Agent Based Models for Enterprise Wide Optimization and Decision Support Optimization and Decision Support Raj Srinivasan Raj Srinivasan Dept of Chemical & Biomolecular Engg National University of Singapore Process Systems & Modeling Institute of Chemical & Engg Sciences [email protected]CMU, 1 Dec 2009 1
52
Embed
Agent Based Models for Enterprise Wide Optimization …egon.cheme.cmu.edu/ewo/docs/SrinivasanSupply_Chain...Agent Based Models for Enterprise Wide Optimization and Decision SupportOptimization
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Agent Based Models for Enterprise Wide Optimization and Decision SupportOptimization and Decision Support
Raj SrinivasanRaj SrinivasanDept of Chemical & Biomolecular Engg
National University of SingaporeProcess Systems & Modeling
p• Soft sensors / alarm mgmt• Process transitions mgmt• Fault diagnosis• Fault tolerant control
Plant Operator• Image based control of particulate processes
2
A Energy Company
Electricity
R fi i
Jet Fuel
Oil & Gas Transportation
Refining
Productionp
Storage & Transportation
Petrol & Diesel
Petrochemicals
3
Supply Chain Management
Crude Oil Operations Scheduling
4
All for the want of a nail…“For want of a nail, the shoe was lost,For want of a shoe, the horse was lost,For want of a shoe, the horse was lost,For want of a horse, the rider was lost,For want of a rider, a message was lost,, g ,For want of a message, the battle was lost,For want of a battle, the kingdom was lost,And all for the want of a nail!”
George Herbert, in Outlandish Proverbs (1640)
5
Integarted Models of Supply Chains & Enterprises
6
PSE 101: Unit Operations
Unit-leveld li d i i l ti t l ti i timodeling, design, simulation, control, optimization
“Physicochemical phenomena”Physicochemical phenomena Virtual units through Objects / Aspects 7
• Low stock, out-of-crudeO ti di t d• Operation disrupted
• Demand unfulfilled• Safeguards
• Safety stock• Safety stock• Mitigating Actions
• More reliable shipper34
Consequence Analysis via Simulation
3PL ReliabilityAverage Customer Satisfaction (%)
i h3PL Reliability
Average Profit ($, million)
High Low
Yes 98 95
No 95 91
Safety stock
High Low
Yes 93 38
No 83 27Safety stockNo 95 91
• Current safety stock cannot make up for poor performance of 3PL provider • Additional safeguards or mitigating actions required e g higher safety
No 83 27stoc
• Additional safeguards or mitigating actions required, e.g. higher safety stocks, emergency crude procurement.
• Probability of crude arrival delay: High 3PL reliability (0 05) Low 3PL reliability (0 10)• Probability of crude arrival delay: High 3PL reliability (0.05) – Low 3PL reliability (0.10)• Safety stock: crude (100 kbbl) – product (20%)• # Simulation runs per scenario: 300• Demand variability across cycles: 25%
Si l ti h i 120 d• Simulation horizon: 120 days
35
Process Control & Supervision
Controller Malfunction
Feedback Controller
Sensor FailureProcess DisturbanceH E
Dynamic ProcessActuator Sensorsu y
Sensor FailureProcess DisturbanceHuman Error
Actuator Faults Structural Failures
Diagnostic System
36
Disruptions in Crude Procurement Process
PostingP ti
EXCHANGE
1. Procurement initiation
2. Market dataFetch quotes from postings on
the exchange
LEGEND
...............
Posting PostingPosting
3. Crude basket
5. List of pickup location
6. Request for bids
OPERATIONS
PROCUREMENT
4. Refined crude basket
p pand pickup date for
each crude
7. Bids received from 3PLs
CBA
OPERATIONS
STORAGE
31 2
AND
OR
7. Bids received from 3PLs
Bid deadline over
8. List of best bids foreach crude9. Place order for
crude
10. Orderconfirmed
SALES
11. Information on crudebought
12. Contractawarded to
respective 3PL13. Order
Confirmation14. TransportInformation
15. TransportInformation
LOGISTICS
CBA3PL
OIL SUPPLIER
37
Supply Chain DashboardA DCS for the Supply Chain######
Customer Satisfaction 100
95 98 100
40
60
80
100
Perc
enta
ge
Revenue from Sales
288
241 238
298
150
200
250
300
350
ales
(K$)
Operating Cost
245 230 243
298
150
200
250
300
350
atin
g C
ost (
K$)
pp y
Brent Crude GasolineRaw Materials Shipment Actual Demand Forecast DemandM d Di l
0
20
Mar '08 Apr '08 May '08 Jun '08
P
0
50
100
Mar '08 Apr '08 May '08 Jun '08
Sa
0
50
100
Mar '08 Apr '08 May '08 Jun '08
Ope
ra
800 800Raw Materials Shipment Actual Demand Forecast Demand Arrival Date Type Status Due 24 Jul 08 Due 31-Jul-08
1 27-Jul-08 B & O & A On time Gasoline 250 Gasoline 2802 3-Aug-08 B & K On time CDU Throughput Jet Fuel 63 Jet Fuel 953 9-Aug-08 B & O & A Scheduled Diesel 125 Diesel 1254 16-Aug-08 B & O & A Scheduled Kuwait Crude Total throughput 200 Jet Fuel Fuel Oil 122 Fuel Oil 154
Brent 90Kuwait 110 Products ShipmentD b i 0 Shi t D t T St t
Mode Diesel
0
400
800
800
1200 400
0
400
800
Dubai 0 Shipment Date Type StatusOman 0 1 G + D WIPArab Light 0 2 J + F WIP
• Product to be delivered on: 120Product Inventory
200250
300350
• Date of Stock-out: 119• Amt of crude Shortfall: 758 kbbl
050
100150
200
0 20 40 60 80 100 120 140
Unchecked DisruptionDisruption Managed
40
Scenario 1: Transportation Delay
Can’t unload into Tank 6
Parcel 7 delay from Day 115 to 121
as it is charging CDU 3
-30 U3-50 U3-50 U3100 P7100 P7
Tank 6 will run out of crude at time 11
Existing schedule is infeasible!41
All for the want of a nail…“For want of a nail, the shoe was lost,For want of a shoe, the horse was lost,For want of a shoe, the horse was lost,For want of a horse, the rider was lost,For want of a rider, a message was lost,, g ,For want of a message, the battle was lost,For want of a battle, the kingdom was lost,And all for the want of a nail!”
» PEDD-LJS completion date is later or at most sameA small modification to a policy could have a big impact.
51
Conclusions
• Dynamics are importantSh t t & l t– Short-term & long-term
– Decisions related to operation, control & design
• Agent based models offer a natural paradigm for• Agent based models offer a natural paradigm for modeling the enterprise– Simulation-optimization strategy for designSimulation optimization strategy for design– Control structure for disruption management
• From PSE to PSE2 (= PSE of Enterprise)o S o S ( S o e p se)– Analogy from PSE are useful
• Representation, Modeling & simulation • Control & supervision