Modeling and Optimization of the Upgrading and Blending Operations

Subscriber access provided by Purdue University Libraries

Article

Modeling and Optimization of the Upgrading and Blending Operations of Oil Sands Bitumen Hossein Shahandeh, and Zukui Li Energy Fuels, Just Accepted Manuscript • DOI: 10.1021/acs.energyfuels.6b00037 • Publication Date (Web): 07 Apr 2016 Downloaded from http://pubs.acs.org on April 8, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Energy & Fuels is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

Modeling and Optimization of the Upgrading and Blending Operations of Oil Sands Bitumen Hossein Shahandeh, Zukui Li∗

Department of Chemical and Materials Engineering, University of Alberta Edmonton, AB, T6G1H9, Canada

Abstract A general framework is proposed for the operations optimization of bitumen upgrading plant in the oil sands industry. Based on simulation results from an upgrading plant in Aspen HYSYS environment, empirical models are developed through statistical analysis for different process units. Each generated correlation is a function of the relevant process unit operating conditions. All of the correlations are further used to develop the upgrading plant optimization model, which is a nonconvex Nonlinear Optimization (NLP) problem. The proposed model is tested on three examples in which different commodity demands are imposed as constraints: (i) no restriction for production, (ii) sweet Synthetic Crude Oil (SCO) production, and (iii) mandatory multiple production. Results demonstrate the efficacy of the proposed framework for the upgrading plant operations optimization.

Keywords: oil sands bitumen, process operations, upgrading plant, nonlinear optimization, operations optimization

∗

Author to whom correspondence should be addressed. Email: [email protected]; Tel: 1-780-492-1107; Fax: 1-780-492-2881

1 Environment ACS Paragon Plus

Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

1. Introduction Canada has large amount of oil reserves and most of the oil resources are in the form of oil sands and situated in the Western Canadian Sedimentary Basin, known as the Athabasca, Peace River, and Cold Lake fields.1 These oil sands fields are mostly located in the province of Alberta where oil sands bitumen production rate was 2.16 million barrels per day for the year of 2014.2 It has been estimated that this value will increase to 3.08 and 3.95 million barrels per day by the year 2020 and 2030, respectively.2

After extraction of bitumen through any surface or in situ approach, there are two alternatives for the next step. The extracted bitumen can be diluted and directly sent to the market as heavy crude oil. On the other hand, it can be upgraded beforehand and then be sold as value-added commodities. During the upgrading process, bitumen is usually processed to produce a commodity which can be processed by conventional refineries, namely, Synthetic Crude Oil (SCO).

The upgrading process can be operated through either hydrocracking or thermocracking. The former one is based on hydrogen-addition and the second one relies on carbon-rejection. In the first technology, the mass fraction of hydrogen is increased and it causes the breakup of large hydrocarbon chains and the formation of new lighter compounds. Based on their boiling points, these lighter compounds are mainly Naphtha (NPH), Light Gas Oil (LGO), and Heavy Gas Oil (HGO). LC-Finning is the ongoing technique based on hydrocracking. In the second technology, the large hydrocarbon chains can be broken into lighter ones through adding thermal energy. Delayed coking and fluid coking are the major running techniques based on thermocracking. Compared to hydrocracking, thermocracking has a lower conversion rate and produces coke which is an undesirable byproduct. Therefore, hydrocracking technology is more favorable due to its capability of yielding high quality distillates.3

Systematic operations optimization is necessary for different parts of oil sands industry, especially the upgrading process. The operating cost of bitumen production is quite high. This is due to the demand of variety of utilities during extraction and upgrading processes, including electricity, steam, hot water, freshwater, natural gas and hydrogen. The margin between SCO 2 Environment ACS Paragon Plus

Page 2 of 31

Page 3 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

price in market and its production cost is thus not that much, and it could be very vulnerable when the oil price is too low. Furthermore, it was found that bitumen upgrading is the most cost and energy intensive section of SCO production.1 Therefore, it is imperative to perform operations optimization for the bitumen upgrading process. However, limited research has been done in the past and the most relevant researches can be summarized as follows.

Modeling and optimization of Canadian oil sands operations was studied by Ordorica-Garcia et al.1, 4 For given oil production demands, different schemes were proposed for SCO and Diluted Bitumen (DilBit) productions. It included different extraction methods (e.g. surface and in-situ), upgrading technologies (such as fluid coking, delayed coking, and LC-fining), and energy producers with fixed capacity (like boiler, hydrogen plant, and power plan). Moreover, a CO2 emissions constraint was imposed during the optimization in order to target the level of greenhouse gas emissions of each energy producer. Owing to availability of historical and industrial data, the basic case study was based on year 2003.1, 4 Betancourt-Torcat et al. proposed an integrated optimization model for simultaneous analysis of energy producers and production schemes.5 Afterwards, they extended the developed deterministic model into a stochastic optimization model in order to account for parameter uncertainty in the natural gas price and steam-to-oil ratio plus imposing a constraint for freshwater withdrawal.6 When natural gas cost could change the optimal configuration drastically, it was found that hydrocracking-based plant is more efficient.3 A set of uncertain realizations were then defined to reach the optimal upgrading scheme meeting different commodity specifications at minimum cost.7 Since the optimization was in the presence of worst-case realizations, uncertain economic and operational factors, it concluded that the proposed model could find the robust scheme.

The research on oil sands upgrading optimization is very limited. Nonetheless, lots of attention has been paid to conventional refineries optimization. For instance, Leiras et al. reviewed refinery planning at different levels (strategic, tactical, or operational), and at different oil chain segments (upstream, midstream, or downstream).8 Key issues, advances, and future opportunities for scheduling, planning, and supply chain management of oil refinery operations were also discussed in another study.9 Regardless of optimization case study, formulating an appropriate and accurate model is the most essential step. Pinto and Moro modeled typical refinery process units10 and presented a general modeling framework for the operational planning of petroleum 3 Environment ACS Paragon Plus

Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

supply chains including processing unit, tank, and pipeline.11 Later on, they extended their study to multiperiod and uncertain case for production planning of petroleum refineries.12 It should be highlighted that it is difficult to apply rigorous mechanistic model in refinery planning and operating optimization. The reason is that the mechanistic modeling makes the optimization problem too complex. Therefore, simplified models have been widely used such as (i) empirical correlations for estimation of different products properties,13-15 (ii) swing cuts approach for distillation units,13, 14, 16, 17 or (iii) artificial neural network.18, 19 These types of equations can be expressed as linear or nonlinear equations.

An interesting research field has been on crude scheduling and allocation optimization. The scheduling part addresses unloading crude oils into storage tanks from ships or tankers – depending on topology of case study – arriving at different times. The allocation part focuses on sending feeds with various rates over time to the distillation columns. The crude oil unloading, storing, and processing in a marine-access refinery was studied by Reddy et al.20 Simultaneous solving of oil quality, transfer quantity, tank allocation, and oil blending were addressed based on a novel Mixed Integer Linear Optimization (MILP) solution algorithm. Mendez et al. proposed an integrated MILP-based approach to optimize gasoline off-line blending and scheduling of a refinery at the same time.21 To avoid Mixed Integer Nonlinear Programming (MINLP) problem, an iterative procedure was presented which could be applied to both discrete and continuous time formulations. Numerical comparisons demonstrated that the proposed approach can converge to the same solutions, but faster, due to linearization. Li et al. worked on the gasoline blending recipe and scheduling decisions.22 A slot-based continuous-time MILP formulation was developed for an integrated recipe, specifications, blending, and storage problem. Several real-life operating features and policies were included in the model.

