase compared to the present existing conditions for high and low-pressure scenarios, respectively. The low-pressure scenario, which requires much less capital investment and operating costs seems more realistic from a merely economic perspective. It was also shown that both scenarios at their optimum points can significantly reduce sulfur dioxide emissions.
Table A1- Response values R1 and R2 computed via Aspen-HYSYS software at operating points for low pressure ET scenario as recommended via response surface method (RSM).
Table A2- Response values R1 and R2 computed via Aspen-HYSYS software at operating points for high pressure ET scenario as recommended via response surface method (RSM).
Figure A1: simplified flowchart of Constrained Genetic Algorithm Optimization with ANDI termination criteria
Al-Amri, A., & Zahid, U. (2020). Design Modification of Acid Gas Cleaning Units for an Enhanced Performance in Natural Gas Processing. Energy and Fuels, 34(2), 2545–2552. https://doi.org/10.1021/acs.energyfuels.9b03671
Barvar, K., Safamirzaei, M., & Shariati, F. P. (2019). Energy Efficiency Enhancement in Amine Unit by Application of Multiple Lean Amine Streams in Amine Contactor. Gas Processing Journal, 7(2), 29–36.
Chilukuri, P., & Bowerbank, G. (2016). Benefits of integrated design of gas processing units to achieve optimal project economics. Society of Petroleum Engineers - Abu Dhabi International Petroleum Exhibition and Conference 2016, 2016-Janua. https://doi.org/10.2118/183542-ms
Dahiya, S., & Myllyvirta, L. (2019). Global SO 2 emission hotspot database RANKING THE WORLD’S WORST SOURCES OF SO 2 POLLUTION Content. Retrieved from https://www.greenpeace.org/static/planet4-africa-stateless/2019/08/5f139f4c-final-global-hotspot-and-emission-sources-for-so2_19th_august-2019.pdf
Dai, Y., Peng, Y., Qiu, Y., & Liu, H. (2019). Techno-Economic Analysis of a Novel Two-Stage Flashing Process for Acid Gas Removal from Natural Gas. Energies, 12(21), 4213. https://doi.org/10.3390/en12214213
Dara, S., A.AlHammadi, A., S.Berrouk, A., Al Khasawneh, F., Al Shaiba, A., & F.AlWahedi, Y. (2018). Carbon footprint reduction of acid gas enrichment units in hot climates: A techno-economic simulation study. Journal of Cleaner Production, 201, 974–987. https://doi.org/10.1016/j.jclepro.2018.08.067
Garmroodi Asil, A., & Shahsavand, A. (2014a). Reliable estimation of optimal sulfinol concentration in gas treatment unit via novel stabilized MLP and regularization network. Journal of Natural Gas Science and Engineering, 21, 791–804. https://doi.org/10.1016/j.jngse.2014.09.033
Garmroodi Asil, A., & Shahsavand, A. (2014b). Selecting Optimal Acid Gas Enrichment Configuration for Khangiran Natural Gas Refinery. Gas Processing Journal, 2(2). https://doi.org/10.22108/gpj.2014.20179
Garmroodi Asil, A., Shahsavand, A., & Mirzaei, S. (2017). Maximization of sulfur recovery efficiency via coupled modification of GTU and SRU processes. Egyptian Journal of Petroleum, 26(3), 579–592. https://doi.org/10.1016/j.ejpe.2016.08.003
Ibrahim, S., Jagannath, A., & Raj, A. (2020). Aromatics oxidation in the furnace of sulfur recovery units: Model development and optimization. Journal of Natural Gas Science and Engineering, 83(August), 103581. https://doi.org/10.1016/j.jngse.2020.103581
Lotfikatooli, L., & Shahsavand, A. (2017a). A Reliable Approach for Terminating the GA Optimization Method. Iranian Journal of Numerical Analysis and Optimization, 7(1), 101–122. https://doi.org/10.22067/ijnao.v7i1.48652
Lotfikatooli, L., & Shahsavand, A. (2017b). Reliable prediction of adsorption isotherms via genetic algorithm molecular simulation. Journal of Molecular Modeling, 23(1). https://doi.org/10.1007/s00894-017-3206-2
Mahdipoor, H. R., & Dehghani Ashkezari, A. (2016). Feasibility study of a sulfur recovery unit containing mercaptans in lean acid gas feed. Journal of Natural Gas Science and Engineering, 31, 585–588. https://doi.org/10.1016/j.jngse.2016.03.045
Mak, J., B.Nielsen, R., King Chow, T., Morgan, O., & Wai Wong, V. (2009). Patent No. 7635408. Retrieved from http://www.freepatentsonline.com/7635408.html
Northrop, S., Seagraves, J., Ramkumar, S., & Cullinane, T. (2019). ExxonMobil’s Experience with Sour Gas Treating and Acid Gas Handling. Abu Dhabi International Petroleum Exhibition & Conference. https://doi.org/10.2118/197137-MS
Palmer, G. (2004). Patent No. 20040226441. Retrieved from http://www.freepatentsonline.com/y2004/0060334.html
Palmer, G. (2006). Patent No. 7147691. Retrieved from http://www.freepatentsonline.com/7147691.html
Rameshni, M., & Santo, S. (2020). Patent No. 20200039825. Retrieved from https://www.freepatentsonline.com/y2020/0039825.html
Seider, W. D., Seader, J. D., & Lewin, D. R. (2018). Product and Process Design Principles: Synthesis, Analysis and Design. John Wiley and Sons, Inc, pp. 472–558.
Walas, S. M. (1998). Chemical process equipment: Selection and design. Boston: Butterworths.
Weiland, R. H. (2008). Acid gas enrichment I: Maximising selectivity. Petroleum Technology Quarterly, 13(2 SUPPL.).
ZareNezhad, B., & Hosseinpour, N. (2008). Evaluation of different alternatives for increasing the reaction furnace temperature of Claus SRU by chemical equilibrium calculations. Applied Thermal Engineering, 28(7), 738–744. https://doi.org/10.1016/j.applthermaleng.2007.06.014
تعداد مشاهده مقاله: 113
تعداد دریافت فایل اصل مقاله: 76