Each individual business with their own unique assets and supply chain optimizing their decisions according to their incentives, abilities and working culture. The latest developments in Planning (Petro) and AI use cases.
2. Maintaining the status quo is not acceptable
TIGHTENING
HEALTH,
SAFETY AND
ENVIRONMENT
REGULATION
is forcing facility
investment in order
to maintain the
license to operate
TECHNOLOGICAL
INNOVATION
AND
DIGITIZATION
is causing
disruption and
consideration of
new operating
models
CHANGING
INDUSTRY
DEMOGRAPHICS
are presenting
leadership, succession
and competency
challenges, with different
worker needs
RISE OF SHARE
HOLDER
ACTIVISM
is driving strong
appetite for financial
stewardship and asset
optimization
PROLIFERATION
OF DATA AND
DATA-DRIVEN
ORGANIZATIONS
is compressing
timeframes for decision-
making and introducing
new digital competitors
3. Proprietary Information 3Proprietary Information 3Proprietary Information 3Proprietary Information
Squeeze down on the
gap between potential
and realized margin
Create more utility for
the end customer
Extend problem-solving
eco-sphere beyond the
plant
Outmaneuver
competition
Manage day-to-day
performance
safely and reliably
Respond rapidly to
market swings
Operate at true
optimum
Why
Digitalize?
4. Proprietary Information 4Proprietary Information 4Proprietary Information 4
Eachindividual businesswiththeirownunique
assetsandsupplychainoptimizingtheirdecisions
accordingto their incentives, abilities
andworkingculture.
Window of Optimization
HYDROCARBON VALUE CHAIN
5. 5Proprietary InformationProprietary Information
Does the refinery have the ability to
produce the products being
demanded?
Approximately how much shortfall
will need to be factored in?
How much surplus is available to
sell, and in what time frame?
LinearProgramming
(LP)tools,suchasPetro,
havebeenusedheavily
toprovideanswersto:
7. Proprietary Information
Production
Planning
Scheduling
Supply
Planning
Investment
Planning
Losses Due to
Uncertainty
Reduced
Backcasting and improved planning technology
with Digitalization and the integration of
artificial intelligence (AI)
DecisionValue
Automation
MINUTES
Ago
HOURS
Ago
DAYS
Ago
YEARS
Ago
SECONDS
Ahead
MINUTES
Ahead
HOURS
Ahead
DAYS
Ahead
MONTHS
Ahead
MONTHS
Ago
NOW
Decision-Making
Time Horizon
Decision Impact
Time Horizon
8. 8Proprietary Information 8
Process and Offsites
Control & RTO
Reconciliation &
Analysis
Production
Accounting
Rigorous
Simulation
Supply Chain
Scheduling
Operational
Planning
Corrected
model
Corrected
data
Corrected
model
Plant
Raw
data
- 1 MONTH - 1 WEEK NOW
Digital Future
• Data driven, automated
identification of optimization
using AI
• Intelligent, automated work
processes
• Automated data management
and tool integration
• Cloud enabled to facilitate;
• Scalability
• Collaboration
• Support
• Rapid enhancements
• Knowledge management
• Integration
• Visualization
11. Proprietary Information
ApproachandMethods
Convergingon theGlobal Optimum- Quickly
11
Model Type Equations Coefficients
Solve Time
(Seconds)
Single-Period Models 5,000-10,000 100,000-200,000 5
Multi-Refinery Models 10,000-20,000 200,000-400,000 10
Multi-Period Models 25,000-100,000 500,000-1,000,000 60
Petro Model Sizes and Solve Times
11
12. Proprietary Information
ApproachandMethods
MaintainingModel Accuracy
12
Petro Validation Tool
Timely refinery optimization supported by
confidence in LP
Continuous validation of LP accuracy
Validation data and SME analysis stored in
database
Data shared in real-time
Monitoring yields, with Petro models as a
reference, results in GM improvements for
Chevron
$13 MM$10 MM$25 MM$78 MM
2015 2016 2017 2018
12
13. Proprietary Information
ApproachandMethods
Model VersionControl andCaseSharing
13
Model and case
architecture enables
version control and case
sharing.