Refinery planning optimizations has been another interesting subject in process systems engineering. During planning optimization, finding the optimal flow rates of streams is the main purpose. These variables might be independently defined or might be specified as a function of operating conditions of associating units. Blending rules can also be incorporated into the model wherever it is necessary. One of the first attempts in this area was carried out with Alhajri et al.13 Their proposed model estimated product properties of Crude Distillation Unit (CDU) and Fluid Catalytic Cracking unit (FCC) when the independent variables were the cut-point temperatures 4 Environment ACS Paragon Plus

Page 4 of 31

Page 5 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

and conversion, respectively. To meet the market specifications for commodities, some constraints and blending rules were imposed in addition to the product demands. Afterwards, a robust optimization methodology was used.23 When flow rates were independent variable in this study, uncertainties in costs, prices, commodities demands, and product yields were addressed. However, this approach was only implemented for a small case study and its capability to solve real cases is unknown. Moreover, effects of operating variables such as pressure and temperature were not taken into account. In another work, Guerra et al. first developed nonlinear empirical models for CDU and FCC process units.14 They then implemented those models in a NLP problem for a small case (with only presence of CDUs) and a medium case (with presence of CDUs and FCC).24 The empirical models related each unit outlets’ properties, yield, flow rate and so forth to the operating conditions.

The presented literature review reveals that studies have mainly focused on conventional refinery operations optimization and there have been limited contributions to oil sands operations optimization. Furthermore, the existing studies of oil sands operations have some limitations. First, the only commodity in these studies was assumed to be sweet SCO and the other alternatives were neglected. Second, the operating conditions of process units were fixed, and hence, specifications (such as nitrogen and sulfur contents) changes were not considered. To address these limitations, a new framework is proposed in this paper for simulation-based modeling and optimization of the upgrading plant with multiple bitumen blend and SCO product alternatives.

This article is organized as follows: section 2 describes the simulation of upgrading plant with Aspen HYSYS. Section 3 presents statistical analysis which is used to generate the correlations of different properties for each unit. Section 4 presents formulation of the operations optimization for the upgrading plant being composed of the process models, commodities specifications constraints and the objective function. The solution strategy is also given in this section to explain how the global optimal point can be achieved. Three different examples are studied in Section 5 to check performance of the proposed approach. Concluding remarks and suggesting extensions for future works are then given in the last section.


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 31

2. Process simulation The upgrading process simulation with Aspen HYSYS is discussed in this section. In this work, instead of using rigorous mechanistic models of bitumen upgrading units, empirical regression models are identified for operations optimization. Note here, R2 is employed to measure accuracy of the regression model. Deriving the correlations requires large amount of sample data of each unit at different operating conditions. In this study, operating units are first simulated via Aspen HYSYS V8.4®, and most of the parameters and assumptions are taken from 25

and 26. In the following paragraphs, the simulation procedure is discussed.

The first step of the simulation is defining the available components. There might be more than 1000 distinctive components in bitumen mixture so finding the information for all the existing components would be time consuming. Another alternative for the simulation of bitumen has been applying hypothetical components. To do so, assay properties are imported from experimental results to the simulator, namely, specific gravity, viscosity, molecular weight, sulfur content, nitrogen content and so forth.27 Necessary experiments for data collection of Western Canadian bitumen sample have been also carried out in two works.28, 29 In the Aspen HYSYS environment, many assays have been predefined.30, 31 In this study, Cold Lake blend2011 (which represents a typical diluted bitumen product of Alberta) is used for simulation of bitumen upgrading plant. Recommended by the software guideline, Peng-Robinson thermodynamic package is also selected as the equation-of-state package.30

Next step is adding operating units into the simulation environment and connecting them together. The simulating facility is for the sweet SCO production (see Figure 1).25,

26, 32, 33

Atmospheric Distillation Unit (ADU) is used first to separate diluent from the feed. Afterwards, a Vacuum Distillation Unit (VDU) is applied for separation of the original petroleum mixture into light ends, NPH, LGO, HGO and vacuum residue. While separated NPH, LGO and HGO are then sent to their respective hydrotreaters, the vacuum residue undergoes the Canmet slurry hydrocracker in which large molecules with high molecular weight are fractioned into smaller molecules with lower molecular weight and boiling point.25,

26

Afterwards, the hydrocracker

lighter products are separated into NPH, LGO, and HGO again and hydrotreated. Moreover, the hydrocracker residue is recycled to the hydrocracker to reach higher efficiency (conversion rate).


Page 7 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

Auxiliary units including pump, heater, and cooler are also considered for pressure and temperature or phase changes. In order to provide some perspective about the upgrading plant, feed operating conditions, specifications of atmospheric and vacuum distillation units, and cut point ranges of different products are reported in Table 1.

Figure 1. Process flow diagram of upgrading plant simulation

Table 1. General specifications for the simulated bitumen upgrading plant Feed Properties25, 26 Temperature

67 ºC

Pressure

101.3 kPa

Flow rate

385,500 kg/h

ADU25, 32 Top pressure Bottom pressure Naphtha recovery

Cut point25

120 kPa

NPH

40-180 ºC

140 kPa

LGO

180-360 ºC

90 %

HGO

360-540 ºC

VDU25, 32 Top pressure Bottom pressure

2 kPa 5 kPa

3. Correlation modeling In this section, the procedure for generating the correlation models of process units is explained. Basically, the empirical models are developed to replace the rigorous mechanistic model including kinetic and equilibrium relations, mass and energy balances. The correlations try to estimate the yield and properties of outlet streams based on operating conditions such as temperature, pressure, Liquid Hourly Space Velocity (LHSV), and cut point of NPH, LGO, and HGO. The important properties include the yield of products, specific gravity, viscosity, sulfur


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 31

content and nitrogen content of streams. Hence, five correlations should be found for each outlet stream for every unit. The process units which need correlations are the separators (ADU and VDUs) and the reactors (hydrocracker and hydrotreaters).

