Dynamic Modeling of Learning in Emerging Energy Industries:
The Example of Advanced Biofuels in the United States
Laura J. Vimmerstedt '*, Brian W. Bush ', and Steven O. Peterson 7
' National Renewable Energy Laboratory
National Renewable Energy Laboratory RSF 300
15013 Denver West Parkway, Golden, CO 80401-3305
“ Corresponding author, 303-384-7346
laura.vimmerstedt@nrel.gov
> Lexidyne, LLC
19 S. Tejon Street, Suite 212
Colorado Springs, CO 80903
603-359-7188
steve.peterson@lexidyne.com
Acknowledgments
This work was supported by the U.S. Department of Energy under Contract No. DE-AC36-08-
GO28308 with the National Renewable Energy Laboratory, 15013 Denver West Parkway,
Golden, CO 80401-3305.
Abstract
This paper (and its supplemental model) presents novel approaches to modeling interactions and
related policies among investment, production, and learning in an emerging competitive
industry. New biomass-to-biofuels pathways are being developed and commercialized to support
goals for U.S. advanced biofuel use, such as those in the Energy Independence and Security Act
of 2007. We explore the impact of learning rates and techno-economics in a learning model
excerpted from the Biomass Scenario Model (BSM), developed by the U.S. Department of Energy
and the National Renewable Energy Laboratory to explore the impact of biofuel policy on the
evolution of the biofuels industry. The BSM integrates investment, production, and learning
among competing biofuel conversion options that are at different stages of industrial
development. We explain the novel methods used to simulate the impact of differing assumptions
about mature industry techno-economics and about learning rates while accounting for the
different maturity levels of various conversion pathways. A sensitivity study shows that the
parameters studied (fixed capital investment, process yield, progress ratios, and pre-commercial
investment) exhibit highly interactive effects, and the system, as modeled, tends toward market
dominance of a single pathway due to competition and learning dynamics.
Key words
Biomass, Biofuel, Learning, Learning Curve, Experience Curve, Policy, Renewable Fuels
Standard, System Dynamics
Introduction
Modeling the emergence of an industry presents challenges in assessing the dynamic interactions
among investment decisions, cost reductions during development (e.g., due to industrial learning
and economies of scale), and levels of utilization and associated production. The Biomass
Scenario Model (BSM) uses novel approaches to address several of these challenges. It was
developed by the U.S. Department of Energy (DOE), partnering with the National Renewable
Energy Laboratory (NREL), in an effort to reduce ambiguity about potential futures and to
improve understanding of the role of policy in industry development (Bush, 2011; Bush et al.,
2008; Newes et al., 2011; Lin et al., 2013; Peterson et al., 2013; Inman et al., 2014). Developed
in isee system’s STELLA®', the model uses a system of coupled ordinary differential equations
and integrates these forward in time. This links variables that represent physical, technical,
economic, and behavioral features of the biofuels supply chain through interdependent rates of
change and feedbacks. The BSM has been extensively validated; in particular, the learning
curves which are used here are based on historical data on learning in ethanol and related
industries (McCurdy ef al., n.d.). The BSM is an analysis platform that is designed to explore the
impact of biofuel policy on the evolution of the biofuels supply chain. It includes modules
representing the U.S. agricultural system; pathways that convert starch, cellulosic, oil crop, and
algal feedstocks into liquid fuels; and downstream distribution and dispensing of fuels.
Today’s biofuels industry includes mature commercial enterprises, such as starch-to-ethanol,
sugarcane-to-ethanol, and biodiesel. Other emerging technology pathways convert cellulosic and
bio-oil feedstocks to ethanol, butanol, or other hydrocarbons; they could develop substantially in
the future but are not yet mature at industrial scales. These biomass-to-biofuel conversion
technology pathways have potential to reduce carbon emissions, to enhance U.S. energy security,
and to increase the degree to which non-food biomass resources supplement or displace corn-
based fuel ethanol production. The Energy Independence and Security Act of 2007 (EISA)
mandates 36 billion gallons of renewable transportation fuel in the U.S. marketplace by the year
2022 (Energy Independence and Security Act of 2007), and includes targets for growth of
advanced biofuels (cellulosic biofuels, biomass-based diesel, and other biofuels that meet
threshold greenhouse gas emissions reductions of 50%). In response to this mandate, successive
U.S. administrations have provided incentives for advanced biofuels development.
In developing the BSM, we sought to provide a simple representation of the mechanisms,
constraints, and trade-offs associated with supply chain evolution based on the physical features
and relationships of the biomass-to-biofuels industrial system. We drew from a broad set of
information sources in order to incorporate data about today’s industry, assess the current status
of less mature pathways, estimate the economic and technical performance of prospective
pathways, characterize relationships and constraints, and represent the effects of choices by key
decision makers. A central feature of the BSM is a dynamic framework that integrates
investment, production, and industrial learning among competing biofuel conversion options.
‘i http://www. iseesystems.com
In this paper, we focus on a subset of the BSM by addressing learning and associated investment
in conversion pathways that are in early stages of development and whose mature techno-
economic performance is unknown—not the mature, established conversion pathways that
dominate current production of biofuels in the United States. We show how the model represents
learning concepts and highlight how our approach differs from the approaches commonly used in
other energy models. We then explain how the BSM couples together investment, production,
and learning processes in the context of competing biomass-to-biofuel conversion technology
pathways. A sensitivity analysis of the simplified model is used to derive insights into the role of
pre-commercial investment, learning rates, mature-technology capital costs, and mature-
technology yields in the competition among pathways for investment and in overall stimulation
of industry development.
Applying learning curve concepts in the BBM
Cost reduction through learning-by-doing is a prominent feature of many analyses of the energy
sector (Junginger et a/., 2010). Learning curves (also known as experience curves) are based on
the empirical observation that in many industries’ unit cost declines along with increases in
production experience. This finding was documented in the 1930s for labor input for airframe
manufacture (Wright, 1936). Beginning in the mid-1960s, analyses used the experience curve
concept to explain the cost behavior and the resultant market share dynamics in competitive
industries (Hax and Majluf, 1982; Kouvaritakis et a/., 2000). In the energy sector, empirical
observation has also supported a learning curve approach. For example, endogenous learning is a
component in studies focusing on ethanol production in Brazil (Van den Wall Bake et al., 2009)
and the United States (Chen and Khanna, 2012; Chum ef a/., 2013) and in studies focusing on
electricity generation, including nuclear, coal, hydropower, wind, and solar photovoltaics
(McDonald and Schrattenholzer, 2001).