Cases or models are
searchable and directly
downloadable from within
the Petro application
Model updates
automatically pushed to all
users - simultaneously
13
Petro Case Sharing in the Cloud
14. Proprietary Information
PETRODevelopmentand
Valueto Chevron
14
Chevron’s PETRO is an industry
leading LP modeling platform
developed over 35 years
through collaboration between
software developers and
refinery planning engineers.
Value is created through use in
all refinery optimization work
processes including support of
capital project evaluation
0
200
400
600
800
1000
1200
1980 1985 1990 1995 2000 2005 2010 2015 2020 2025 2030
PETRO Value Generated Per Year
($ Millions)
1990
Recursion-
based models
2007
Terminal-
Distribution
Models
2014
Code
Re-write
Petro 2020
Application Database, Case
Sharing in the Cloud, and
Visualization
1985
PC version
released
2000
Windows
version
released
2002
Multi-period
Models
2004
Multi-location
14
15. 15Proprietary Information
Process and Offsites
Control & RTO
Reconciliation &
Analysis
Production
Accounting
Rigorous
Simulation
Supply Chain
Scheduling
Operational
Planning
Corrected
model
Corrected
data
Corrected
model
Plant
Raw
data
5
Plant-wide
optimization
OptimizationAI UseCases
1
Demand and
price forecasting
2
Reconciliation
and variance
analysis
3
Automation
4
Automation of
production
scheduling
6
Opportunity Engine
16. AI study
Demand and
price forecasting
Challenge
Feedstock and product prices
and product demands are
increasingly volatile.
Optimization of the value chain
with poor price and demand
forecasts is sub optimal.
17. Solution
Use ML algorithms to forecast
demand. High performance
computing and AI we can optimize
and compare multiple scenarios to
reduce risk.
Find the predictive model that leads
to the minimal prediction error such
as a neural network (e.g. MONMLP).
Provide the model predictions to
the planners to improve multi-
period optimisation.
Predicted
Actual
18. AI study
Automation of LP
& Scheduling
Model Updates
from Rigorous
Model
Challenge
For speed, Planning and
Scheduling Models are simplified
(often linear) from rigorous models.
Linear models result in artificial
constraints and are not able to see
all opportunities.
Maintaining the rigorous models
and recalibrating to actual plant
performance takes time and
expertise.
19. Solution Benefits
An ML surrogate model that improves
the linear models can be generated.
This can be done by using synthetic
data from the Rigorous Model to train
the ML model.
Use past model performance and
current model deviations and an AI
Classifier to find the pattern of
deviations that characterize a poor
model and trigger and update of the
relevant part of the model.
Decrease Model Update Times due
to better definition of
Improvements areas
Reduced need for SME participation
ML Planning/Scheduling Model that
accounts For a larger range of
Operating conditions
Maximize Model Accuracy by
Better Definition Of Re-calibration
Needs
20. 20Proprietary Information
AutomationofLP & SchedulingModelUpdates
fromRigorousModel
Rigorous
Refinery Model
Planning /Scheduling
Model Monitoring
Actual
Plant Behavior
Lab Data
Plant Data
Data Clean-Up
Select data-set
for
Recalibration
Activate
Model Update
Signal
Historical
ML Model
Performance
Train
Source of
Inaccuracies
Explained
AI Classifier
Monitoring Recalibration
Planning /Scheduling
ML Model (*)
Train
(*) ML structure suitable for client’s needs
Historical
Rigorous Model
Performance
Train
Rigorous
Model
Validation
21. 21Proprietary Information 21
Summary
Optimizing the hydrocarbon value chain is
critical to achieving Commercial Excellence
in an increasingly VUCA world
New technology such as Cloud and AI
provides an opportunity to digitally
transform the value chain by:
• Creating new business models
• Expanding the scope of optimization
• Improving execution
• Increasing agility