For distillation units, Petroleum Distillation Column module is chosen for the simulation in which the cut points of withdrawing streams are the independent operating variables. The pressure is assumed to be fixed for both ADU and VDU.25,

26

The variables are known as

Effective Cut Point (ECP) in Aspen HYSYS environment. It should be noted here, when there are p withdraws, only the pth and p+1th ECPs are effective on the properties of the pth withdraw. For example, specific gravity of withdraws can be formulated as: SG p = f ( ECPp , ECPp +1 )

(1)

∀p ∈ P

where SG is specific gravity and P is the set of VDU products, which includes NPH, LGO, and HGO. In addition, empirical equations were found from literature5 for steam requirement of ADU and VDU based on their inlet mass flow rate. The corresponding operating costs of separation units are incorporated into the optimization problem. steam M ADU = 0.30 M in

(2)

steam M VDU = 0.07 M in

(3)

For the hydrocracking unit, a Petroleum Shift Reactor and a Petroleum Distillation Column modules are selected.25, 26 These modules are incorporated with spreadsheet in order to manually tune some outlet properties (i.e., yield, sulfur and nitrogen distributions) via available correlations from literature. For example, conversion rate of residue in the hydrocracker, hydrodesulfurization, and hydrodenitrogenation are given as follows. residue u

CR

 = 1 − 1 + k A PH 2 

( )

β

V (1 − ε )   VAB0 

−1

CRuHDS = a + b ( CRuresidue ) + c ( CRuresidue )

2

u = hydrocracker

(4)

u = hydrocracker

(5)

HDN

CRuHDN = Ratio HDS × CRuHDS

u = hydrocracker HDN

where k A , β , ε , a, b, c, and Ratio HDS are all reported in the references.25, 26

Sulfur distribution in the outlets of hydrocracker is expressed as 8 Environment ACS Paragon Plus

(6)

Page 9 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

γ p = a p ln ( CRuresidue + d p ) + bp ( CRuresidue ) + c p ( CRuresidue )

2

∀p ∈ P, u = hydrocracker

where ap, bp, cp, and dp are available in the previous works.24,

25

(7) Temperature, pressure and

LHSV are the variables for hydrocracker, and ECPs are the variables for the VDU. During hydrocracking, there is no separation and breakup of large hydrocarbon chains into new lighter compounds is only taken place. Therefore, we cannot develop desired correlations for each NPH, LGO, and HGO. As these three products are separated through the following VDU, it would be better to present correlations of these two units as a single one. Note here, the simulations of these units are carried out separately, and it is just assumed that they are united during generation of empirical models. By this way, each property can be formulated based on operating variables for each product; however, the number of variables is larger than a single unit: SGu , p = f (Tu , Pu , LHSVu , ECPp , ECPp +1 )

(8)

∀p ∈ P , u = hydrocracker

where the index of hydrocracker for u represents the hydrocracker and the following VDU units. For hydrotreaters, hydrodesulfurization and hydrodenitrogenation are the main reactions taking place with catalysts. Conversion rates of these two reactions were studied through experiments and corresponding correlations were proposed for NPH-hydrotreater,34 LGO-hydrotreater,35 HGO-hydrotreater.36, 37 The following empirical models are used for hydrodesulfurization and hydrodenitrogenation of NPH-hydrotreater, LGO-hydrotreater, and HGO-hydrotreater (given in equations 9-10, 11-12, and 13-14 respectively).

(

CRuHDS = 1 − exp k HDS × P β

(

HDS

CRuHDN = 1 − exp k HDN × P β

HDN

LHSV α

HDS

LHSV α

)

HDN

u = NPH

)

(9)

u = NPH

CRuHDS = 97.82 + 2.62 × Tu − 0.87 × Pu − 2.46 × LHSVu − 1.76 × Tu2 − 2.79 × LHSVu2 +1.53 × Tu × LHSVu

(10) (11)

u = LGO

CRuHDN = 97.63 + 4.45 × Tu + 0.95 × Pu − 5.87 × LHSVu − 4.57 × Tu2 − 2.08 × Pu2 − 4.44 × LHSVu2 + 4.68 × Tu × LHSVu

(12)

u = LGO

CRuHDS = 110.97 + 3.15 × Tu − 31.04 × LHSVu − 3.09 × Pu + 9.41× Tu × LHSVu + 0.70 × Tu × Pu + 3.15 × LHSVu × Pu − 2.87 × Tu2 + 8.01× LHSVu2 + 3.77 × Pu2 CRuHDN = 124.14 + 37.04 ×Tu − 95.96 × LHSVu + 3.02 ×Pu − 6.13 × Tu × LHSVu + 6.25 × Tu × Pu + 4.77 × LHSVu × Pu + 3.08 × Tu2 + 26.05 × LHSVu2 + 2.55 × Pu2


u = HGO u = HGO

(13) (14)

Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 31

For the hydrotreaters, the remaining properties are developed from the simulation results obtained from Aspen HYSYS. Note here, correlations here are based on the variation of each term instead of the actual value: ∆SGu = f (Tu , Pu , LHSVu )

∀ u ∈ U \ {hydrocracker}

(15)

For the hydrocracker and each hydrotreater, hydrogen is also added, and the way to estimate its flowrate was presented in the reference (see equations 16 and 17).25 H u = 2000 × CRuresidue

H u = f ( Fu )

u = hydrocracker

(16)

∀u ∈U \ {hydrocracker}

(17)

Furthermore, before these units, pump and heat exchanger are installed to adjust the temperature and pressure. The required work for the pump is mainly a function of the flow rate and the pressure (equation 18), and the required duty is dependent on the flow rate and the temperature (equation 19). Wu = f ( Pu , Fu )

∀u ∈ U

(18)

Qu = f (Tu , Fu )

∀u ∈ U

(19)

where F is the mass flow rate, and U is the process unit set which includes hydrocracker, NPH, LGO and HGO hydrotreaters.

After identifying important properties and their related operating variables, simulation should be performed at different operating conditions in order to have enough samples for the generation of correlations. The first step is determining a valid range for each independent variable. Independent variables of each unit and their ranges ( xmin , xmax ) are provided in Table 2. For the ECPs, beginning and ending boiling point temperatures of each product is subtracted and added by 10 ºC, respectively. Pressure, temperature, and LHSV ranges of reactors are adapted from literature at which the obtained correlations are valid.

Table 2. Operating variables to be optimized and the corresponding ranges Variable

VDU25

Hydrocracker25

Hydrotreater NPH34


LGO38

HGO38

Page 11 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

Temperature (˚C)

–

420-480

260-280

330-350

350-370

Pressure (MPa)

–

10-18

3-5

6.9-12.4

6.1-10.2

–

0.2-1

1-2

0.5-2

0.5-2

ECP1 (˚C)

30-50

30-50

–

–

–

ECP2 (˚C)

170-190

170-190

–

–

–

ECP3 (˚C)

350-370

350-370

–

–

–

ECP4 (˚C)

530-550

530-550

–

–

–

-1

LHSV (hr )

For the VDUs, three levels of variation are selected for each ECP. When each product is only dependent on two consecutive ECPs, total number of simulations are 9 (=3×3) and full factorial method is adopted for the design of simulations. For the NPH-hydrotreater, three levels of variation are selected for pressure and LHSV, and four levels of variation are assigned for temperature due to its wider range. As a result, total number of simulations are 36 (=3×4×3) and full factorial method is chosen again for the design of simulations. For the LGO- and HGOhydrotreater, four levels of variation are selected for all variables owing to their wider ranges of changes. Full factorial method is still the best choice for design of simulations since the total number of simulations is still not too large (64=4×4×4).

