Transcript Production Functions - Massachusetts Institute of Technology

Screening Models
Richard de Neufville
Professor of Engineering Systems
and of
Civil and Environmental Engineering
MIT
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 1 of 21
Outline





Issue 1: Which Flexibilities add the
most value to project?
Issue 2: Why is this a challenge?
Concept of Screening Model
Development of Screening Models
Types:
Bottom-up
— Simulator
— Top-Down
—

Use in Practice
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 2 of 21
Definition of Flexibilities:
Why a Problem? Possible types…
Answer depends on:
 Nature of System –
mines vs. manufacturing;
— small vs. large quantities
—

Kinds of Uncertainties
—

Intensity of Uncertainties
—

State of Technology? Or of Demand?
Slow or Fast Evolution (Subway vs. Google)
Cost of Implementing Flexibilities
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 3 of 21
Definition of Flexibilities:
Why a Problem? Complexity
The curse of dimensionality again!
• Too many combinations to explore
• Why?
Complexity of design itself – creation of a
single design for a oil platform may take a full
day with “oil and gas” model…
— Crossed with need to examine many scenarios
scenarios of uncertainty over time – cannot in
practice simulate hundreds of patterns
—

We simply cannot explore design space
analytically
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 4 of 21
Concept of Screening Model (1)



A rapid way to explore design space
systematically
Substitute for designer “experience” or
“intuition” -- an engineering approach
Metaphor:
—
High Altitude flight over unknown territory,
looking for special features
—
Can be complete
—
(But of course can miss some possibilities)
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 5 of 21
Concept of Screening Model (2)

The image:
A few intuitive designs
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
or
Systematic Search?
Richard de Neufville
Screening Models

Slide 6 of 21
Concept of Screening Model (3)
• Screening Models not a substitute for detailed models
• They define set of designs for detailed analysis
Screening
Models
Rapid Analysis of
Performance of
Possible Designs
Complex Models
Shortlisted
Candidate
Designs
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Detailed Analysis
of Short-listed
Candidate
Designs
Richard de Neufville
Screening Models
Final
Design

Slide 7 of 21
Development of Screening Models

Desirable features;
Rapid Analysis of Many Possibilities
— Rank Designs reasonably accurately
—

How important is accuracy? Not much!
Accuracy is not their function.
— Their suggestions will be checked by analysis
—

This is an important distinction
Practicing professionals want the “real thing”
— Tendency needs to be resisted
—
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 8 of 21
Types of Screening Models

Bottoms-up:
—

Simplified versions of detailed descriptions
Simulators
Mimic detailed descriptions
— Not necessarily “simulations” …
—

Top-down
—
Conceptual Representations of system
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 9 of 21
Bottoms-up Approach


Example: River Basin Development
Full Analysis involves
River Flow Model (channel, dams, diversions)
— Hydrologic Model (rainfall, snow melt, etc)
— Economic Model
(Value of Power, Irrigation…)
— Stochastic seasonal patterns of water flow, use
—

Screening Model
Average Annual Flows
— Optimization possible
— Identifies reasonable possibilities
—
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 10 of 21
Bottoms-up Example (Wang)
Price of
Electric Power
Renminbi /
KWH
Optimal Characteristics of Dams
Value of
Site 1
Project
Renminbi. 106
Power
MW
Site 2
Volume
Power
9
MW
10
M
3
Site 3
Volume
Power
9
MW
10
M
3
Volume
10
9
M
3
0.10
0
0
0
0
0
0
0
0.13
367
3,,600
9600
1,700
25
0
0
0.16
796
Same
Same
Same
Same
0
0
0.19
853
Same
Same
Same
Same
1,564
6,593
0.22
1,607
Same
Same
Same
Same
1,723
9,593
0.25
2,196
Same
Same
Same
Same
1,946
12,242
0.28
2,796
Same
Same
Same
Same
1,966
12,500
0.31
3,396
Same
Same
Same
Same
1,966
12,500
Source: Adapted from Wang (2005) p. 188.
Step 1: Optimize for Range of Conditions
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 11 of 21
Bottoms-up Example (Wang)
Site
Source of Option Value
Timing
Design
1
Yes
NO
2
Yes
NO
3
Yes
Yes
Source: Adapted from Wang (2005) p. 189.
Step 2: Identify Factors that might enter optimal
design in different cases – these provide flexibility
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 12 of 21
Simulator Models




Idea is represent overall performance of
detailed model using a simpler model
We focus on output of the system, not on
replicating its internal workings
Largely a statistical exercise – to fit
simple model with few parameters, to
output of detailed model
Two versions
—
Direct and Indirect
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 13 of 21
Direct Approach

Also known as “response surface model”

Four Steps
1. Select major factors (Xi) and vary them over a
range (e.g.: Oil prices = 20, 40, 60, 80, 100, 120
$/bbl)
2. Run the detailed model with these values and
obtain overall results (e.g. NPV of Project)
3. Do statistical analysis to fit the factors (Xi) to
output
4. Result is Screening Model: Output = f (Xi)
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 14 of 21
Indirect Approach



Use first principles to construct simplified
models of components of detailed model
(e.g. mass balance equations)
Assemble simple sub-models to create a
complete model
Validate by simulation
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 15 of 21
Validation of Indirect Model (Lin)
Model Prediction and Actual Production Profiles
350
total fluids production rate [m bd]
actual
300
oil production rate [m bd]
actual
gas production rate[m m s cf/day]
Oil /gas /water production rates
actual
250
water production rate [m bd]
actual
200
150
100
50
0
0
5
10
15
Years of production
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 16 of 21
Top-down Approach




Focus is on how major parts of system
influence output
Appears to be most useful when we have
dynamic systems that evolve over time
Best known examples use “Systems
Dynamics”
Preparing a good SD model requires a
great deal of effort (Steel took about 2
years on model of Kenya power system)
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 17 of 21
Example of Top-down model (Steel)
Top-down model of how consumers respond to market
state of market and power supply – and vice-versa
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 18 of 21
Use of Screening Models
3 Approaches



Conceptual – get planners to “think
outside the box” : Local hospital
Optimization (e.g. Wang)
—
Optimize for one configuration
—
Repeat for Others
—
Observe which components change
Patterned search – Like optimization, but
not algorithmically driven
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 19 of 21
Examples of Search Patterns


Possible Dimensions
Phased Design -- smaller units (instead
of larger ones)

Modular – “plug and play” easy additions

Design for Expansion – space, strength
—


Parking Garage; Bridges over Hudson, Tagus
Platform Design – Chassis for cars (Suh)
Shell Design – empty space available for
future use (Mt. Auburn Hospital)
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 20 of 21
Summary



Screening Models Very Useful in Identifying
Opportunities for Flexibility
Are complementary, not competitive, with
detailed models of system
Feed results into detailed models, and thus
guide their direction
Engineering Systems Analysis for Design
Massachusetts Institute of Technology
Richard de Neufville
Screening Models

Slide 21 of 21