The system dynamics literature describes the use of learning curves in models of strategy
dynamics, technological development, industry growth, and investment decision-making. For
example, Naim and Towill (1994) present various learning curve formulations and their
application in system dynamics models. Fiddaman’s work on integrated climate-economy system
dynamics models (1997) explicitly accounts for learning-by-doing. Morrison (2005) uses an
extension of learning curve theory to show the potential emergence of positive feedbacks in
productivity in the context of throughput constraints. Sterman et a/. (2007) describe a system
dynamics model with learning curves that is used to explore “get big fast” strategies under
conditions of bounded rationality for actors in the system. Learning curves are also a prominent
feature of a system dynamics model of the photovoltaic industry (Jeon and Shin, 2014). This
model uses system dynamics and Monte Carlo analysis to examine how the value of PV
technology varies over time in response to uncertainty and volatility in 11 key input drivers.
A single-factor learning curve is commonly used for analyses in the energy sector. Less
commonly used are multiple-factor learning curves that incorporate, for example, research and
development investment (Junginger ef al., 2010). In the single-factor formulation, each doubling
of experience yields a constant percentage decrease in unit costs. The relationship is often
expressed as a power law:
Y=ax? qd)
Y current unit cost
X cumulative production
a unit cost of initial unit
b slope of the function when plotted on a log-log scale
PR=2- Q)
LR=1-—PR (3)
PR is the progress ratio, the relative cost after each doubling of cumulative
production
LR is the learning rate, the relative cost reduction after each doubling of
cumulative production
A single-factor learning curve yields results such as those shown in Figure 1.
AN Sa ———Progress Ratio = 0.9
~~ Progress Ratio = 0.8
30
Progress Ratio = 0.7
Cost Index (uintless)
Bw
66
(0) 1 2 3 4 5 6 7 8 9 10 11 12
Cumulative Production (widgets)
Figure 1. The chart illustrates a single-factor learning curve relationship under three different progress ratios. With
each doubling of cumulative production, the cost index is reduced by a constant percentage based on the progress
ratio.
The learning curve approaches in the literature generally observe a historical progress ratio,
which may then be extrapolated to the future. For a single factor learning curve for unit cost, this
would result in an implicit ultimate cost of near zero. In the BSM, we use a learning curve
approach to enable us to interpolate between the characteristics of current and future technology
for each technology pathway under consideration. In doing so, we have adapted the single-factor
learning curve in several ways. First, instead of using an implicit asymptote of zero for unit cost,
in the BSM we define an explicit asymptote of a “mature technology.” As discussed in Newes et
al. (2011), we introduce a variable called maturity, M:
4
es for E=L* 7
*\\ Ind
(mF) " @
Me= iB
My otherwise
L* = max{L,E,} (5)
Mp is initial maturity
L is the minimum experience for learning
L* is the effective minimum experience for learning
E is the cumulative experience
Eo is the initial cumulative experience
PR is the progress ratio
From Newes ef al. (2011).
Second, we adapt the single-factor learning curve approach that typically is connected to a single
technical attribute (unit cost). Instead of a single measure, we define a vector of technical
attributes that are assumed to depend on industry maturity. This vector includes the following
attributes, each of which has a mature value input and a current state-of-industry value that is
calculated. As the industry matures, each value is assumed to approach the mature value or nt"
plant characteristics.
* Process yield
¢ Feedstock throughput capacity—the degree to which facilities are able to perform at
nameplate capacity
* Capital cost—the premium in capital cost, beyond the n” plant estimate, which would be
observed if development of a facility was begun today
¢ Investor risk premium—the additional premium, beyond normal hurdle rate, that
investors would require for investment in the facility
¢ Access to debt financing—the portion of the expected facility capital cost that would be
financed via borrowing (as contrasted with equity investment).
Accordingly, we define a multiplier m that is arrayed by technical attribute i, where n denotes input
n" plant values. This multiplier is a function of maturity, as described next. Generally, the value of
the technical attributes A in any simulated year is calculated as:
A is a vector of technical attributes (various units)
m is a multiplier (unitless)
Tis the index of technical attributes
nis the ultimate n” plant attributes
O0<m (7)
i
Figure 2 summarizes these interactions in a causal loop diagram.
t. +
+
+
+ Investment
fas hy +
R
®W R Conversion Facilities
oe iw) R
Access to tR)
Risk Process ; +
Debt ; Pravin. Yield Throughput Capital Cost tR) iF
Financing
= Production
- ‘ Phi Ne, we
Maturity ay
Figure 2. A causal loop diagram summarizing the interactions of maturity, the vector of technical attributes,
investment attractiveness, conversion facilities, and production.
Third, instead of a single learning curve metric, we have applied learning curves in a cascading
fashion to multiple industry development scales (pilot, demonstration, pioneer, and commercial).
Pilot and demonstration facilities are constructed without the intention of producing fuel for
commercial sale. For most technologies in the BSM, commercial scale is over 2,000 short tons
per day throughput capacity. Pioneer scale refers to the early commercial developments that are
assumed to be approximately one-third the scale of commercial facilities and to have higher
capital costs. As a result, at any point in simulated time, the performance and cost characteristics
of a pioneer or commercial-scale facility are estimated and used to inform the model’s
investment allocation. To accomplish this, maturity is calculated at each developmental scale,
and the multiplier m in the equation above is calculated based on the impact of each development
scale:
m,= CM, +[D.M, +F(1-Mp))xd-M- ) (8)
0<Msl (9)
Mc is commercial maturity and Mp is demonstration maturity
Cis the vector of mature commercial multipliers
D is the vector of mature demonstration multipliers
P is the vector of mature pilot multipliers
Mature multipliers at commercial, demonstration, and pilot scale (C, D, and P) define the
maximum performance on each technical attribute that can be achieved through experience at
that scale.