For the hydrocracking part, as explained before, we unite hydrocracker and its separator in one unit for the correlations development. Hence, number of variables for estimation of each product properties is five (temperature, pressure, LHSV, and two ECPs). Furthermore, five levels of variation are assumed for temperature, four levels of variation are used for pressure and LHSV, and three levels of variation are adopted for ECPs. A full factorial design leads to 720 (=5×4×4×3×3) simulations, which is time-consuming. So Latin Hypercube Sampling method is used to narrow down the number of experiments to only 10% of actual ones. Existence of a recycle stream from the second VDU to the hydrocracker (see Figure 2) makes the developed correlations for this unit less accurate than others. Full factorial analysis was carried out for this unit first, but with fewer levels for the ECPs. Nevertheless, the results were not more accurate. Therefore, where enhancing the accuracy of correlations was not feasible for this unit, number of simulation runs is tried to be as less as possible in order to at least reduce the computational time.


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 31

After performing all the assigned simulations at different operating conditions, all the data is exported to Design-Expert® software in order to generate the correlations. Since the order of variables values are quite different, coded variables are calculated and used instead of actual levels of the variables according to the following equations.35 Notice that, the actual values are used for the LHSVs (0.5–2) due to their closeness to coded values (–1 to +1). xcoded =

2 x − ( xmax + xmin ) xmax − xmin

(20)

R2 is then taken into account to measure how close the simulation data are to the fitted regression curve. Wherever it is possible, simple linear correlations are chosen. Otherwise, quadratic terms are included to get higher accuracy. Regardless of the degree of generated polynomial, ineffective terms in each correlation are excluded in order to have simpler equations for optimization. R2 of 0.99 or higher is achieved for all properties in different units except the hydrocracker in which few properties could only get R2 higher than 0.9.

All the correlation models are provided in the supplementary material.

4. Optimization The upgrading plant operations optimization problem can be stated as follows. The given information includes a fixed DilBit feed flowrate with known properties, a set of operating units, a set of correlations for each operating unit to estimate properties of products obtained from previous section, a set of correlations for each operating unit to estimate the required work, duty, and hydrogen consumptions obtained from previous section, the demands of a set of commodities being composed of DilBit (the bitumen diluted with naphtha), sour SCO (mixture of untreated NPH, LGO, and HGO), sweet SCO (mixture of treated NPH, LGO, and HGO), SynDilBit (mixture of sweet SCO and DilBit), SynBit (mixture of sweet SCO and Bitumen), a set of specifications including viscosity, specific gravity, sulfur content and nitrogen content for each commodity, the costs of feed, commodities, and utilities. The objective is to determine the optimal operating condition of process units and flow rates to get the maximum profit.

Illustrative upgrading plant


Page 13 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

As shown in Figure 2, a multiproduct upgrading plant is considered based on hydrocracking technology. The optimization model is then formulated accordingly. DilBit feed can undergo the upgrading process or it can be sold directly to the market. The separated diluent from ADU is sent back to the extraction plant. The SCO is a mixture of NPH, LGO, and HGO, and it might be produced as sour or (and) sweet SCO(s). To produce SynDilBit or SynBit, sweet SCO is mixed with DilBit or bitumen, respectively. Several mixers and splitters are included in the plant to make the multi-production purpose possible. Depending on whether there is any demand for a specific commodity or not, some of the interconnection might be zero. Three intermediate products of NPH, LGO, and HGO are the main outlets from VDU and hydrocracker. After mixing together, they can be sent to hydrotreaters for the sweet SCO production, or directly used for the sour SCO production.

Diluent DilBit DilBit

Sour SCO ADU

VDU Q

Q, W,H2

NPH LGO Q

Hydrotreater (NPH)

SynDilBit

HGO Q, W,H2 Hydrocracker + VDU

Sweet SCO Hydrotreater (LGO)

NPH LGO

Q, W,H2 SynBit

HGO Hydrotreater (HGO) Q, W,H2

Bitumen

Figure 2. Multiproduct bitumen upgrading and blending plant

Mathematical optimization model The model presented in this section is based on the following assumptions: the feed specifications are fixed, the arrangement of units is fixed, the hydrocracking technology is chosen for upgrading, and the generated correlations are valid in the range of operating conditions from Table 2. The proposed optimization model mainly consists of equality


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 31

equations: (i) total mass balance for each unit, (ii) properties correlations of outlets from each operating unit, and (iii) blending rules after each mixer. Inequality equations are for product demands and their specifications. The formulated model is a nonconvex NLP problem due to the nonlinearity appearing in correlation models and the blending rules. For instance, appearances of bilinear terms in the correlations are nonconvex terms. Detailed model is presented in the following parts for the upgrading plant shown in Figure 2.

For each operating unit (Figure 3(a)), there is one equation with respect to total mass balance. This equation is for estimation of outlet flow rates which can be expressed through the yield. M _ out p , u = X p , yield × M _ in p , u

∀p ∈ P , u ∈ U

(21)

On the other hand, there are five correlations for each individual outlet from any operating units, namely, yield, specific gravity, viscosity, nitrogen content and sulfur content. These correlations are function of the operating conditions. A comprehensive list of correlation models can be found in the supplementary material document. Notice that, the variables in these equations are coded variables as explained before. For example, the following correlations are used for the VDU (equation 22), hydrocracker (equation 23), and hydrotreaters (equation 24), respectively: X p , pr = f ( ECPp , s , ECPp +1, s )

∀p ∈ P, pr ∈ PR, s = VDU1

(22)

X p, pr = f (ECPp,s , ECPp+1,s , LHSVu , Tu , Pu ) ∀p ∈ P, pr ∈ PR, u = hydrocracker, s = VDU2

(23)

X p , pr = f ( LHSVu , Tu , Pu )

(24)

∀ p ∈ P , pr ∈ PR , u ∈ U \ {hydrocracker}

As an example, yield of LGO withdrawing from VDU is: X LGO , yield = 0.21387 − 4.13841×10−3 × ECPLGO,VDU + 0.017477 × ECPHGO,VDU

(25)

− 1.81078 ×10−3 × ECPLGO ,VDU × ECPLGO,VDU + 4.75853 ×10−5 × ECPHGO,VDU × ECPHGO,VDU

M_outp,u . . .

M_inp,u

M_outp,u

(a)

M_in 1

M_out1

. . .

M_out

M_in

. . .

M_outi

M_in i

(b)

(c)

Figure 3. Schematics of units (a: process unit, b: mixer, c: splitter)


Page 15 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

The mass balance for mixers are modelled as M _ out = ∑ M _ ini

(26)

i

For nitrogen and sulfur content, blending rules are based on mass compositions.13 The blending rule of specific gravity is based on volumetric flow rate. in X prout = ∑ M _ ini × X pr,i / M _ out ,

∀pr ∈ {Nitrogen,Sulphur}

(27)

i

out in X SG = ∑ CVi × X SG ,i

(28)

i

The viscosity is converted to blending index (equation 29) and then blended using volumetric fraction.39 The blending index for mixture is converted back to the viscosity (equation 31). BI i =

in log(X viscosity,i )

(29)

in 3+log(X viscosity,i )

BI Blend = ∑ CVi × BI i

(30)

i

3× BI Blend (1-BI Blend )

out X viscosity = 10 

(31)

For splitters, only the mass balance is modelled. M _ in = ∑ M _ outi

(33)

i

For commodities, quality constraints should be imposed to meet market quality requirements. Parameters for this part are adapted from literature40 and they are shown in Table 3. X fp , pr ≤ Spec fp , pr

