RefiningNZ Rigorous Long Residue Inventory Modelling
Marcos Monguzzi, RefiningNZ, Commercial Programmer Strategy & Supply Optimisation
Charles Taylor, Honeywell, Senior Consultant
2
Presentation Overview
• New Zealand Refining Company Profile
• Role Played by RPMS for Refinery and Processors
• Business need for Long Residue inventory modelling requirements
• Process Overview
• Inventory Modelling Approach
• Problems identified and their resolution
• Business impact / benefits
3
Company Profile
• is New Zealand’s only oil refinery
• JV Tolling 120kbbl/d refinery
• Customers-Shareholders: BP, Chevron, Mobil, Z Energy
• Our goal is to be NZ Oil products Supplier of choice, by being the most competitive source of supply- in terms of reliability, cost and environmental footprint
• We produce nearly 74% of NZ Oil products requirements
• We implemented RPMS and Assay2 in 2009 to replace GEMMS, together with custom tools to assist with our business model
4
Company Profile
Feedstock purchase & delivery to Refining NZ
Imported
product
purchase &
delivery
International markets
26%
Product coastal
distribution
74%
Product
pipeline to Wiri
Exports &
bunker fuel
Refining
44%45% 5%
Wiri terminal Truck loading
International companies
Coastal terminals
New Zealand
supply chain
41.2 Mbbls
NZ Oil companiesRefining NZLegend:
6%
Notes: - Values and percentages are actual Refining NZ production and market estimates for 2011.
- Values include petrol, diesel, jet fuel & kerosene, fuel oil and roading bitumen.
35% 4%35%26%
New Zealand product market (~50.5 Mbbls)
Petrol (40%), Jet fuel (16%), Diesel (36%), Fuel oil (6%), Bitumen (2%)
5
RPMS Use in the Planning Process
• Multi-period model (3*2months) including Inventories, Used by the refinery and the four customers for Manufacturing program submission and acceptance (RNZ=Σ Customers):
– Refining NZ
– Chevron NZ
– BP NZ
– Mobil Oil NZ
– Z Energy
• Used by the customers for crude ranking and purchasing (Multi-Case, Single or Multi-Period models)
• Used for Crude and Production Allocation to individual customers (Single Case, Single Period)
6
Business Need for Long Residue Inventory Modelling
• Market constrained by excess fuel oil production. Excess FO production is exported at significant discount due to high freight costs
• Customer may be asked, or opt, to build Long Residue in storage for later processing to offset processing of light crudes or downstream unit shutdowns
• Average intake quality varies from period to period, resulting in significant differences to Atmospheric Residue quality
Gap Identified in migration project from GEMMS to RPMS
Business needs Long Residue inventory qualities to be representative of the crudes from which it was derived
Not a standard functionality in RPMS and hence involved Honeywell APS CoE
7
Crude and Vacuum Tower Configuration
Complex processing:
•2 Physical Crude towers
•2 Physical Vacuum towers
•Bitumen mode
•Atmospheric residue is
directed to VDUs for
further distillation or it
can be reprocessed via
CDU to satisfy column liquid loading while on lighter feeds
Crude
Distiller
1
Crude
Condensate &
Low Sulphur
Crudes
+ Slops
Naphtha Minus:
75-185º CP
Liquid Mass
Rate 3126t/d
Density 0.711
Kerosene:
185-250º CP
Liquid Mass
Rate 2052t/d
Density 0.811
Light Gasoil:
250-300º CP
Liquid Mass
Rate 1523t/d
Density 0.853
Heavy Gasoil:
300-350º CP
Liquid Mass
Rate 2088t/d
Density 0.868
Long Residue:
350º + CP
Liquid Mass
Rate 4180t/d
Density 0.931
S.W.
Max Intake Actual
13,000t/d
Imported / RNZ
Long Residue
T10
/T12 NHDT
U250
T16
/T12
KHDS U5500
T43
to 49 HDS3 U5800
T140
to 144 HV2 U6100
T122,
123,124
JET A1 5-25%
T24,
26,27,54
F.O.
