Enterprise-Wide Optimization for Operations of Crude-Oil Refineries: closing the procurement and scheduling gap

Brenno C. MenezesPostDoc Research Scholar

Carnegie Mellon University

Pittsburgh, PA, US

Jeffrey D. Kelly

Owner

industri@lgorithms

Toronto, ON, Canada

Enterprise-Wide Optimization for Operations of Crude-Oil

Refineries: closing the procurement and scheduling gap

EWO, Pittsburgh 22nd Sep, Berlin 20th Oct, Barcelona Nov 7th, Salamanca Nov 10th, 2016.

Ignacio E. Grossmann

R. R. Dean Professor of Chemical Engineering

Carnegie Mellon University

Pittsburgh, PA, US

Faramroze Engineer

Senior Consultant

SK-Innovation

Seoul, South Korea

1

- Quantitative Analysis on Opportunities for Planning and Scheduling Integration

- Shows the Scheduling Problem Developed for SK-Energy (3rd biggest refinery among 635)

Procurement

Scheduling

Close thedecision-making GAP

in terms of crude-oil quality

month

week

2

Outline

1- Decision-Making in the Crude-Oil Refining: Identify the Enterprise Optimization

3- Enterprise-Wide Optimization (EWO) for Operations of Crude-Oil Refineries

2- Planning and Scheduling Integration: Opportunities for EWO

Challenges

Commercial and Academic Approaches

4- Whole Crude-Oil Scheduling: from crude unloading to product delivery

Feedstock Storage Assignment (MILP) FOCAPO

Crude-Oil Blend Scheduling (MILP+NLP) FOCAPO

Key Aspects in the Modeling and Solution IMPL

5- Conclusions/Next Steps on Scheduling

MINLPRelax y [0,1]as (0,1) in NLP

Re-Planning; Planning + Feedback from Scheduling

Off-Line and On-line Scheduling: Re-Scheduling

3

Time

Supply Chain

Refinery

Process Unit

second hour day month year

RTOControl

on-line off-line

Operational Planning

Tactical Planning

Strategic Planning

Simulation

LP Optimization

LP Optimization

Operational Corporate

week

Decision-Making in the Oil-Refining Industry: Challenges

SchedulingNLP

NLP

SQP

SLP

MILP+NLP SLP

LP

PIMS-AO

Spiral Plan

Production Scheduler

MILP+NLP SLPQPE

Structure(space)

EWO Meeting, Sep 22nd, 2016.

Commercial Solutions

4

Time

Supply Chain

Refinery

Process Unit


RTOControl

on-line off-line


Tactical Planning

Strategic Planning


week

1st: Logic variables (MILP)

2nd: Nonlinear Models (NLP)Optimization

(MILP and NLP)

ExpansionInstallation

Modes of operation

MaintenanceCleaning (Decoking)Catalyst Change

Integration

time

space

Scheduling

(Menezes, Kelly, Grossmann & Vazacopoulos 2014)

(Menezes, Kelly & Grossmann, 2015)Simulation

Decision-Making in the Oil-Refining Industry: Challenges

Integration

Structure(space)

(unmeasured or off-line)

(measured or on-line)


Academic Approaches

Scheduling

5

Time

Supply Chain

Refinery

Process Unit


RTOControl

on-line off-line


Tactical Planning


week

1st: Crude Procurement (NLP)

2nd: Modes of Operation (NLP)

Planning and Scheduling Integration: EWO Opportunities

3rd: Ship Scheduling (MILP)

Storage Assignment (MILP)

Crude Blend Scheduling (MILP+NLP)

time

Integration in time (unmeasured or off-line)

Structure(space)

Crude Basket

Planned Operations

Procurement

Scheduling

Close the GAP


Scheduling

6

Time

Supply Chain

Refinery

Process Unit


RTOControl

on-line off-line

Tactical Planning


week

1st: Crude Procurement (NLP)

2nd: Modes of Operation (NLP)

3rd: Ship Scheduling (MILP)

Storage Assignment (MILP)


time

Integration in time (unmeasured or off-line)

Structure(space)

Planned Operations

Procurement

Scheduling

Close the GAP


Crude Basket


Planning and Scheduling Integration: Re-Plan + Feedback

Re-Planning to reduce the crude basket and operations from 1 month to 4 weeks

modes of operationCampaign (1 month)