(32)

Table 3. Commodities specifications Specification

Sour SCO

Sweet SCO

SynDilBit

SynBit

Sulfur content (%wt)

3.180

0.180

2.510

3.150

Nitrogen content (wtppm)

1515.0

691.5

4456.4

2689.0

Specific gravity

0.9396

0.8576

0.9377

0.9347

Viscosity (cSt)

96.4

7.2

172

179

To make sure that the volumetric blending ratio in each commodity is also in the valid range, lower and upper bounds are also imposed on the commodities (see equation 33 and Table 4). θ i ≤ CVi ≤ ϕi

(33)


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 31

Table 4. Lower and upper bounds for volumetric blending ratio of commodities Sour SCO

Sweet SCO

SynDilBit

SynBit

θi

ϕi

θi

ϕi

θi

ϕi

θi

ϕi

NPH

0.05

0.30

0.05

0.30

–

–

–

–

LGO

0.20

0.65

0.20

0.65

–

–

–

–

HGO

0.20

0.65

0.20

0.65

–

–

–

–

Sweet SCO

–

–

–

–

0.51

0.77

0.53

0.80

Components

The objective function of this study is the profit as given below, and its economic parameters are reported in Table 5. The profit is calculated as the summation of incomes from selling different commodities minus energy and feed costs. Profit = ∑ V fp Price fp − V feed Price feed − ∑ H u Price Hydrogen − ∑ Wu Price Electricity − ∑ Q u Price Steam fp

u

u

(34)

u

Table 5. Price data Parameters

Price

Hydrogen ($ kg-1)

1.73

Steam ($ tonne-1)

20.041

Electricity ($ kWh-1)

0.0763563

Sour SCO ($ bbl-1)

40

Sweet SCO ($ bbl-1)

60

SynDilBit ($ bbl-1)

50

-1

SynBit ($ bbl )

45

DilBit ($ bbl-1)

10

Solution strategy In this section, the solution strategy is explained. The optimization model presented in the previous section is a nonconvex NLP. Therefore, getting trapped in a local optimum point is possible by implementing a local optimization solver. To avoid this problem, global optimization is applied in order to search for the optimal solution within a predefined time limit. Specifically,


Page 17 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

the optimization problem is modeled in GAMS42 and the solver BARON43 is used for the global optimization.

Introducing tight bounds for all variables in the model is very important to reach the global optimality of nonconvex optimization problem. This is due to use of these bounds in the convex envelopes for under- and over-estimating the nonconvex terms of the problem. First, lower bounds of mass and volume flow rates are zero, and their upper bounds are set equal to feed mass and volume flow rates. The reason is that none of the streams can have higher flow rates than the inlet feed. Second, some variables are based on percentage so their lower and upper bounds can be easily fixed at 0 and 100, respectively. Thirds, –1 and +1 are lower and upper bounds of coded operating variables. Finally, for the rest of variables, the objective function is replaced with the corresponding variable and BARON solver is implemented. After doing optimization for each variable, safe bounds are identified.

There are some points about the LHSV variables which should be clarified. This variable shows the residence time of the inlet fluid in a vessel, and it can be calculated by dividing the inlet volumetric flow rate to the vessel volume (see equation 35). In this work, hence, the LHSV can be concluded as a variable since the volumetric flow rate of inlet can change. LHSVu =

Vuin Volumeu

∀u ∈ U

(35)

In this study, the inlet flow rate of reactors are variables, and so LHSV is a dependent variable in the adapted correlations for the conversion rates. However, there is not any information about capacity of the reactors in previous studies. To resolve this issue, the following procedure is found applicable. First, after formulating the whole model in GAMS, the main objective function of profit is replaced with the inlet volumetric flow rate of each reactor. Notice here, the LHSVs are assumed independent variables at this step. Second, the maximum and minimum of inlet volumetric flow rate to the reactors are obtained by performing the maximization and minimization of the model. Third, the maximum and minimum of reactors volume can be calculated employing equation 35, since the upper and lower bounds of LHSVs were reported for the each available correlation of conversion rate (see Table 2). Fourth, the average value of the


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 18 of 31

found maximum and minimum volumes is assigned as the pre-designed reactor capacity during the main case optimization.

5. Case studies Three different cases are presented in this section to illustrate the proposed optimization framework. They are all implemented in GAMS and solved with a desktop computer (single core of Intel® i5-4590 @ 3.30 GHz, 8 GB RAM). Furthermore, all of the solutions are performed within a time limit of 10 hours and optimality gap stopping criterion 1%. The decision variables in the proposed model are the operating conditions of the process units, and mass and volume flow rates. Results of each example are interpreted accordingly.

Example 1 In this case, there is no demand on any specific commodity. A combination of the sour and sweet SCOs, SynDilBit, and SynBit, without any restriction on their flow rates, can be produced in order to achieve more benefits. After global search, optimal profit of 51,191.5 ($ hr-1) is found for this case. When there is no pre-specified demand for commodities, the SynDilBit is the only commodity which is produced under optimal conditions. According to Table 5, the SynDilBit is the second most expensive selling commodity. Due to all the imposing constraint and lower yields of light compounds, production of sweet SCO is not the most economical alternative here, even though its selling price is the highest among the all commodities. Note here, when 1516.7 (m3 hr-1) SynDilBit is produced, 37% of feed (DilBit) should be sold directly to the market unprocessed. The blending ratio of sweet SCO and DilBit is also illustrated in Figure 4.


Page 19 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

Figure 4. Optimal product composition and the blending ratio – example 1

In Table 6, comparison of current and required values for different specifications is reported for the SynDilBit. All the four specifications are upper bounds, and as it can be seen, current values are all lower than them. In the sense of optimization, the nitrogen content of sweet SCO and the sulfur content of SynDilBit are the active constraints, since they are at the highest allowable values.

Table 6. Comparison of product qualities and their specifications – example 1 Variable

SynDilBit solution

spec

Viscosity (cSt)

14.7

172


1997.0

4456.4


1.7

2.51

Specific gravity

0.889

0.9377

Before interpreting the solution results of operating conditions, the effects of pressure, temperature, and LHSV on the process need to be clarified. The higher the pressure, temperature, and LHSV, the more conversion can be achieved for the desired reactions in the hydrocracker or hydrotreaters. Nevertheless, increasing the pressure and temperature leads to more electricity and steam costs, respectively. Moreover, a larger LHSV results in increase of both electricity and steam costs simultaneously. Optimal operating conditions are reported in Table 7. Some points should be highlighted here by comparing the results in Table 7 with operating ranges given in 19 Environment ACS Paragon Plus

Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 20 of 31

Table 2. Temperatures are the most effective variables, and they are set near or at their upper bounds for all units. The LHSVs are all approximately in the middle of defined range. There is not a clear trend for the pressure. Its optimal value in NPH-hydrotreater is at the lower bound value. The reason is that pressure was not an effective variable in the developed correlations,34 so lowest value leads to lowest electricity cost. When the pressure is in the middle for the LGOhydrotreater, they are at the highest extreme for the hydrocracker and HGO-hydrotreater. Based on ECPs values, one can understand that the NPH is separated in a wider range in the first VDU. This trend is vice versa for the HGO separation. For the LGO withdrawing, the ranges of cut points have nearly the same length, but the LGO withdrawing from the second VDU is lighter.

