163 How linear ming helps to develop the Czech Refining Co. Hugo Kittel, Ph.D.. Strategy and Long Term Technical Development Manager Czech Refining Co. (CRC), O.Wichterleho 809, 278 52 Kralupy n.vlt. tel.+420 315 718306, e-mail hugo.kittel@crc.cz Introduction This paper discusses application of linear ming technique (LP) in refineries generally and describes the role of LP in management of the Czech Refining Co. (CRC) specifically. Selection of the refinery modelling tool is critical to the accuracy and execution efficiency. LP represents well proven practice in refineries. The main advantages are complexity, universality, accuracy, and quick response. LP model was officially instituted in the CRC ten years ago. Capability of this model has been improved progressively, in cooperation with software licensor and shareholders. Transformation of CRC to a join venture refinery in 2003 strengthened significantly role of LP modelling on operative level. Furthermore, new challenges for CRC related to production of clean fuels, soaring consumption of diesel, implementation of biocomponents, and request of shareholders to increase profit of CRC significantly fostered application of LP for searching response to these challenges and definition of potential new development alternatives. Some of these activities are listed in this paper. Implementation of linear ming in refineries A lot of different reasons exist why to model for optimization of refineries [ROMANOW, 2001]. Specialists claim that the great part of the efficiency increase reported by refineries in the last decade is due to application of modeling tools. Optimization of refinery operations, due to its complexity and huge economic facet, represents a really difficult task. Modeling the entire refinery accurately requires consideration of several thousands components, regardless toll implemented. Modeling tools vary considerably in sense of rigorousness, complexity, and methods based on - from spreadsheets to high fidelity models [NAGEL, GUERRA, 2002]. Selection of a tool is critical to the accuracy and execution efficiency. Linear ming technique (LP) can be characterized as follows [ANDERSON, SWEENEY, WILLIAMS, 2000; HARTMANN, 2002; TUCKER, 2001]: It was developed at the end of 40s to help managers make decisions. G.B.Dantzing is universally recognized as a father of LP he developed simplex algorithm. Utilization of LP has expanded adequately to growth of computers performance. 1333
Original LP models were non-recursed, no pooling considered, yield driven, with fixed stream properties. LP modeling progressed implementing non-linear distributive recursions technique, improving pooling (linkage between quality of upstream changes and downstream pool), embedding process simulation, and enhancing multi-period and multi- side capabilities. In consequence to recursion technique, process unit representation changed from multiple mode type yielded structure to base-delta representations. Local optimums or non-convergence can be problems now, resulting from intense application of recursion technique and can contribute to lack of confidence. A LP model is significantly influenced by initial and legacy model structure, follow-up maintenance, and modeler knowledge and experience. Accuracy depends on implemented model structure, regular updating of data, and repeated tuning The biggest models involve 7000 rows and to 600 recursions now. Development is focused to provide more accuracy, realism, plug-and-play features (to make easier modification of model), and user s friendliness. More commercial systems are available. For the review of the main pros and cons of LP modeling of refineries see the exhibit 1: Exhibit 1: Pros and cons of LP modeling of refineries Pros Cons Well proven and generally accepted technique by refiners Linearity: Facilitates constructing complex and quickly Linearity. Many processes in refineries show strongly responding models of refineries non-linear features. Fortunately some non-linearity can Complexity, as concerns number of units, type of operations, and feature of data involved. It allows really entire refinery modeling: Universality, as concerns scope of issues solved Versatility; material balances (inclusive refinery wide systems - hydrogen balance, fuel gas supply, and sulfur balance) and economics (GRM, cost of utilities etc.) are parallel outputs. It gives very quick response, in seconds Multi-sites and multi-period models can be developed It plays role of a rule setter in joint venture refineries be successfully mastered Due to complexity optimum is often flat; existence of sub-optimums cannot be excluded Data intensive approach, in consequence to linearity and complexity High quality data requested Tough analysis of results, more time demanding than calculation itself Limited predictive capability As concerns implementation of LP in refineries, following point are stressed in the literature [BRUNO, HILEMAN, 2005; HARTMAN, 2004; RAGHAV and others, 2006; SAHDEV, M.K.; JAIN, K.K.; SRIVASTAVA, P.; SLOLEY, FRASER, 2003; TUCKER. 2001; VALLEUR, M.; GRUE, J.L, 2004]: Each refinery LP model situation is unique. LP model represents particular refining configuration and involves economic drivers, availability of feed (crude diet) and utilities, 1334