With these three modifications to the single-factor learning curve, we have developed a novel
modeling approach to modeling learning that accounts for performance in multiple technical
attributes that approach non-zero ultimate values through development at multiple scales. This
approach is best suited to situations wherein the modeler can readily assume knowledge of the
ultimate technical attributes of the technology and the mature technology characteristics, and
would like to explore scenarios for the rate at which current technology approaches this mature
performance. We are not aware of other literature describing this application of learning curves.
In situations where mature technology characteristics can be estimated, this approach replaces
the implicit assumption in much of the literature of ultimate near-zero cost (if unit cost is the
attribute). For the BSM, we operationally define “mature technology” by drawing from
engineering design studies that estimate capital investment, process yield, and fixed and variable
operating costs for the “n" plant” (where n is a large number) of a given pathway. Because the
BSM relies on design studies to provide estimates for n"" plant techno-economics, its results are
subject to uncertainties in yield and cost estimates as well as scale effects of assumed
commercial plant sizes.
Figure 3 shows an illustrative example of the correspondence between production (or
experience), maturity, and values of technical attributes over the simulated time. We use
different units of experience at different scales, as shown in panel a. As shown in panel b, the
maturity multipliers include experience at each scale, increasing as cumulative production grows
over time and asymptotic to one. Panel c illustrates how current (or state-of-industry) values for
two of the technical attributes change as they approach mature values as cumulative production
increases over time.
Because the BSM is a forward-looking simulation model of technologies that are not yet
commercialized, there is no empirical data from which to estimate many of the progress ratios
that we use. Instead, it uses assumed values that are informed by statistically estimated values for
progress ratios from the literature about related technologies, such as starch ethanol production
and, more generally, the chemical process industry. In their review of retrospective studies of
learning in the energy sector, McDonald and Schrattenholzer (2001) identify a high degree of
variability in estimates for learning rates—both within a given technology and across differing
energy technologies. They suggest that modelers should consider the uncertainty in the value of
learning rates (McDonald and Schrattenholzer, 2001). For the BSM, this points to the importance
of including sensitivity analysis of learning rates in this paper. Model results are not predictions,
but should be viewed as plausible scenarios subject to uncertainties in model structure and data
inputs.
(a) Cumulative Experience
—Cumulative Pilot a0 | I H
Experience (yr) 100 | ! |
— Cumulative Demo i |
Experience (yr) 50 '
—Cumulative Industry 0 H |
7 1 ] 1
Output (107 gal) 0 6 | 10 115 20
H | year |
| \
I \
(b) Maturity Indices |
\
1 | |
—Pilot Maturity |
(unitless)
—Demo Maturity 0.5
(unitless)
—Commercial
Maturity (unitless) 0 : , ’
ie) | 5 | 10 | a5 20
year ;
. . !
(c) Selected Technical Attributes
\ | th
—Process Yield factor | |
(unitless) 45 H | i
au es
—Capital Cost 1 |
Growth factor 05 ' ;
(unitless) ou ! 4
0 5 10 15 20
year
Figure 3. (a) The first panel illustrates experience profiles at pilot and demo scales, as well as commercial scale
(measured in industry output). Pre-commercial curves are based on accumulation of operating time at pilot and
demo scales. Cumulative industry output grows as a quadratic, reflecting constant linear growth in annual industry
output beginning in year 14. (b) The second panel shows change in maturity index with increasing cumulative pilot
experience, demonstration experience, or commercial production over time. The corresponding experience or
production curves are shown in panel a. Pilot experience precedes d experience, which precedes
commercial production. Maturity is correlated with doublings of experience at each scale. The maturity index is an
abstract, unitless multiplier that is multiplied by metrics of mature technological performance to estimate the state of
the industry. (c) The third panel i a ding effect of ion on selected technical attributes.
Technical attributes improve with maturity, based on a distinct calculation for each measure. The technical attributes
shown are a subset of those used for commercial biorefin . At a given year in simulated time, investment techno-
economics will reflect the cumulative effect of pilot, demo, and commercial maturity levels. Technical attributes
progress differently at different industry-development scales.
Coupling learning, production, and investment in the BBM
Production and investment in the BSM depend upon performance in each technical attribute, as
calculated in the learning process. In the BSM, techno-economic information drawn from design
reports and engineering studies is used to select the input assumptions for the initial and mature
industry (or n"' plant) technical attributes vector (listed above), the financial and performance
characteristics for different biomass-to-biofuel conversion technology pathways. Over simulated
time, the model’s learning curve logic determines how the state of each pathway moves from
today’s characteristics toward ni" plant maturity. Together with other inputs, notably feedstock
cost and product price, the state of each pathway defines streams of expected costs and revenues
associated with a prospective project investment based on costs and revenues in the simulated
year, without investor foresight. In the BSM, we discount these expected costs and revenues
streams to the current simulated year to calculate the net present value (NPV) of the prospective
investment, as follows:
NPV, =R,-[, +C,+T)] (10)
NPY is the net present value of investment in a biorefinery of technology i
R is the NPV of expected revenue less expected operating costs for technology j
T; is the initial equity investment for technology j
Cis the present value of the debt for technology j
T is the net present value of the taxes less tax credits for technology j
The net present value calculation uses the required rate of return, which changes with maturity,
and is calculated according to:
aR
Ry,.= yt 11
io Da ay
Ran is the expected revenue less expected operating cost, discounted by the
required rate of return d over n time periods
R,is the net revenue in time period t
d is the required rate of return
t is the current time period
nis the final time period
Then, we use a logistic function to allocate investment among different conversion pathway
options or non-biofuels options based on the relative financial attractiveness of each. This
approach is often used to model consumer choice situations (Train, 2009); in the BSM, we use
logistic formulations to allocate limited resources, such as plant construction capacity, among
competing uses.
10
o(ktenPV |)
5 = 5D (12)
S is the share construction capacity allocated to pathway j
k is a constant
c is a constant scaling factor
As suggested in Figure 4, reinforcing feedbacks (also called “positive” feedbacks) emerge from
the coupling of investment, production, learning, and the maturation of a set of performance
attributes.