EWO for Operations of Crude-Oil Refineries: Old Fashion Planning


Process Units

FuelFarms

Raw Material

Crude basket

Instrumentation, Advanced Process Control and RTO

Scheduling

Tactical Planning

on-line

off-line

model data

Crude

Procurement Crude-to-Fuel TransformationFuel

Sales

1m 2m 3m

3m

model data

model data

(NLP)

(NLP)

(3 months)

Orders (decisions): Feedforward


Tactical:

1m

Operational:

m = month

Scheduling:

collaboration


EWO for Operations of Crude-Oil Refineries: Proposed Re-Planning

Week for crude arrival


Process Units

FuelFarms

Raw Material

Crude basket


Scheduling

Tactical Planning

on-line

off-line

model data

Crude


Sales

1m 2m 3m

1w 2w 3w 4w

blocked production

3m

model data

model data

(NLP)

(NLP)

(3 months)Crude basket4 weeks+ 2m

(1st weekin days/shifts + 3 weeks)

Ship Rearrangements

Orders (decisions): Feedforward Key Indicators (performance): Feedback

(4 weeks) (4 weeks)


4w+2m 4w

Tactical: Operational:

7 days 21 shifts

m = month

w = week

4w

Scheduling:

collaboration


EWO: Re-Planning + Feedback from Scheduling



Process Units

FuelFarms

Raw Material

Crude basket


Scheduling

Tactical Planning

on-line

off-line

model data

Crude


Sales

1m 2m 3m

1w 2w 3w 4w

2h 4h 166h 168h

blocked production

FSA

CBS

model data

model data

SHS = Ship Scheduling

(NLP)

(NLP)


Ship Rearrangements


(4 weeks) (4 weeks)

(MILP)

(MILP+NLP)

SK example : 7 days/2h time-step


Conclusions:

• Reduce the crude-oil procurement decision from amonth to a week

• Feedback from Scheduling for detailed Planning within time-steps of days/shifts


FSA = Feedstock Storage Assignment

collaboration


EWO for Operations of Crude-Oil Refineries: Off-Line Scheduling



Process Units

FuelFarms

Raw Material

Crude basket


Scheduling

Tactical Planning

on-line

off-line

model data

Crude


Sales

1m 2m 3m

1w 2w 3w 4w

1d 2d 29d 30d

2h 4h 166h 168h

blocked production

SHS

FSA

CBS

model data

model data

SHS = Ship Scheduling

CBS = Crude-oil Blend Scheduling

(NLP)

(NLP)


CDU IBP/FBP

Ship Rearrangements


(4 weeks) (4 weeks)

(MILP)

(MILP)

(MILP+NLP)

SK example : 7 days/2h time-step



(NLP) = 100 K to 0.5 million (MILP) = 30 K to 50 K

Number of Variables:


13

Scheduling

Structure

Time

Supply Chain

Refinery

Process Unit


RTOControl

on-line off-line

week

space

Measured updates/feedback:a) Fill-Draw delays in Storage Tanksb) Data Rec in feed tank real composition and properties

SK example developed in SK-Innovation Internship: 7 days/2h time-step in Discrete-Time

Integration in space (measured or on-line)

Re-SchedulingOn-line SchedulingReal-time SchedulingData-Analysis with Feedback

Data-Analytics + Decision-Making:a) Distillation Blending and Cutpoint Opt. (ASTM D86 => TBP + flows)

b) Data-Driven RTO (LP to relate the majors pairs of Depend./Independ. Variables): SSD, SSDR, SSGE, SSGO