logistical operational constraints, process unit simulation, product blending, and product portfolio optimization. Preparation of data for LP models has been automated (for example transfer of crude assays into LP assays). Vendors crude assays can be utilized. Now has been added features allowing side-cut overlap, utilizing of multiple assays table, and assisting in calculation of properties. Swing cuts for cut point optimization were implemented, substituting multiple quality streams. Multiple capacity constraints are considered for unit simulation. Commercial data, based on real operational results, are utilized rather than laboratory data now to adjust LP model. Exploited model needs to be trusted. Therefore, to have real values for a refinery, LP model must be accurate, regularly up-dated (yields, stream properties), well documented, and properly filed. Successful implementation requires well organized communication and cooperation across a company, with partners, technical service providers, and other external institutions to gather all necessary data. Usually GRM is maximized as an objective for modeling of refineries. However, objective function can be set differently, for example to minimize operative costs, maximize performance of selected units, etc. Simulators help to generate shift vectors for scenarios where the refinery has little or no operating data, what is typical situation for development projects. However these simulators must be tuned somewhere as well. Exploiting of LP modeling can lead to hundred million CZK of annual profit resulting from right managerial decision. For application of LP modeling in refineries exist following important areas: Planning for different time horizons, e.g. evaluation of future performance of existing assets, fulfilling production premises of owners. Backcasting, e.g. comparing past performance and allocating product to Processors, which is in parallel very important activity for continuous improvement of LP model s accuracy. Rationalization activities, e.g. looking for new opportunities in the frame of existing technological scheme (so called what if cases ) - alternative streams, diverse modes of operation representing new practices, several blending strategies, variant product slates, improved process unit efficiency, capacity improvements (revamps and / or decommissioning), alternate pricing scenarios, etc. Development projects, first of all assessing impact of new process units (often called case / configuration studies ) or radically new product specifications (clean fuels). Assessing benefits of rationalization and development activities, it is important to prepare a robust base case. 1335
CRC practice in linear ming In the Kralupy refinery a simple LP model was available already in 1974 as part of software package of IBM 1800 control computer. CRC has started to utilize the LP technique intensely and professionally in 1996, in connection to its privatization. Change to a joint venture refinery mode in 2003 has significantly strengthened the role of LP modeling for management of CRC. Exhibit 2: The important inputs and outputs of CRC LP models Crucial information Inputs Outputs related to LP modeling Crude oil assays and availability of crude Volume of feeds illustrates exhibit 2. The oils Internal transfers between refineries main current applications Logistical constraints Data book, describing performance of Balance of products Consumption of utilities of LP modeling in CRC are as per exhibit 3. CRC technological units Variable costs developed and exploits Availability of units maintenance plan GRM multi-sites LP model Material premises volume and quality two refineries or three of products refinery complexes are Financial premises exchange rates, price of crude oils and refinery products embraced in one LP model, to simulate actual operation of the Company. It exploits multi-period capability of software as well. Strategic Plan and Development Projects represent a specific example of LP model application, because new processes or practices are thoroughly investigated as target of modeling, which is not standard and even more not allowed practice in operative planning. For this specific purpose so called CRC Development LP Model (DLP) has been created in the last three years, to support strategic decisions and to help develop new ideas. The features of this model can be characterized as follows: It involves promising technologies, not available in CRC now (alkylation, steam reformer, hydrotreating of heavy or residual streams, delayed coking, etc.) It implements new technological regimes, catalysts, and additives not exploited in current operations (for example high conversion mode of