Industry Development Maturity in Pioneer Scale Financials
i ‘ ; P terms of... i . A
Multiple Regions » | Multiple Regions
Multiple Scales Process Yield Pro Forma Financials
+ Pilot Input Capacity Net Present Value of “Next” Plant
> . Demo |__} Capital Cost Growth
OLAS Investor Risk Premium -— — — —
commercial Debt Financing Access Scaler
Learning Curve Dynamics Multiple Technologies/Multiple Regions
»
Fuel Production | Pro Forma Financials ri
Net Present Value of “Next” Plant
Industry Production and Capacity
Multiple Technologies/Multiple Regions
: Net Present Value of “Next” Plant
Pioneer and Commercial Scale
Capecty nadiions Allocation of Plant Construction Capacity
Initiation of Construction of Discrete Plants
Figure 4. The chart is an overview of key concepts, i and p in the BSM. Rei eedbacks
around investment, production, and learning drive bi ss-to-biofuel conversion i pathway and industry
evolution.
Sensitivity study design
The BSM is a dynamically complex model that is rich in feedback and nonlinearities (Peterson et
al., 2013). To investigate the impact of learning, investment, and techno-economic inputs on
biofuel industry development, we extracted from the BSM a simplified model, focusing on
biomass-to-biofuel conversion processes that occur at biorefineries. While the BSM considers
ten U.S. regions, this simplified model considers only one region and only three conversion
technology pathways (labeled “A,” “B,” and “C”), as opposed to more than ten conversion
technology pathways in the larger model. Costs of feedstocks entering the biorefinery and prices
of products that the biorefinery produces are defined as scenario inputs. The model is configured
such that the three hypothetical pathways under consideration differ in their n” plant techno-
11
economic characteristics, their initial levels of maturity, and their pre-commercial investment
profiles. These baseline conditions serve as a springboard for a sensitivity analysis in which we
vary select techno-economic attributes, learning rates, and pre-commercial investment scenarios
(Table 1). The sensitivity ranges are relatively narrow (McDonald and Schrattenholzer, 2001),
but sufficient to cause substantial variation in the key result, technology market share.
The techno-economic attributes were selected to mimic real trade-offs among capital, feedstock,
and other operating costs, without specifically representing existing technologies. For example,
there are actual technologies with higher first-costs and lower operating costs, higher feedstock
costs and higher co-product sales, lower operating costs and lower yield, and lower maturity and
lower expected future performance. In the baseline condition:
¢ Pathway A: This conversion technology pathway represents an extremely promising
pathway that is very early in its development.
* Pathway B is quite promising as well (although its expected yield at maturity is lower
than A) and is closer to commercialization than A.
¢ Pathway C is very close to commercialization but has a very high capital cost and
significantly lower yields. On the other hand, pathway C appears to have a favorable
variable cost structure relative to the other two options.
The sensitivity variations perturb the attractiveness of the pathways along one or more
dimensions. Initial technical maturity (Mp, p, or c) and progress ratio (PR) are the learning
attributes in the sensitivity variation. Initial technical maturity is the initial value of the maturity
index that was described in the learning section. It has a value between 0 and 1, where 1
represents full n‘” plant maturity and associated performance. The selected progress ratio value of
PR = 0.75, and the sensitivity value of PR = 0.79 are within the range of historical progress ratios
for starch ethanol and chemical processes. Pre-commercial investment projects at pilot and
demonstration scales are projects not intended for commercial production that contribute to the
initial learning of each pathway, as described above. Simulations are conducted against a policy
backdrop designed to stimulate initial take-off of all pathways. Policies consist of loan
guarantees and capital cost grants for pioneer-scale facilities that expire in 2022. An initial
product subsidy applies to each gallon of production, up to the first 1 billion gallons of industry
output; this is reduced to a nominal amount for the duration of the simulation. It is important to
note that this is a hypothetical policy set. The policies are derived from a scenario library
developed elsewhere in the BSM project. For details on these scenarios, see Inman et al. (2014).
12
Table 1. Conditions for baseline simulation.
Pathway A | Pathway B | Pathway C | Sensitivity
Baseli Baselii Baseli Variation
Mature Feedstock Input | 2,000 2,000 2,000
Industry (dry short ton/day)
Techno- Fixed Capital | 300,000,000 | 300,000,000 | 400,000,000 | 1.15 x
economics Investment ($) baseline
Fixed Operating | 15,000,000 | 15,000,000 | 15,000,000
Cost ($/yr)
Other (non- | 50,000,000 | 40,000,000 | 5,000,000
feedstock)
Variable
Operating Cost
($/yr)
Co-Product Sales | 5,000,000 16,000,000 | 10,000,000
Revenue ($/yr)
Process Yield | 100 90 66 0.85 x
(gal/dry short ton) baseline
Initial Pilot 0.1 0.5 0.85 0.85 x
Technical Demonstration 0 0.5 0.75 baseline
Maturity (unit- (pathway
less) “B” only)
Commercial 0 0 0
Progress Ratio | Pilot 0.75 0.75 0.75 1.05 x
(1/doubling) Demonstration 0.75 0.75 0.75 baseline
Commercial 0.75 0.75 0.75
Pre- Pilot pre-2012:2 | pre-2012:2 | pre-2012:2
Commercial 2012: 4 2012: 4 other yrs: 0
Investment 2013: 4 2013: 4
(projects/yr) 2014: 4 2014: 4
Demo pre-2012:0 | pre-2012:0 | pre-2012:0
2014: 4 2014: 4 other yrs: 0
2015: 4 2015: 4
2016: 4 2016: 4
For the sensitivity study, baseline conditions are modified as shown in the final column. Each combination of
conditions was simulated, resulting in two conditions each for three (fixed capital i , process
yield, and progress ratio) for each of three pathways and two conditions of a fourth parameter (initial pre-
commercial maturity) for pathway B only.
Results and insights
Figure 5 shows simulation results under baseline conditions. With this set of baseline policy
conditions and input values for feedstock cost and product price, the model generates a phased
evolution in which pathway C is first to mature. Over time, growth in pathway C stabilizes,
while pathway B, and then pathway A, grow to significant volumes (Figure 5).