Table 7. Optimal operating conditions of process units – example 1 Variable

VDU1

Temperature (˚C)

Hydrotreater

Hydrocracker and VDU2

NPH

LGO

HGO

–

477.0

280

350

370

Pressure (MPa)

–

18

3

8.4

10.2

LHSV (hr-1)

–

0.48

1.15

1

0.86

ECP1 (˚C)

30

38.1

–

–

–

ECP2 (˚C)

190

170

–

–

–

ECP3 (˚C)

370

354.8

–

–

–

ECP4 (˚C)

530

550

–

–

–

Example 2 In the second case, the only option of production is the sweet SCO. The Sour SCO, SynDilBit, and SynBit are not produced at all. The imposing constraint here is slightly different. The flow rates of sour SCO, SynDilBit, and SynBit are fixed at zero, as the flow rate of sweet SCO is greater than or equal to zero. Global solution of 33,750.4 ($ hr-1) is found in this case, which is lower than the corresponding value in the example one. This is because of applying stricter constraints for final products. Since the sweet SCO is the only acceptable product in this case, a larger portion of feed is separated from the beginning of the process without undergoing any processing unit. Note here again, the


Page 21 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

sweet SCO is the most valuable commodity in this study. Due to commodities specifications constraints, it would be impossible to process the whole feed flow rate into just a single product. Hence, large amount of DilBit is sent directly to the market. Production rates of 864.4 and 1562.1 (m3 hr-1) are found for sweet SCO and DilBit, respectively. Figure 5, illustrates the blending ratios of selling products and sweet SCO. Four specifications constraints are met, and nitrogen and sulfur contents of sweet SCO are the active constraints (see Table 8). Table 9 includes the optimal operating conditions of the example two. Like the former case, the temperatures are the most effective variables in the process units. The LHSVs are again all approximately in the middle of defined range. The Pressures and ECPs also have the same trends as previous case.


Table 8. Comparison of product qualities and their specifications – example 2 SCO Sweet

Variable

solution

spec

Viscosity (cSt)

7.0

7.2


691.5

691.5


0.18

0.18

Specific gravity

0.853

0.8576


VDU1

Hydrotreater

Hydrocracker and VDU2

NPH


LGO

HGO

Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 22 of 31

Temperature (˚C)

–

477

280

350

370

Pressure (MPa)

–

17.58

3

8.38

10.2

LHSV (hr )

–

0.48

1.15

1

0.86

ECP1 (˚C)

30

38.7

–

–

–

ECP2 (˚C)

190

170

–

–

–

ECP3 (˚C)

370

355.6

–

–

–

ECP4 (˚C)

530

550

–

–

–

-1

Example 3 The third example is something between the first and second cases in terms of strictness for production demands. Here, small demands of 20 (m3 hr-1) are assigned for all the four commodities, namely, sweet and sour SCOs, SynDilBit, and SynBit. In the optimization model, volumetric flow rates of the mentioned streams are greater than or equal to 20. Under these new circumstances, 48,593.1 ($ hr-1) is found as the best possible answer for the objective function through the global optimization. It is worth to mention that the profit for this case is less than the first case owing to the fact that selling the small amounts of all commodities are forced. Like examples one and two, a portion of feed cannot be processed and it is sold as DilBit. The optimal flow rates of sour and sweet SCOs, SynDilBit, SynBit, and DilBit are 20, 20, 1390.4, 20, and 975.8 (m3 hr-1), respectively. The volumetric blending ratios of selling products are demonstrated in Figure 6. Like former cases, four specification constraints are met here. The viscosity, sulfur and nitrogen contents of sweet SCO and the nitrogen content of sour SCO are the active constraints here (see Table 10). The optimal operating conditions for this example are reported in Table 11. The same trends exist here for all pressure, temperature, and LHSV and ECPs.

Table 10. Comparison of product current qualities and their specs – example 3 Variable

Sweet SCO solution spec

Sour SCO solution

spec

SynDilBit solution spec


SynBit solution spec

Page 23 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

Viscosity (cSt)

7.2

7.2

3.0

96.4

14.9

172

28.1

179

Nitrogen (wtppm)

691.5

691.5

1515.0

1515.0

1996.1

4456.4

2192.2

2689.0

Sulfur (%wt)

0.18

0.18

1.26

3.18

1.72

2.51

1.90

3.15

Specific gravity

0.854

0.8576

0.862

0.9396

0.889

0.9377

0.907

0.9347



Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 24 of 31


VDU

Hydrocracker

Temperature (˚C)

–

Pressure (MPa)

Hydrotreater NPH

LGO

HGO

475.0

279.6

350

370

–

18

3

8.67

10.2

LHSV (hr-1)

–

0.47

1

0.92

0.87

ECP1 (˚C)

30

38.3

–

–

–

ECP2 (˚C)

186.6

170

–

–

–

ECP3 (˚C)

360.8

351.3

–

–

–

ECP4 (˚C)

530

539.8

–

–

–

Finally, problem size and computational time of examples are discussed here. The number of variables and equations are the same for the three examples (250 and 269, respectively), and the only difference between them is demand constraints. The computational times for the examples presented in the previous section are shown in Table 12. Even though the size of three cases are the same, the optimization can be carried out much faster in the second example compared to the other two by using stricter constraints for the commodities demands.

Table 12. Comparison of the presented examples Example Optimality gap

CPU time (s)

Objective ($ hr-1)

1

21.5

36,000

51,191.5

2

1.0

303

33,750.4

3

19.8

36,000

48,593.1

Validation of optimization results In this section, validation of the optimal solutions obtained by GAMS is presented. Since the empirical models are employed during the optimization, the optimal results need to be resimulated in Aspen HYSYS environment to make sure that the formulated model is accurate enough. To do so, the optimal solution of the third case is used which is the most general case due to producing all the commodities with minimum values. Independent variables including


Page 25 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

ECPs, temperatures, pressures, and flow ratios of mixtures are imported in the simulator, and LHSVs (as dependent variables) and commodity properties and volume flow rates are exported for the validation. Comparison of the simulation and optimization results is provided in Table 13. The average of errors between the simulation and optimization results is 5.1%. Accordingly, the simulation results are in agreement with those achieved through the optimization, and it can be concluded that the generated correlations are sufficiently precise to be applied in the optimization model.

Table 13. Comparison of simulation and optimization results Commodities

Properties

Sour SCO

Sweet SCO

SynDilBit

SynBit

3.3

9.4

14.0

30.7

3.0

7.2

14.9

28.1


1555.5

646.5

1945.3

2149.0

Nitrogen content† (wtppm)

1515.0

691.5

1996.1

2192.2

Sulfur content* (%wt)

1.23

0.16

1.67

1.86

Sulfur content† (%wt)

1.26

0.18

1.72

1.90

Specific gravity*

0.861

0.904

0.917

0.939

†

Specific gravity

0.862

0.854

0.889

0.907

Flow rate* (bbl hr-1)

21.0

19.6

1376.3

19.6

Flow rate† (bbl hr-1)

20.0

20.0

1390.4

20.0

Dependent variable

Hydrocracker

LHSV* (hr-1) LHSV† (hr-1)

Viscosity* (cSt) †

Viscosity (cSt) *

*

Hydrotreater NPH

LGO

HGO

0.50

1.07

0.96

0.89

0.47

1.00

0.92

0.87

From Aspen HYSYS simulation and † from GAMS optimization.