Visbreaking, ZSM-5 / GSR / Desox additives for FCC). Capacities need not to strictly comply with approved data-book; in some cases even more can go above possible revamp of units. Streams can be routed differently. Crude oils diet on input to LP model is reduced however new crude oils can be tested. Other feeds list is reduced however new feeds can be considered (bioethanol, RME; external alkylate). Portfolio of products is reduced to the main ones, however new products can be considered (ETBE, alkylate, benzene, etc.). Seasonality is approximated by average properties (crude oils pour point) and products expected for different seasons are blended in parallel. Premises are defined directly as common for all Processors. Model is tightly joined with other CRC s managerial decision tools (Justification of new investments, computing benchmark indexes, etc.). 1336
Exhibit 3: Applications of LP modeling in CRC Application Characteristic Horizon / Period MANAGERIAL PLANNING Strategic Plan (SP) Development Projects (DP) Business Plan (BP) Operative Plan (OP) Allocation of products, backcasting (AP) Base document for developing of CRC Checks future possible exploitation of CRC s assets Implements mission and visions of CRC Reacts to challenges of changing oil industry environment Calculates benchmark indexes, comparing these indexes with goals set by shareholders, identifies gaps, and defines activities covering these gaps. Initialized to reflect the trends of the oil industry, to cover performance gaps, or to elaborate step change ideas. Need on additional data (mined from literature, research studies, test runs on existing units, or from dialogue with licensors and engineering companies) The problem solving potential of LP modeling favorably exploited Base document for managing of CRC Detailed material balance and financial results Asses impact of winter and summer quality of motor fuels and seasonal demand picks on performance of CRC OPERATIVE PLANNIG Obligatory document The identical LP model is utilized parallel by Processors, to define their individual feasible demand and by CRC, to prepare final consolidated solution. In special case not feasible request can be converted into feasible solution running joint model. Consolidation of premises for joint CRC LP model needs mastery in refinery operations. ALLOCATION Final obligatory allocation of materials and cost to shareholders Checks-up the LP model accuracy and gives historical review of CRC s performance LP model exploited simply as a material balance model. 5 years / one year Future oriented 3 10 years Future oriented SP as a measure 1 year / 1 month Future oriented SP as a measure Month / day Future oriented BP as a measure Month, year Backwards orient. OP and BP as measures Exhibit 4: Implementation of LP in CRC Activity Model type # of specialists SP, DP Development 0,5 BP Operative planning 0,5 OP Operative planning 3 AP Operative planning 4 Maintenance and upgrade All models in use 2 Total Two models 8 Exhibit 5: Characteristics of CRC Development LP Model Characteristic Number of row 2712 Number of columns 2951 No zero cells (%) 0.68 Number of recursion coefficients 877 LP models in CRC are developed, maintained, and run as described in the exhibit 4. From mathematical point of view CRC Development LP Model can be characterized as per exhibit 5: DLP has been exploited to solve following interesting issues: To assess implementation of bioethanol blended into mogas directly and / or in form of ETBE. Results of ETBE batch run have been utilized to populate the model with 1337
necessary data [KITTEL, 2002]. Under studied price scenario, direct blending of bioethanol and continuing in MTBE production has represented the best choice, e.g. different result has been achieved in comparison to common practice in Europe now. Furthermore, negative impact of bioethanol on blending value of isomerate has been highlighted. Similar study has been prepared for direct blending of RME into diesel. Significant negative impact on balance of kerosene, from kerosene derived JET production, and GRM, has been stated, all related to high density or purchase price of RME. To analyze competitive position of the Hydrocracking unit vs. the FCC complex within CRC for production of clean fuels. Revamp alternatives of both main CRC s conversion technologies, consuming partially similar type of feed, have been assessed considering all existing (reformate, isomerate, FCC gasoline, MTBE) and potential (bioethanol, ETBE, alkylate, RME) components of motor fuels. In scenario of soaring consumption of diesel, Hydrocracking unit revamp will deliver higher benefit [KITTEL, PELANT, 2004]. Continuing in task above, impacts of varying capacity and conversion of the Hydrocracking unit on production of motor fuels, maximal possible yield of diesel, and GRM have been deeply researched. Yields and quality shift vectors for LP model have been taken from an external feasibility study. Hydrocracking unit represents significant