13
(a) Baseline Model Output: Production by Pathway
_ 10
5
= 8
&
§ 6
@ Pathway A =
24
m Pathway B 3
$2
m Pathway C S
io}
0
2011 2016 2021 2026 2031
year
(b) Baseline Model Output: Commercial-Scale NPV
250
_ 150 /
a
——= Pathway A 8 50
=)
——Pathway B 8 -5020 31
——= Pathway C 3
-150 =
-250
year
(c) Baseline Model Output: Commercial Maturity
1
2°
0
I
LF
Lf).
L Lf
ae
2
a
—— Pathway A
—Pathway B
9
iN)
Maturity (unitless)
°
B
——— Pathway C
°
Figure 5. The figure shows baseline results by conversion technology pathway. (a) The first panel shows baseline
production. For the purposes of this study, pathway characteristics were selected to show industry growth for all
three p: under the policy itions that were modeled. (b) The second panel shows baseline
results for commercial-scale NPV by conversion technology pathway. Modeled decisions to invest in commercial-
scale facilities are based on NPV for a new commercial facility constructed in each year. NPV is calculated based on
capital costs, operating expenses, revenue from product sales, and effects of incentives, all of which change over
time. Positive NPV in Figure 8 corresponds with early industry growth in Figure 7. There are 2 discontinuities in
each NPV trace. The first is the result of expiration of FCI loan guarantees. The second is the result of expiration of
14
fuel product startup subsidies that applies to the first 1 billion gallons of industry output. (c) The third panel shows
baseline results for ial maturity by i hnol pathway. The chart shows that pathway C is
initially more mature and obtains early investment that advances its maturity. Pathways A and B are able to compete
sufficiently such that they also mature.
The results from this sensitivity study (see Figure 6) illustrate competition among biomass-to-
biofuel conversion technology pathways, interaction among the sensitivity parameters studied,
and a tendency for a single pathway to gain most of the technology market share. Each
horizontal, colored line represents a different amount of biofuel production in the year 2030 from
one of the conversion technology pathways. The numerical x-axis labels indicate simulation
condition values that are held constant within that column while other sensitivity parameters are
varied. In any given column, there are many different year 2030 production levels for each
pathway. These represent results of different simulations for sensitivity parameters other than the
one for which the column is labeled.
ie
i]
a
u
1}
i}
mt
ii
w
Mh
a
t)
|
tt
u
i
Biofuel Production [gal/ yr]
= =
= & ge =
0.85 1.001.00 1.051.00 1.051.00 1.050.85 + —1.000.85 1.000.85 —1.001.00 —1.181.00
Initial Maturity Progress Ratio Progress Ratio Progress Ratio
8 A B c
1.161.00 1,15
Yield A Yield 8 Yield C FCLA FCI B FoI
Variable for Input Sensitivity
Technological Pathway
A BB mc
Figure 6. The chart is a summary of sensitivity study results. Interpretation of this chart is discussed in the text. In
each column the baseline case is at 1.00 on the x axis. Each horizontal bar in the chart represents the biofuel
production for one of the three pathways (indicated by the color or the bar) in the year 2030 under one set of
conditions in the sensitivity study. Each of the 10 columns labeled with a sensitivity variable name at the bottom
separates the simulation results for a low value of that input variable from those for a high value of that variable,
such that all of the simulations in the study are shown once in each of the 10 columns. The slanted lines indicate
how the mean output of each pathway varies as a function of the sensitivity variable for that column.
Competition
Competition among the pathways shapes results of the sensitivity study. In general, sensitivity
cases that reduce the performance of a given conversion technology pathway affect fuel
production from that pathway the most, and the other pathways tend to benefit. Conversely,
improved performance primarily increases production from the improved pathway, with
15
secondary effects decreasing production from the other pathways. The examples below give
trends in one direction only, but each of these could be expressed either as an increase or a
decrease.
See Figure 6, column 1, “Initial Maturity B.” Sub-column labeled “0.85” has lower initial
maturity than sub-column labeled “1.00.” Production in “0.85” is lower on average than
production in “1.00” as indicated by the slope and endpoints of the orange line.
Production from conversion pathway B drops dramatically if its initial maturity is
lowered, and production from conversion pathways A and C increases modestly on
average.
See Figure 6, columns 2-4, “Progress Ratio A, B, or C.” Sub-columns labeled “1.04”
have higher progress ratios than sub-columns labeled “1.00” for pathway A, B, or C,
respectively. An increase in progress ratio for pathway A, B, or C causes production for
that conversion pathway to plummet, as indicated by the slopes and endpoints of the blue
line in the column “Progress Ratio A,” by the orange line in the column “Progress Ratio
B,” and by the green line in the column “Progress Ratio C.” The other pathways increase
production slightly in each case, as shown by the other two line colors. Pathway A is
more sensitive to changes in its progress ratio than the other pathways, primarily because
of its initial immaturity, as indicated by the steeper slope of the blue line in the column
“Progress Ratio A” relative to the slopes of the orange line in pathway B and green line
in pathway C.
See Figure 6, columns 5-7, “Yield A, B, or C.” Sub-columns labeled “0.85” have a lower
yield than sub-columns labeled “1.00.” A drop in yield results in a dramatic loss in
production capacity for the affected conversion pathway, shown by the slopes and
endpoints of the blue line in the “Yield A” column, the orange line in the “Yield B”
column, and the green line in the “Yield C” column. The drop in yield in a given
conversion pathway generally increases the capacity of the other pathways. This is shown
in the figure because the slopes of the two lines for the unaffected pathways are negative,
but the slope for the affected pathway is positive. For example, in the “Yield A” column,
production for pathway B (orange) and pathway C (green) is higher in sub-column “0.85”
than in sub-column “1.00.” The increase may or may not be sufficient to make up the
production capacity loss. For example, a drop in yield in pathway A or C could cause an
increase in pathway B that is sufficient to make up the overall production loss because
production from pathway B generally can increase more rapidly and attain higher levels
than from pathways A and C.