6. Conclusion In this article, a novel optimization framework has been proposed for bitumen upgrading plant. The hydrocracking-based upgrading plant is first simulated with Aspen HYSYS software. The process is then simulated under various operating conditions in order to get adequate amount of 25 Environment ACS Paragon Plus

Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 26 of 31

data for correlation modeling. Afterwards, empirical models are generated to estimate properties of outlet stream as function of the operating conditions. According to the developed model, the proposed optimization problem is a nonconvex NLP. For the global optimization, tight bounds on the variables are required, and they are defined by physical inspection of the plant configuration (global minimization and maximization with different variables as objective function) or basic mathematical logics (such as conversion rate or composition of contents which should be in the range of 0–100 or 0–1, respectively). An upgrading plant that can produce multiple bitumen and SCO products is investigated. To show that the proposed model can be used for industrial purposes, three different cases are considered and the results show that the proposed approach is effective for the upgrading plant operations optimization.

The proposed modeling and optimization framework can be extended for multiperiod production planning and scheduling of the oil sands upgrading plant. Furthermore, the model of thermocracking unit can also be incorporated if its experimental data is available so both types of upgrading plants can be studied at the same time. In addition, to improve the decision robustness considering the inaccuracy or uncertainty in the correlation modeling, we can include uncertainty into the correlation model to reflect the modeling inaccuracy, and construct an uncertain optimization problem. The problem will be solved through uncertainty optimization technique such as robust optimization or stochastic optimization. Optimization under uncertainty of the utility costs or selling production costs is also meaningful for future work.

Supporting Information Empirical correlation models for the major process units of bitumen upgrading plant.

Author Information *Email: [email protected]; Tel: 1-780-492-1107; Fax: 1-780-492-2881 (Z. Li)

Acknowledgements The authors gratefully acknowledge the financial support from Alberta Innovates - Technology Futures (AITF) and The Natural Sciences and Engineering Research Council of Canada (NSERC).


Page 27 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

Nomenclature (i)

(ii)

(iii)

(iv)

Indices p

Intermediate product

s

Separation unit

u

Reaction unit

fp

Final product

ut

Utility

pr

Property

i

Stream (used in blender, mixers, splitters)

Sets P

NPH, LGO, HGO

S

VDU1, VDU2

U

NPH-hydrotreater, LGO-hydrotreater, HGO-hydrotreater, hydrocracker

FP

Sour SCO, Sweet SCO, SynDilBit, SynBit, DilBit

UT

Hydrogen, Steam, Electricity

PR

Yield, Specific gravity, Viscosity, Sulfur content, Nitrogen content

Parameters

α,β,k

Kinetic coefficients for HDN and HDS conversions

θ ,ϕ

Lower and upper bounds for volumetric blending ratio

γ

Sulfur distribution in the hydrocracker outlets

ap, bp, cp

Kinetic coefficients for sulfur conversion in the hydrocracker

a, b, c, d

Regressed coefficients of sulfur distribution in hydrocracker outlet

Feedpr

Feed specifications

Priceut

Utility price

Pricefp

Final product price

Volume

Reactor volume

Ratio

Conversion ratio of HDN to HDS

Specfp,pr

Final products specifications

Variables CR

Conversion rate


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

M_inp,u

Inlet mass flow rate

M_outp,u

Outlet mass flow rate

Mfp

Final products mass flow rate

V_inp,u

Inlet volume flow rate

V_outp,u

Outlet volume flow rate

Vfp

Final products volume flow rate

CVi

Volume composition

in out X pr ,i , X pr

(v)

Steam properties

Qu

Duty requirement

Wu

Work requirement

Hu

Hydrogen requirement

Tu

Temperature of operating unit

Pu

Pressure of operating unit

LHSVu

LHSV of operating unit

ECPs,p

Cut point range

BI

Blending index

CV

Volume compositions

Superscripts HDS

Hydrodesulfurization

HDN

Hydrodenitrogenation

Residue

Residue from VDU

Steam

Required steam in ADU and VDUs


Page 28 of 31

Page 29 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

References 1. Ordorica-Garcia, G.; Croiset, E.; Douglas, P.; Elkamel, A.; Gupta, M., Modeling the Energy Demands and Greenhouse Gas Emissions of the Canadian Oil Sands Industry. Energy & Fuels 2007, 21, (4), 2098-2111. 2. Crude Oil: Forecast, Markets & Transportation; Canadian Association of Petroleum Producers: June 2015. 3. Charry-Sanchez, J.; Betancourt-Torcat, A.; Ricardez-Sandoval, L., An optimization energy model for the upgrading processes of Canadian unconventional oil. Energy 2014, 68, 629-643. 4. Ordorica-Garcia, G.; Elkamel, A.; Douglas, P. L.; Croiset, E.; Gupta, M., Energy optimization model with CO2-emission constraints for the Canadian oil sands industry. Energy and Fuels 2008, 22, (4), 26602670. 5. Betancourt-Torcat, A.; Gutierrez, G.; Elkamel, A.; Ricardez-Sandoval, L., Integrated Energy Optimization Model for Oil Sands Operations. Industrial & Engineering Chemistry Research 2011, 50, (22), 12641-12663. 6. Betancourt-Torcat, A.; Almansoori, A.; Elkamel, A.; Ricardez-Sandoval, L., Stochastic Modeling of the Oil Sands Operations under Greenhouse Gas Emission Restrictions and Water Management. Energy & Fuels 2013, 27, (9), 5559-5578. 7. Gomes, U.; Patil, B.; Betancourt-Torcat, A.; Ricardez-Sandoval, L., Optimal Infrastructure of the Upgrading Operations in the Oil Sands under Uncertainty: A Multiscenario MINLP Approach. Industrial & Engineering Chemistry Research 2014, 53, (42), 16406-16424. 8. Leiras, A.; Ribas, G.; Hamacher, S.; Elkamel, A., Literature review of oil refineries planning under uncertainty. Int. J. Oil, Gas and Coal Technology 2011, 4, (2). 9. Shah, N. K.; Li, Z.; Ierapetritou, M. G., Petroleum refining operations: Key issues, advances, and opportunities. Industrial and Engineering Chemistry Research 2011, 50, (3), 1161-1170. 10. Pinto, J. M.; Joly, M.; Moro, L. F. L., Planning and scheduling models for refinery operations. Computers & Chemical Engineering 2000, 24, (9–10), 2259-2276. 11. Neiro, S. M. S.; Pinto, J. M., A general modeling framework for the operational planning of petroleum supply chains. Computers and Chemical Engineering 2004, 28, (6-7), 871-896. 12. Neiro, S. M. S.; Pinto, J. M., Multiperiod optimization for production planning of petroleum refineries. Chemical Engineering Communications 2005, 192, (1-3), 62-88. 13. Alhajri, I.; Elkamel, A.; Albahri, T.; Douglas, P. L., A nonlinear programming model for refinery planning and optimisation with rigorous process models and product quality specifications. Int. J. of Oil, Gas and Coal Technology 2008 1, (3), 283-307. 14. Guerra, O. J.; Le Roux, G. A. C., Improvements in Petroleum Refinery Planning: 1. Formulation of Process Models. Industrial & Engineering Chemistry Research 2011, 50, (23), 13403-13418. 15. Muñoz, J. A. D.; Aguilar, R.; Castañeda, L. C.; Ancheyta, J., Comparison of Correlations for Estimating Product Yields from Delayed Coking. Energy & Fuels 2013, 27, (11), 7179-7190. 16. Alattas, A. M.; Grossmann, I. E.; Palou-Rivera, I., Integration of nonlinear crude distillation unit models in refinery planning optimization. Industrial and Engineering Chemistry Research 2011, 50, (11), 6860-6870. 17. Menezes, B. C.; Kelly, J. D.; Grossmann, I. E., Improved swing-cut modeling for planning and scheduling of oil-refinery distillation units. Industrial and Engineering Chemistry Research 2013, 52, (51), 18324-18333. 18. Gueddar, T.; Dua, V., Disaggregation-aggregation based model reduction for refinery-wide optimization. Computers and Chemical Engineering 2011, 35, (9), 1838-1856. 19. Gueddar, T.; Dua, V., Novel model reduction techniques for refinery-wide energy optimisation. Applied Energy 2012, 89, (1), 117-126.