EWO for Operations of Crude-Oil Refineries: On-Line Scheduling

Crude Transferring

Refinery Units Fuel Deliveries

Fuel Blending

Crude Dieting

Crude Receiving

Hydrocarbon Flow

FCC

DHT

NHT

KHT

REF

DC

B

L

E

N

SRFCC

Fuel gas

LPG

Naphtha

Gasoline

Kerosene

Diesel

Diluent

Fuel oil

Asphalt

Crude-Oil Management Crude-to-Fuel Transformation Blend-Shop

Charging or

Feed Tanks

Whole Scheduling: from Crude to Fuels Crude-Oil Scheduling Problem

Receiving or

Storage Tanks

Transferring or

Feedstock Tanks

VDU

1996: Lee, Pinto, Grossmann and Park (MILP), discrete-time

2004: Randy, Karimi and Srinivasan (MILP), continuous-time

2009: Mouret, Grossmann and Pestiaux: MILP+NLP continuous-time

2014: Castro and Grossmann: MINLP ; MILP+NLP, continuous-time2015: Cerda, Pautasso and Cafaro: MILP+NLP, continuous-time(336h: 14 days; binary ≈ 4,000; continuous ≈ 6,000; constraints ≈ 100K; CPU(s) ≈ 500)

14

2016 Goal: solve the SK Ulsan refinery scheduling for a week (38 crude, 2 pipelines, 23 storage tanks, 11 feed tanks, 5 CDUs)

1. JD Kelly, BC Menezes, IE Grossmann, F Engineer, 2017, FOCAPO.

2. JD Kelly, BC Menezes, F Engineer, IE Grossmann, 2017, FOCAPO.EWO Meeting, Sep 22nd, 2016.

MINLP -> MILP + NLP

MINLP Relax y [0,1]as (0,1) in NLP

Current Benchmark

DICOPT (5,000 binary variables)

Crude Transferring


Fuel Blending

Crude Dieting

Crude Receiving

Hydrocarbon Flow

FCC

DHT

NHT

KHT

REF

DC

B

L

E

N

SRFCC

Fuel gas

LPG

Naphtha

Gasoline

Kerosene

Diesel

Diluent

Fuel oil

Asphalt


Charging or

Feed Tanks

Whole Scheduling: from Crude-Oils to Fuels Crude-Oil Blend Scheduling Problem

Receiving or

Storage Tanks

Transferring or

Feedstock Tanks

FSA

VDU

(MILP+NLP)

PDH Decomposition (logistics + quality problems)

Includes logistics details

1996: Lee, Pinto, Grossmann and Park (MILP), discrete-time

2004: Randy, Karimi and Srinivasan (MILP), continuous-time

2009: Mouret, Grossmann and Pestiaux: MILP+NLP continuous-time

2014: Castro and Grossmann: MINLP ; MILP+NLP, continuous-time2015: Cerda, Pautasso and Cafaro: MILP+NLP, continuous-time(336h: 14 days; binary ≈ 4,000; continuous ≈ 6,000; constraints ≈ 100K; CPU(s) ≈ 500)

15

(MILP)

2016 Goal: solve the SK Ulsan refinery scheduling for a week (38 crude, 2 pipelines, 23 storage tanks, 11 feed tanks, 5 CDUs)

Minimize the Quality Variation Feedstocks -> Storage TanksReduces optimization search space for further scheduling

2nd Crude Blend Scheduling Optimization (CSBO)

Yields Rates (crude diet, fuel recipes, conversion)

(Menezes, Kelly & Grossmann, 2015)


2. JD Kelly, BC Menezes, F Engineer, IE Grossmann, 2017, FOCAPO.EWO Meeting, Sep 22nd, 2016.

MINLP -> MILP + NLP1st Feedstock Storage

Assignment (FSA)

FSA

CBSO

Crude Transferring


Fuel Blending

Crude Dieting

Crude Receiving

Hydrocarbon Flow

FCC

DHT

NHT

KHT

REF

DC

B

L

E

N

SRFCC

Fuel gas

LPG

Naphtha

Gasoline

Kerosene

Diesel

Diluent

Fuel oil

Asphalt


Charging or

Feed Tanks

Whole Scheduling: from Crude-Oils to Fuels Crude-Oil Blend Scheduling Problem

Receiving or

Storage Tanks

Transferring or

Feedstock Tanks

FSA

VDU

(MILP+NLP)

PDH Decomposition (logistics + quality problems)

Includes logistics details

17

(MILP)1st Feedstock Storage Assignment (FSA)

Minimize the Quality Variation Feedstocks -> Storage TanksReduces optimization search space for further scheduling

2nd Crude Blend Scheduling Optimization (CSBO)

Yields Rates (crude diet, fuel recipes, conversion)

(Menezes, Kelly & Grossmann, 2015)



2. JD Kelly, BC Menezes, F Engineer, IE Grossmann, 2017, FOCAPO.

CBSO

18

Crude-Oil Blend Scheduling: Illustrative Example

336h: 14 days discretized into 2-hour time-period durations (168 time-periods)

The logistics problem (MILP): ZMILP=695.68,333 continuous + 3,508 binary variables

3,957 equality and 15,810 inequality constraints

Non-Zeros: 59,225 ; Degrees-of-freedom: 7,884

CPU(s): 176.0 seconds / 8 threads in CPLEX 12.6.