See Figure 6, columns 8-10, “FCI A, B, or C.” Sub-columns labeled “1.15” have a higher
capital cost than sub-columns labeled “1.00.” An increase in the capital cost of pathway
A causes much lower production capacity from pathway A, as shown by the negative
slope of the blue line in the “FCI A” column. Pathways B and C only increase their
production capacity a little in response, with the orange and green lines having a small
positive slope. Conversely, the production capacity of conversion pathway A benefits
greatly from increases in the capital cost of the other pathways, especially for increases in
FCI C, as shown by the relatively large positive slope of the blue lines (for pathway A) in
column “FCI C.” Pathways B and C show drastic loss in production at higher capital
16
costs, with large negative slopes for the orange (B) and green (C) lines in columns “FCI
B” and “FCI C,” respectively. Pathway B, and especially pathway C, are less influenced
by increased capital costs in the other conversion pathways, as shown, for example, by
the relatively low positive slope of the green lines (for pathway C) in the “FCI A” and
“FCI B” columns.
Note that this simple model does not include competition among pathways for feedstock or for
end users. Studies with the BSM have indicated that these can be significant factors under some
scenario conditions.
Interaction
The input parameters exhibit interacting effects. Every combination of input parameters (fixed
capital investment, process yield, progress ratios, and pre-commercial investment) shows
interaction in its effects on simulation results. These interactions are not simple linear
relationships, but rather, nonlinear dependencies rooted in the feedback mechanisms inherent in
the dynamics of coupled learning and investment. Small differences in progress ratio, cost, and
yield shape the evolution of the system through the feedback structure of the model in a complex
and nonlinear manner.
Reducing a pathway’s assumed performance does not necessarily lead to increased production
capacity for competing pathways, but in certain situations, an increase may be dramatic. For
example, Figure 6 shows a small subset of the sensitivity studies in which dramatic changes may
be observed. These situations are likely conditions that are favorable for industry growth, such
that the growth opportunity will be met by the most competitive pathway, and lower
performance of one pathway will open an opportunity for another.
Precom Mat B= 0.85 Precom Mat B= 4
P FOIA=4 FoI A= 1.15 FOLA=1 FOIA= 1.15
S foles i Foles 118 Fcie= 1 FOB= 115 Foes 1 oles 115 Foleat Fels 115
2 2 pce 4| 148 Foe Force 1 148 FolG=1 118 14a roice 1 tas roc 115
2
~0851 2
5 2
Fs 2 ht ch
3 s
<
3 o
Oo, 1 @
2 L
Technology A ersanersis: B fersenieay c
Figure 6. The chart is a subset of sensitivity analysis results showing examples of significant change with change in
pathway performance.
17
Market dominance
In our sensitivity analysis, the case where all three pathways exhibit high output is the base case,
which was selected with that goal. There are only a handful of cases where two exhibit high
output. In the cases where two pathways exhibit high output, each pathway must remain
sufficiently competitive with the other to attract investment during key periods of market
development. This tendency toward market dominance is a key feature of the behavior of this
model, given the reinforcing feedback as encoded in the logistic, techno-economic, and learning
parameters used in this study. In Figure 7, points at the vertices of the triangle are dominated by
a single pathway. The figure 7 shows that many of these single-pathway simulations are also
those with higher biofuels production output (shown by larger, darker blue circles). A few of the
high-output cases have high production from 2 pathways (points along one of the sides of the
triangle). High output cases do not have pathway C as the largest-producing pathway, but include
cases with dominance from A, from B, or combined production with either A or B having the
largest production. Figure 8 zooms in on a small subset where more than one pathway develops.
100% Technology C
Total Biofuels Production [gal/yr] &
60,000,000
2,000,000,000
4,000,000,000 e
6,000,000,000
8,000,000,000 : g
9,000,000,000 wk | °
60M 98 a)
.
. e
‘
2
°
J |e # %
‘ * ®
°
g . 4
'
Ps .
° - aes
lo fs
8 "4 ¥
- . % 5
e, *
@ 2 Oy 2.2 2 °Q
° ° ° ° b/
C25 o-.—t2 oS
100% Technology A 100% Technology B
Figure 7. This figure shows total biofuels production in 2030 and the distribution of biofuels production among
pathways. The points that fall at the vertices show dominance by a single pathway; points that fall along a side have
production from the two p ‘S$ COr ing to the two adjoining vertices; and points in the middle share
production among all three pathways.
18
Precom Mat B / FCIA / FCIB/ FCIC
Precom Mat B = 1
< a o FCLA= 1
z z z FCIB= 1 FCI B= 1.15
o 3 3
= = = FCI C= 1 1.15 FCLC=1 1.15
ao
©
1 0.85 0.85 mo
+
ao
c S
S a
= ©
a
S
3 2 /
© +
Ss am oo am
7 a
3 i)
e-4 a
Kg 2.
a
2 1 0.85
o +
3
8 a ZL SL
8 S
ao
©
1 fea}
+
a ZL 7 27 Al
S ..
8 3.8 8 8 3 8 8
8 R18 g 8 RR 8
Mi Technology A Mi Technology B Technology C
Figure 8. The chart is a subset of sensitivity analysis results showing
production.
of multiple p:
In this discussion, market dominance refers to dominance of a technology, not dominance of a
single firm. This conceptualization of market dominance is consistent with a competitive market.
This is application of the “competitive exclusion principle” (Meadows, 2008) and an example of
the “success to the successful” system archetype as described in Senge (1990). Figure 9 shows
this as a causal loop diagram, illustrating in greater detail the reinforcing feedback shown in
Figure 2. Pathways can dominate through combinations of superior mature techno-economics,
higher learning rates, and greater initial maturity. This is one of the fundamental mechanisms by
which a pathway that initially attracts slightly more investment could develop more rapidly and
ultimately dominate the market. The pathway that initially attracts more investment develops
more biorefineries, produces more fuel, and advances down the learning curve through these
experiences. This increase in maturity, translated into more favorable cost and performance,
makes it even more attractive to future investment. Annual investment in construction of new
biorefineries is assumed to be constrained. Pathways compete directly for this investment, so the
reinforcing feedback can cause favorable pathways to attract most of the investment and
dominate the market. Even without complete market dominance, successful pathways have a
tendency to suppress less favorable ones, so it is rare that all pathways succeed equally well. This
can only happen in the unlikely situation where technological parameters are tuned to maintain
sufficient competitiveness among all pathways.