Energy & Fuels

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

20. Reddy, P. C. P.; Karimi, I. A.; Srinivasan, R., Novel solution approach for optimizing crude oil operations. AIChE Journal 2004, 50, (6), 1177-1197. 21. Méndez, C. A.; Grossmann, I. E.; Harjunkoski, I.; Kaboré, P., A simultaneous optimization approach for off-line blending and scheduling of oil-refinery operations. Computers and Chemical Engineering 2006, 30, (4), 614-634. 22. Li, J.; Karimi, I. A.; Srinivasan, R., Recipe determination and scheduling of gasoline blending operations. AIChE Journal 2010, 56, (2), 441-465. 23. Leiras, A.; Hamacher, S.; Elkamel, A., Petroleum refinery operational planning using robust optimization. Engineering Optimization 2010, 42, (12), 1119-1131. 24. Guerra, O. J.; Le Roux, G. A. C., Improvements in Petroleum Refinery Planning: 2. Case Studies. Industrial & Engineering Chemistry Research 2011, 50, (23), 13419-13426. 25. Gemayel, J. E. Integration and simulation of a bitumen upgrading facility and an IGCC process with carbon capture. University of Ottawa, Ottawa, 2012. 26. Frandsen, A.; Hogan, D.; McCuskey, S. In Situ Bitumen Hydrocarbon Extraction and Upgrading; Chemical Engineering Department: Florida Institute of Technology, Melbourne, USA, , 2015. 27. HYSYS® 2004.2: Tutorials & Applications. Aspen Technology, Inc.: Cambridge, USA, 2005. 28. Castellanos Díaz, O.; Sánchez-Lemus, M. C.; Schoeggl, F. F.; Satyro, M. A.; Taylor, S. D.; Yarranton, H. W., Deep-Vacuum Fractionation of Heavy Oil and Bitumen, Part I: Apparatus and Standardized Procedure. Energy & Fuels 2014, 28, (5), 2857-2865. 29. Sánchez-Lemus, M. C.; Schoeggl, F.; Taylor, S. D.; Růžička, K.; Fulem, M.; Yarranton, H. W., DeepVacuum Fractionation of Heavy Oil and Bitumen, Part II: Interconversion Method. Energy & Fuels 2014, 28, (5), 2866-2873. 30. FAQ: Modeling Heavy Oils Using Aspen HYSYS®. Aspen Technology, Inc.: Burlington, Ontario, Canada, 2010. 31. Jump Start: Assay Management in Aspen HYSYS® Petroleum Refining Aspen Technology, Inc.: Burlington, Ontario, Canada, 2015. 32. Chen, J.; Munteanu, M. In Optimizing bitumen upgrading scheme - Modeling and simulation approach, 12AIChE - 2012 AIChE Spring Meeting and 8th Global Congress on Process Safety, Conference Proceedings, 2012; 2012. 33. Alvarez-Majmutov, A.; Chen, J.; Munteanu, M., Simulation of bitumen upgrading processes. Petroleum Technology Quarterly 2013, 18, (2), 37+39-43. 34. Yui, S.; Chan, E., Hydrogenation of Coker Naphtha with Nimo Catalyst. In Studies in Surface Science and Catalysis, Kevin, J. S.; Emerson, C. S., Eds. Elsevier: 1992; Vol. Volume 73, pp 59-66. 35. Owusu-Boakye, A.; Dalai, A. K.; Ferdous, D.; Adjaye, J., Maximizing Aromatic Hydrogenation of Bitumen-Derived Light Gas Oil: Statistical Approach and Kinetic Studies. Energy & Fuels 2005, 19, (5), 1763-1773. 36. Ferdous, D.; Dalai, A. K.; Adjaye, J., Hydrodenitrogenation and Hydrodesulphurization of Heavy Gas Oil Using NiMo/Al2O3 Catalyst Containing Phosphorus: Experimental and Kinetic Studies. The Canadian Journal of Chemical Engineering 2005, 83, (5), 855-864. 37. Ferdous, D.; Dalai, A. K.; Adjaye, J., Hydrodenitrogenation and Hydrodesulfurization of Heavy Gas Oil Using NiMo/Al2O3 Catalyst Containing Boron: Experimental and Kinetic Studies. Industrial & Engineering Chemistry Research 2006, 45, (2), 544-552. 38. Gray, M. R., Upgrading Oilsands Bitumen and Heavy Oil. The University of Alberta Press: Edmonton, Alberta, Canada, 2015. 39. Centeno, G.; Sánchez-Reyna, G.; Ancheyta, J.; Muñoz, J. A. D.; Cardona, N., Testing various mixing rules for calculation of viscosity of petroleum blends. Fuel 2011, 90, (12), 3561-3570.


Page 30 of 31

Page 31 of 31

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Energy & Fuels

40. crudemonitor.ca: maintained by Crude Quality Inc, Canadian Association of Petroleum Producers (CAPP) and Small Explorers and Producers Association of Canada (SEPAC) as stakeholders. http://www.crudemonitor.ca/home.php. 41. Costa, A.; Bakhtiari, B.; Schuster, S.; Paris, J., Integration of absorption heat pumps in a Kraft pulp process for enhanced energy efficiency. Energy 2009, 34, (3), 254-260. 42. Brooke, A.; Kendrick, D.; Meeraus, A.; Raman, R., GAMS: A user’s guide, release 2.50. GAMS Development Corporation: 1988. 43. Sahinidis, N., BARON: A general purpose global optimization software package. Journal of Global Optimization 1996, 8, (2), 201-205.


Modeling and Optimization of the Upgrading and Blending Operations

Recommend Documents