supplier of Chemopetrol s Steam Cracker unit (SCU) and maximization of diesel production reduces hydrowax delivery to SCU. There is a possible conflict of interest. Results of simulation, based on lots of LP model runs, have confirmed advantage of revamp (+25%), high conversion regime (from 55% to 73%) and delivered possible production of diesel. GRM from simulation has been utilized as base for justification of investment. The DLP was further exploited to assess: Possible implementation of alkylation technology Revamp of existing MTBE unit Increase propylene production in FCC unit Desulphurization of feed for FCC now, which represents the most sophisticated and complex application of the DLP in CRC till now. The most illustrative example of DLP role represents CRC s solution for Clean 10 ppm S. This project started in CRC in year 2000. Original proposal, based on a 3-cut splitter (3CS) of FCC gasoline and next processing of cuts by already existing technologies, was later identified as not optimal for full 10 ppm S mogas production scenario. Therefore CRC re-initialized the project during 2003. Answers after following issues were looked: Which will be an optimal technological scheme? Which will be the role of 3CS in this scheme? Should CRC start the construction activities on 3CS or to stop the investment? Will be an additional Selective desulphurization (SDS) needed? If yes, which will be the best SDS technology for CRC? Which will be an optimal SDS capacity? 1338
Phasing of change (Two steps 50 and 10 ppm S or target directly 10 ppm S in mogas)? Which will be related risks? FCC complex, commissioned in 2001 in the Kralupy refinery, processes mixture of vacuum distillates and long residue from sweet crude oils. Feed is not pretreated. Original desulphurization scheme of FCC gasoline illustrates the exhibit 6. Exhibit 6: Simplified original scheme of FCC mogas desulphurization MEROX LPG, C3= Originally, heavy gasoline was hydrotreated unselectively, then re-blended and full gasoline stream meroxed utilizing non extractive mode. This mode fit neither 50 ppm S scenario nor 10 ppm S. HDS To meet mandated term for 50 ppm S mogas (January 1, 2005), 3CS construction was only 2500 ppm S 340 ppm S viable solution, starting in 2003. Scheme with 3CS illustrates the exhibit 7. Interesting part of this solution, exploited in CRC now, represents routing of FCC heart cut gasoline to Naphtha Hydrotreater (NHT). Reactor of NHT unit was equipped with a new liquid quench, to manage highly exothermic character of reactions. Facility was commissioned on the end of 2004, see the exhibit 8. The role of DLP was to assess impact of this scheme on mogas production and quality, to calculate volume of 10 ppm S mogas available from this solution, and to delivers arguments to justify the investment. Main fractionator 180 ppm S Exhibit 7: Scheme with 3-cut splitter 50 ppm S LPG, C3= MEROX-E 20 ppm S Main fractionator HDS 2500 ppm S 180 ppm S 340 ppm S 3-cut splitter 590 ppm S NHT Redistillation Steam cracker 1 ppm S SRR 180 ppm S 1339
Exhibit 8: CRC s 3-Cut Splitter To achieve full 10 ppm S production, three schemes have been further researched, applying the DLP: Scheme 1: Utilizing existing 3CS, adding new small SDS unit, relocating MEROX put into extractive mode, and relocating existing HDS unit with added new stabilization section, see the exhibit 9. Scheme 2: Utilizing 3CS as 2CS only, adding new SDS of moderate capacity, relocating MEROX put into extractive mode, and mothballing existing HDS unit, see the exhibit 10. Scheme 3: Adding new full range SDS unit, mothballing new 3CS, existing HDS unit, and MEROX (or alternatively use it downstream of the new SDS, to remove recombinant mercaptans), see exhibit 11. Exhibit 9: 10 ppm S, scheme 1 utilizing 3-cut splitter The role of DLP was clear check feasibility, calculate material balance of different schemes, deliver economical impacts (GRM). LPG, C3= 15 ppm S MEROX-E Main fractionator 3-cut splitter SDS 10 ppm S HDS 15 ppm S Diesel 1340
Exhibit 10: 10 ppm S, scheme 2 utilizing 3-cut splitter as 2-cut splitter LPG, C3= 15 ppm S MEROX-E 10 ppm S Main fractionator 2-cut splitter SDS Exhibit 11: 10 ppm S, scheme 3 not utilizing 3-cut splitter LPG, C3= Main fractionator SDS 10 ppm S Four technologies have been assessed in all three schemes, utilizing technological data from bids of licensors and CAPEX normalized according to list of equipments. Feasibility have been checked, OPEX, GRM, NPV, and IRR have been calculated for each scheme and technology, based on DLP outputs. Individual schemes have been assessed as per exhibit 12 (for selected technology only). Based on these results: Scheme 1 has been selected finally as the most proper for CRC, due to the best IRR and advantages stated in the exhibit 12. Decision represents an apparent trade-off between CAPEX, OPEX (3CS is energy demanding), and blending value of components. Licensor for technology has been selected 1341