The implications of the tendency toward market dominance of one or two pathways for actual
industry development are challenging to assess. In this simple model, the tendency for market
19
dominance may occur in part because of the small number of pathways considered, and perhaps
also because the model does not sufficiently represent details of different market segments,
which might provide competitive advantage to multiple pathways, each in its respective niche.
Production A
# SE Progress Ratio A
Capacity A Maturity a
+
[Delay | n — Plant
attractiveness A
Investment A re
a, Relative Attractiveness A : B
maple B <a — Plant
Le Ts emia B
cau B Maturity ie
“Was * — Ratio B
a Production B
Figure 9. The chart shows key factors in the p between p ys for i . Attractiveness of
investment in a given pathway is determined by mature industry techno-economics, as modified by current state of
maturity. Progress ratio governs the speed by which production translates into increases in maturity. Lock-in
dynamics can result from differences in (1) initial maturity, (2) progress ratios, or (3) nt plant techno-economics
such as capital cost or yield.
The converse of market dominance—pathway failure—is also common. If a pathway does not
reach some minimal level of learning-based feedback, it is unable to attract sufficient investment
and therefore fails to develop. There are thresholds for progress ratios, and to a lesser extent, for
techno-economics and initial maturity, in their contribution to this reinforcing feedback between
industry development and learning-based cost reductions.
Production capacity and maturity both contribute to learning, but on different timescales,
resulting in a sequence of feedback responses. Production capacity, when utilized, directly and
immediately increases learning from experience, a function of production. This is especially
influential in the early years of pathway growth. Maturity influences learning; increased maturity
facilitates investment in new plants, but guarantees neither production capacity construction nor
production of biofuels due to external conditions and competition between pathways.
20
Limitations
This analysis has several important limitations. First, the learning model boundaries include only
the biomass-to-biofuels conversion portion of the supply chain, eliminating interactions with
feedstock supply and fuel demand. This portion of the supply chain is illustrative of the interplay
between investment, technology development, and production. Second, the concept of nf plant
performance is not grounded in purely physical limitations, so quantitative values for n" plant
performance are uncertain. Third, learning rates (or progress ratios) are highly uncertain, and the
full range of possibilities is not explored here. Fourth, learning rates may not reflect omitted
variables. Fifth, modeled markets are simplified, eliminating niches that may counteract the
tendency to market dominance. Sixth, this novel formulation of learning curves is not validated
elsewhere in the literature. Despite these limitations, we offer this paper and the associated
model as a potentially valuable approach to understanding emerging industry development.
Conclusions
The simplified learning model illustrates implementation of the novel approach to learning that is
used in the full BSM model. This approach contrasts with the single factor learning model in (1)
progress toward mature performance instead of an implicit zero asymptote of (2) multiple
technical attributes instead of only unit cost at (3) multiple development scales. The sensitivity
study shows that the parameters studied (fixed capital investment, process yield, progress ratios,
and pre-commercial investment) exhibit highly interactive effects. The system tends to show
market dominance of a single pathway due to competition and learning dynamics. While these
results do not predict biofuels industry development, they illustrate the types of behavior that
could occur for different combinations of techno-economic and learning parameters for biomass-
to-biofuel conversion technology pathways. This modeling approach could be broadly useful in
representing the interactions among investment decisions, cost reductions during development,
and utilization and production in the emergence of other industries. The results could inform
future analytic work to explore the robustness of the finding of a tendency toward market
dominance, and the policy implications of this kind of system behavior could be considered.
The coupling of learning and investment, within the context of a dynamic model, can generate
results that show a variety of potential behaviors of the system. These may inform the
development of a shared understanding among stakeholders of possible effects of policy,
external factors, decision making (e.g., by investors, growers, and consumers), infrastructure
constraints, and learning on the development trajectory of the domestic biofuels industry.
Exploration of learning and investment scenarios for biofuels industry development could help
inform consideration of biofuels as part of U.S. energy and climate strategy. For example, the
system behavior showing a tendency toward dominance of single pathways could prompt
discussion about whether this outcome appears likely or is an artifact of the model formulation,
how mitigating circumstances (e.g., more detailed representation of different market segments)
could overcome this tendency, and social costs or benefits of such an outcome. Similarly, the
wide range of potential learning rates could focus discussion on rates of progress that are or are
not conducive to further public or private investment. Through its uses as an analysis platform
and scenario generator, the BSM can inform the pursuit of these biofuels goals.
21
Appendix
Table Al. Indices
for Equations
Index Symbol Definition Variables Using Index
0 Initial value M
n Ultimate value A
i Attributes of technologies A
CDP Scale (Commercial, Demo, Pilot) M
J Technology pathways NPV, S
Table A2. Selected Equations
Eq. No. or | Equation and Variable Name in Model Units Variable Name in Paper
Symbol
: for E=L*
1-(1 a2) az)
4 M= E
M, otherwise
Pilot Maturity Unitless Pilot maturity
0< M<1_ | Demo Maturity Demonstration maturity
Commercial Maturity Commercial maturity
Min Pilot Experience for Learning year Minimum experience for learning
L, Min Demo Experience for Learning
Min Cume Industry Output for Learning
0< PR<1 Pilot Progress Ratios, Demo Project Ratios, Project Ratios | 1/doubling Progress ratio
= = Commercial
Cumulative Pilot Experience year Cumulative experience
E Cumulative Demo Experience year
Cumulative Industry Output billion gallons
22
Eq. No. or | Equation and Variable Name in Model Units Variable Name in Paper
Symbol
6 A,=m,A,,,
A State of Industry Process Yield gal/short ton Process yield
(vector of FS Throughput C, FS Throughput P short ton/day Feedstock throughput capacity
. Expected FCI C, Expected FCI P Capital cost growth
technical ol Ds 0 camer ;
attributes) Req’d Rate of Return as % Investor risk premium
Expected Equity Fraction Access to debt financing
O<m State of Industry Multipliers (one for each technical unitless multiplier
' attribute A)
8 [m=CM.+1DM, +PU-M,))KO-M, )
Cc Mature Commercial Multipliers unitless Mature commercial multiplier
D Mature Demo Multipliers unitless Mature demonstration multiplier
P Mature Pilot Multipliers unitless Mature pilot multiplier
(one for each technical attribute A)