The quality problem (NLP): ZNLP=701.919,400 continuous variables

14,862 equality and 696 inequality constraint

Non-Zeros: 26,430 ; Degrees-of-freedom: 4,538

CPU(s): 16.8 seconds in the IMPL’SLP

engine linked to CPLEX 12.6.

Crude blend scheduling(MILP+NLP)

Clustering (MILP)

MILP-NLP gap: 0.09% with only one PDH iteration.


IMPL (Industrial Modeling and Programming Language) using Intel Core i7 machine at 2.7 Hz with 16GB of RAM

SK Refinery Example

19

The logistics problem (MILP):33,937 continuous + 29,490 binary variables

8,612 equality and 72,368 inequality constraints

Non-Zeros: 531,204; Degrees-of-freedom: 63,427

CPU(s): 215 seconds (3.58 min) in 8 threads CPLEX 12.6.

The quality problem (NLP):143,316 continuous variables

88,539 equality and 516 inequality constraint

Non-Zeros: 138,812; Degrees-of-freedom: 54,788

CPU(s): 539 seconds (8.98 min) in the IMPL’SLP

engine linked to CPLEX 12.6.

MILP-NLP gap :12% to 10% with two PDH iteration.

Units: 5 CDUs in 9 modes of operation + 4 Blenders

Tanks: 35 among storage and feed tanks. 7 days: 168-hours discretized into 2-hour time-period durations (84 time-periods).


IMPL (Industrial Modeling and Programming Language) using Intel Core i7 machine at 2.7 Hz with 16GB of RAM

IMPL’s Pre-solver reduces non-zeros from 2 to 0.5 million

20

Key Aspects in the Modeling and Solution


3. Quality model1

a. Reformulation to reduce %NLP

and non-zeros (reduce number of

local solutions and their variance)

1. UOPSS

a. flows and setups with LO and UP

b. LO and UP for flows and yields

c. (Unit+Mode)=Mode m

d. Valid Cut: m-arrow-m’

2. Logistics model1

a. Fill-draw delay

b. Uptime for blenders and arrows

c. Multi-use

Flow

Volume-BasedProperties

Weight-BasedProperties

21

Key Aspects in IMPL’s computing skills

5. Complex Number to find derivatives and derivatives by groups of

variables with the same sparsity pattern.

4. Reverse Polish Notation (symbolic representation in the NLP problems

instead of sparse)

6. Computer memory configuration and pre-solving before starting the

solution. (save memory in the CPU)

Conclusion on the Scheduling Problem

22EWO Meeting, Sep 22nd, 2016.

Novelty:

• Segregates crude management in storage assignment1 and crude blendscheduling.2

• Phenomenological decomposition in logistics (MILP) and quality (NLP)problems applied in a scheduling problem.

Impact for industrial applications:

• UOPSS modeling, pre-solving, and parallel processing permitted to solve an2h time-step discrete-time formulation for a highly complex refinery (34crude-oils, 24 storage tanks, 9 feed tanks, 5 CDUs): for 7 days (84 time-periods)



23

Next Steps:

• Add upgrading units and their tanks (RFCC hydrotreaters, RFCC, VDU)

• Add cutpoint optimization instead of modes of operation in CDUs

• Crude-Oil Blender x Sequential Blending to prepare the Feed Tanks.

• Factors using bulk qualities as an LP between the Storage and Feed Tanks tobe used in the logistics (MILP): to have a more robust PDH decomposition

• Whole Scheduling: from Crude-Oils to Fuel Deliveries

• Initialization, Synchronization, Real-time Scheduling with feedback


Conclusion on the Scheduling Problem

Thank You

24

Q?&A!

www.industrialapplications.club

www.induapps.club

[email protected]

Gracias