8
HVU Structure
H2 from HMU
HCU
Hydro Cracker
Unit
Max Feed
4700t/d
70-80t/d H2 97%
Off Gas to
HMU
HVU1
High Vacuum
Unit
Max Intake
1100t/d
HVU2
High Vacuum
Unit
Max Intake
6000t/d
BBU
Bitumen
Blowing Unit
Max Feed
4500t/d
Long Residue from
CD1 & CD2
Imported Residue
from Storage
Long Residue from
CD1 & CD2
NZR Residue from
Storage (T140’s)
S.W.
Short
Residue
BDU
Butane
Deasphalting
Unit
Max Feed
2100 t/d
DAO
Tankage
Waxy
Distillat
e
C4
HC Tops
HC GO
HC Kero
S.W.
H2S
C4 From
Storage
S.W.
Asphalt
New
Route
Bitumen
Flash
Distillate
B45
T28
B180 T40
VGO to HDS3
Feed tankage
Tankage
Residue from HV1
on GP Mode
Bitumen Mode
(CD2 only)
General Purpose
Mode
Bitumen Feed from
Tankage and/or CD2
9
Long Residue Processing – Requirements - Example
• If residue quality is defined in CRDW tables (Assay library), changes to the quality and yields of it will not be accounted for on a multi period model
• Hence LP can create a ‘free’ upgrade in quality or yield for the LR
Period A Qty [kt] API %VGO %WAX %SR TAN S %w N2 ppm
Opening Stock Basrah Resid 40 30.3 6.24% 47.75% 46.01% 0.07 3.00% 3210
To refinery Basrah Resid -20 30.3 6.24% 47.75% 46.01% 0.07 3.00% 3210
from refinery Murban Resid 35 39.9 10.34% 65.52% 24.13% 0.12 1.20% 1285
closing stock Basrah/Murban Resid 55 36.4 8.85% 59.06% 32.09% 0.10 1.85% 1985
Period B Qty [kt] API %VGO %WAX %SR TAN S %w N2 ppm
Opening Stock Basrah/Murban Resid 55 36.41 8.85% 59.06% 32.09% 0.10 1.85% 1985
To refinery Basrah Resid -40 36.41 8.85% 59.06% 32.09% 0.10 1.85% 1985
from refinery Murban Resid 35 39.9 10.34% 65.52% 24.13% 0.12 1.20% 1285
closing stock Basrah/Murban Resid 50 38.6 9.80% 63.17% 27.03% 0.11 1.44% 1540
10
Long Residue Processing - Requirements
• Enable Long Residue (LR) to go to storage from multiple operating modes
• LR inventory required for Fuels mode and Bitumen operating modes
• To retain HVU yields and qualities for the LR in inventory, such that HVU yields and qualities are rigorously simulated when the LR is processed. Effectively: – retaining visibility of the crude processed from which the LR in storage
was generated,
– such that yields and qualities of HVU products when processing LR ex storage would be the same as when processing the LR generated directly from crude processed on the CDU
11
Long Residue Inventory Modelling Approach
Processing Straight Run LR Direct from CDU • Define HVU products as implicit products (implicit =
deferred cuts) • HVUs then explicitly defined as standard process
submodels • HVU yields derived from ‘implicit yields’ defined in
Assay – defined as new qualities in the assay
• HVU product qualities recursed in normal manner
12
Long Residue Inventory Modelling Approach
Processing LR ex Inventory • HVU yields also derived from ‘implicit yields’ defined in Assay (these
are recursed as properties through inventory)
• Determining the properties of the HVU product cuts for feed stock processed via inventory is not so easy….
• The product cut qualities are lost as soon as they go into inventory (or RPMS looses visibility of the Assay data)……..
13
Solving the inventory problem
• Solving inventory problem with implicit cuts: – Need to retain the individual implicit stream qualities through the
inventory blend – eg 1SD spg, sul etc. • Achieve this by modelling new set of properties for each implicit cut
for the Vacuum Unit:
• New qualities can be then be blended in the normal manner in inventory
• Individual implicit HVU yield qualities can then be calculated via customised recursion structure in the new HVU process models (Similar to current QPROPRT structure in Deasphalting / Platformer models).