10 NPV, =R,-[I,+C,+T,]
NPV NPV Investment C $ Net present value
R NPV Revs Net of Op Cost C $ Expected revenue — operating costs
vs Initial Equity Investment C $ Initial equity investment
Cc PV Loan Payment C $ Present value of debt
rd NPV Taxes C $ NPV of taxes — tax credits
e(ktenPV;) This equation summarizes several
12 S$, = ———_— calculations in the model.
a] , @(KFCNPV))
S share to biofuels Unitless Share of construction capacity
k offset for attractiveness weighting unitless Constant
c invest attractiveness weighting unitless Constant scaling factor
23
References
Bush, B. 2011. Biomass Scenario Model (BSM) Development & Analysis, presented at the 2011
Office of the Biomass Program Analysis and Sustainability Activities Platform Peer Review,
April 4, 2011.
Bush, B., Duffy, M., Sandor, D., Peterson, S. 2008. Using system dynamics to model the
transition to biofuels in the United States. Presented at the Third International Conference on
System of Systems Engineering, Institute of Electrical and Electronics Engineers, Inc. (IEEE),
Monterey, California.
Chen, X., Khanna, M. 2012. Explaining the reductions in US corn ethanol processing costs:
Testing competing hypotheses. Energy Policy 44, 153-159. doi:10.1016/j.enpol.2012.01.032.
Chum, H.L., Warner, E., Seabra, J.E.A., Macedo, I.C. 2013. A comparison of commercial
ethanol production systems from Brazilian sugarcane and US corn. Biofuels Bioprod.
Biorefining n/a—n/a. doi:10.1002/bbb.1448.
Energy Independence and Security Act of 2007. 2007. Pub L No. 110-140, 121 Stat 1492.
Fiddaman, T.S. 1997. Feedback Complexity in Integrated Climate-Economy Models.
Massachusetts Institute of Technology.
Hax, A.C., Majluf, N.S. 1982. Competitive Cost Dynamics: The Experience Curve. Interfaces
12, 50-61. doi:10.1287/inte.12.5.50.
Inman, D., Vimmerstedt, L., Newes, E., Bush, B., Peterson, S. 2014. Biomass scenario model
scenario library: definitions, construction, and description (Technical Report No. NREL/TP-
6A20-60386). National Renewable Energy Laboratory, Golden, Colorado.
Jeon, C., Shin, J. 2014. Long-term renewable energy technology valuation using system
dynamics and Monte Carlo simulation: Photovoltaic technology case. Energy 66, 447-457.
doi:10.1016/j.energy.2014.01.050.
Junginger, M., Sark, W., Faaij, A. (Eds.). 2010. Technological Learning in the Energy Sector:
Lessons for Policy, Industry and Science. Edward Elgar, Cheltenham, UK; Northampton, MA.
Kouvaritakis, N., Soria, A., Isoard, S. 2000. Modelling energy technology dynamics:
methodology for adaptive expectations models with learning by doing and learning by searching.
Int. J. Glob. Energy Issues 14, 104-115.
Lin, Y., Newes, E., Bush, B., Peterson, S., Stright, D. 2013. Biomass Scenario Model
Documentation: Data and References (Technical Report No. NREL/TP-6A20-5783 1). National
Renewable Energy Laboratory, Golden, Colorado.
24
McCurdy, M., Hegazy, Y., Norland, D., Victor, N., n.d. An Examination of Technology
Learning Method and its Applications Tailored to Industrial Technologies for Converting
Lignocellulosic Biomass to Fuels (No. SR-6A20-63543). National Renewable Energy
Laboratory, Golden, CO.
McDonald, A., Schrattenholzer, L. 2001. Learning rates for energy technologies. Energy Policy
29, 255-261. doi:10.1016/S0301-4215(00)00122-1.
Meadows, D. 2008. Thinking in Systems: A Primer. Chelsea Green Publishing.
Morrison, J.B. 2005. Implementation as Learning: An Extension of Learning Curve Theory, in:
Proceedings of the 23rd International Conference of the System Dynamics Society. Presented at
the 23rd International Conferences of the System Dynamics Society, Boston, MA.
Naim, M.M., Towill, D.R. 1994. System Dynamics and Learning Curves, in: Proceedings of the
12th International Conference of the System Dynamics Society. Presented at the 12th
International Conference of the System Dynamics Society, Stirling, Scotland, pp. 164-173.
Newes, E., Inman, D., Bush, B. 2011. Understanding the Developing Cellulosic Biofuels
Industry through Dynamic Modeling, in: Dos Santos Bernardes, M.A. (Ed.), Economic Effects of
Biofuel Production. InTech.
Peterson, S., Newes, E., Inman, D., Vimmerstedt, L., Hsu, D., Peck, C., Stright, D., Bush, B.
2013. An Overview of the Biomass Scenario Model. Presented at the 31st International
Conference of the System Dynamics Society, Cambridge, Massachusetts.
Senge, P.M. 1990. The Fifth Discipline: The Art and Practice of the Learning Organization.
Doubleday/Currency.
Sterman, J.D., Henderson, R., Beinhocker, E.D., Newman, L.I. 2007. Getting Big Too Fast:
Strategic Dynamics with Increasing Returns and Bounded Rationality. Manag. Sci. 53, 683-696.
doi:10.1287/mnsc.1060.0673.
Train, K. 2009. Discrete choice methods with simulation, 2nd ed. ed. Cambridge University
Press, Cambridge ; New York.
Van den Wall Bake, J.D., Junginger, M., Faaij, A., Poot, T., Walter, A. 2009. Explaining the
experience curve: Cost reductions of Brazilian ethanol from sugarcane. Biomass Bioenergy 33,
644-658. doi:10.1016/j.biombioe.2008. 10.006.
Wright, T. 1936. Factors Affecting the Cost of Airplanes. J. Aeronaut. Sci. 3, 122-128.
25