Exhibit 12: Selection of the best scheme of FCC gasoline desulphurization Scheme CAPEX OPEX NPV IRR Pros Cons 1 100% 100% 100% Base Utilization of all already existing units Opportunity to optimize blending of FCC cuts and to increase diesel production Minimal CAPEX High complexity High OPEX, related to high complexity 2 111% 82% 107% -6% Moderate CAPEX Moderate OPEX Lower number of units in Regret investment into HDS operation 3 160% 109% 94% -62% Low complexity Low OPEX (however later not proven by simulation) High CAPEX Significant regret investments, related to the new 3CS, MEROX, and HDS unit Not proven scheme in the time of decision making A SDS unit, implemented according to the exhibit 9, will be commissioned in the middle of 2007. DLP has significantly contributed to give answer to all issues related to 10 ppm mogas production in CRC, see the exhibit 13: Exhibit 13: CRC response to 10 ppm S mogas Question Response Which will be an optimal technological scheme? The scheme 1, as per exhibit 9 The role of 3CS in this scheme? Start the construction activities or cancel the investment? 3CS exploited as a fast track solution for 50 ppm S mogas production, it will continue in operation under 10 ppm S mogas scenario as well Will be an additional SDS needed? Yes, definitely! If yes, which will be the best SDS technology for CRC? It was selected based on DLP results Calculation of the optimal SDS capacity? Will be related to the selected scheme 1 Timing of change (Two steps 50 and 10 ppm S or target Two steps, 3CS already in operation, SDS will be directly 10 ppm S in )? Which will be related risks? Conclusion commissioned by 2007 Minimal; mandated quality achieved, existing units will continue in operation, and CAPEX delivered in level acceptable for owners LP modeling in CRC is exploited intensely for different types of activities. Separate operative planning and development models have been worked out in the past few years, reflecting first of all specialized request on LP modelling after transformation of CRC to the joint venture refinery. CRC s Development LP model has been applied to assess impact of the main challenges for CRC for example biofuels, increasing demand of diesel, and production of clean mogas. Model represented the central tool applied to identify for CRC the best technological scheme of FCC gasoline desulphurization and select the best technology. The scheme including new 3-cut splitter, new SDS unit, relocation and reconstruction of two other units will be fully realized in the middle of 2007. 1342
References 1. ANDERSON, David R.; SWEENEY, Dennis J.; WILLIAMS, Thomas A.: "An introduction to Management Science". South-Western College Publishing, Cincinnati, 2000. ISBN 0-324-0321-8 2. BRUNO, Joe; HILEMAN, Michael J.: "Canadian refiner uses improvement to increase refining profits". Oil Gas Journal 2005/103/09/56 3. HARTMANN, J.C.M.: "Planning models for joint venture refineries function as a contract". Hydrocarbon Processing 2002/81/08/13 4. HARTMAN, J.C.M.: "Ramp up LP models with audits and transparent reporting methods". Hydrocarbon Processing 2004/83/09/91 5. JÄNICKE, W.: "Computergestützte Produktionsplannung für Erdölraffinerien". Erdöl und Kohle 1992/108/05/225 6. KITTEL, Hugo: "Experience with production of ETBE in CRC". Preprints of the International Conference Motor fuels 2002. Vyhne, Slovakia, 17-20.6.2002 (in Czech) 7. KITTEL, Hugo; PELANT, Pavel: "Hydrocracking versus Fluid Catalytic Cracking for production of clean fuels". Preprint of the International CHISA Conference 2004, Prague 2004-08-24/28. 8. NAGEL, Ewe; GUERRA, Maria, Jesus: "Rigorous modeling". Hydrocarbon Engineering 2002/07/03/69 9. RAGHAV, O.P. and others.: "Refinery LP modeling". Petroleum Technology Quarterly, 2006/11/Q1/95 10. ROMANOW, Stephany: "What is optimization for the HPI?". Hydrocarbon Processing 2001/80/06/11 11. SAHDEV, M.K.; JAIN, K.K.; SRIVASTAVA, P.: "Petroleum Refinery Planning and Optimization Using Linear Programming". http://www.cheresources.com/refinery_planning_optimization.shtml 12. SLOLEY, Andrew; FRASER, Alastair "Maximizing benefits". Hydrocarbon Engineering 2003/08/03/65 13. TUCKER, Michael A: "LP Modeling - Past, Present and Future". NPRA 2001 Annual Meeting. March 18-20, 2001, Marriott Hotel, New Orleans, L.A. 14. VALLEUR, M.; GRUE, J.L.: "Optimize short-term refinery scheduling". Hydrocarbon Processing 2004/83/06/46 Index of abbreviations 2CS 3CS AP BP CAPEX CRC DLP DP ETBE FCC GRM HDS IRR LP MEROX-E MTBE NHT NPV OP OPEX RME SCU SP SDS 2-Cut-Splitter 3-Cut-Splitter Allocation of Products Business Plan Capital Expenditures Czech Refining Company CRC s Development LP Model Development Project Ethyl Tertiary Butyl Ether Fluid Catalytic Cracking Gross Refinery Margin Hydrodesulphurization Internal Rate of Return Linear Programming Mercaptane Oxidation in Extractive mode Methyl Tertiary Butyl Ether Naphtha Hydrotreater Net Present Value Operative Plan Operating Expenses Rape-seed Methyl Ester Steam Cracker Unit Strategic Plan Selective Desulphurization 1343