14
Long Residue Inventory Modelling Approach
LR Inventory Blending • Blend equation for HVU yields defined as fraction of Long Residue
• Blend equation for new ‘Child’ or psuedo properties introduced for each HVU product stream and stream quality are determined as follows (example for spg of cut 1FW)
15
New ‘Pseudo’ Properties for LR inventory modelling
• Current model has 25 true properties • 72 new Pseudo properties for LR modelling structure – example for HVU#1 below
16
Model Over view
New HVU process
models
New CDU’s for
reprocessing
Atmos Resid
Atmos Resid
inventories- IV8, NZ2
17
Revised Long Residue Potential Modelling Concerns
Concerns Prior to Implementation
• Drastically increases the size of the matrix and slows down the convergence time.
• Step change in the number of properties represented in the model
• It may also exacerbate Local Optimum Phenomena.
• The user is forced to define properties of material in the inventory as defined in the Assay (different code names carried in the Parent Process Stream itself ). This is quite cumbersome.
18
Modelling Problems Endured
• Model convergence became a challenge
• Hence needed to audit model: – Tuning recursion tolerances
– General model settings
• Aligned model custom structures and constraints with best practices
• Development of a calculator to determine pseudo property values to be entered for Long Residue in inventory.
• Local Optima became an issue requiring implementation of ‘best practices’ to minimise occurrence
Handling Local Optima
20
Local Optima – back ground
• All LP models have local optima solutions,
• As users we are just not always aware that they are happening – only aware of local optima when you find one.
• Caused by:
– Non linearity of refinery models
– Use of penalties with high initial values can be a major cause of local optima
– Property tolerance setting have some influence as optimiser can stop searching (as satisfied tolerance criteria) before a better optimum has been found
– Bad starting point for initial estimated stream qualities
21
Minimising Local Optima
• Having a well tuned model will help to reduce local optima
• Use of SLPA (Successive Linear Programming) for complex model will limit the step size the solver can make in searching for an optimum solution. Hence preventing solver from ‘jumping over’ a better optimum solution
• Tighter property tolerances will help – having loose tolerance is equivalent to having a non converged solution. For non converged solutions ‘you do not really know where you are’
• Alignment of starting costs for penalties with crude price, such that optimiser considers breaking penalties in its search for an optimum solution
• Removing penalties from model can also help if users are able to resolve infeasibilities.
22
User Influence Over Local Optima #1
• Initial stream qualities are generally automatically calculated by RPMS; – Initial Crude stream qualities / yields are recalculated based upon
entries in T. CRCHG
• Initial estimates for intermediate blends can be updated in TRACKB
• Use of Case Stacking initiated from a good planning case helps
23
Updating T. TRACKB - #1
• T. TRACKB can be updated with the latest solution using RPMS “Copy Matched” functionality
• Step 1: Open T. TRACKB In the BASECASE model
• Step 2: Open T. #SQU in the solution file – USRAHOC.WRK
24
Updating T. TRACKB - #2
• Step 3: after selecting ‘File’, the user is presented with the popup box below. Navigate to the ‘WORK’ directory and open file USRAHOC1.WRK
• Step4: type in text ‘#SQU’ as shown and open this table
25
Updating T. TRACKB - #3
• Step5: Select Excel view option such that both tables #SQU and TRACKB are displayed side by side
• NB #SQU contains all the stream properties for the latest model solution.
• Step 6: select all contents of the sheet #SQU (rh side below)
26
Updating T. TRACKB - #4
• Step 7: From RDFactory menu select ‘Copy Matched’ function
27
Updating T. TRACKB - #5
• Step 7: From RDFactory menu select ‘Copy Matched’ function
• Step 8: Make sure that option ‘Ignore Empty Cells’ is selected. Select the destination file you wish to copy the data to and press ‘OK’
• If you forget to select t the ‘Ignore empty Cells option then close you Excel file and start again.
28
Updating T. TRACKB - #6
Contents of TRACKB will then be updated automatically. Then save and close T. TRACKB and continue with your model runs
29
Checking the Solution for Local Optima
• Investigate whether you can find a better solution by:
• Changing the starting point (T. TRACKB updates)
• Relax / tighten up constraining variables to see if model solution moves in the right direction.
30
Thank you for your attention