AES-dagarnaKatrineholm, 6-7 May 2009

Leo SchrattenholzerIn Memoriam

Technology Learning forEnergy Technology Policy

Clas-Otto Wene

Frequency of learning rates for energy technologies(McDonald & Schrattenholzer, 2001)

0

1

2

3

4

5

6

7

8

-14 -10 -6 -2 2 6 10 14 18 22 26 30 34 38 42 46

Learning Rate

Fre

qu

en

cy

20%5%

First compilation for energy technologiesRenewables, fossil, nuclear, energy efficiency

Industry level

PV Power Modules 1976-2001

1

10

100

0.1 1.0 10.0 100.0 1000.0 10000.0Cumulative Global Shipments (MWp)

Glo

bal A

vera

ge P

rice

(US

D(2

001)

/Wp)

Oil crisesSolar vision

Growth 84%

Commercialoff-grid

Growth 12%

"Roof-top"programmes

Growth 35%

(PHOTEX data from Strategies Unlimted)

Technology Learning measured by Experience CurveThree decades, four orders of magnitude

and a deployment roller-coaster

TechnologyLearningSystem

M$ Wp

Price = const · (Cum. Ship)-E

Learning Rate = 1 – 2-ELearning Rate = 20%

Technology Learning: Measurement and Energy Policy

Technology Learning: deploying technologies in competitive markets increases skills and stimulates private R&D, leading to cost reductions and improved technical performance.

Experience/learning curves: measures technology learning when technical properties remains same

Deployment Policy

No Learning without Market Action

Scenario Modelling Path dependence leading to ε/Ω solutions

Cybernetic TheoryTechnology Learning as

eigenbehaviour

How to design Deployment Programmes stimulatingindustry internal processes at low cost to tax payers?

Cumulative Sales

Cos

t

IncumbentCost-efficientTechnology

Challenger

Niche Markets for the Challenger

A

B

Special efforts to create niche markets (labelling, feed-in tariffs)? Is the niche market curve flat enough? Contributions from industry in “A” to have the benefits in “B”?

Japan's PV-Roof Programme: Use of Niche Markets

1.00

10.00

100.00

0.01 0.10 1.00 10.00 100.00 1000.00 10000.00

Cumulative Installations in Japan's PV-Roof Programme (MW)

Pric

e an

d C

ost

(US

D(1

997)

/Wp)

Installation Price

Investor's Cost

Experience Curve PV-modules and BOS

3.0 USD/Wp

1.1 USD/Wp

Using Niche Markets to stimulate Learning Investmentsfrom private sources

(Example Japan Residential PV Systems, IEA (2000))

NicheMarkets

Deployment from New Building Code

0

5

10

15

20

25

30

35

1990 1992 1994 1996 1998 2000 2002Year

Sal

es (

mill

ion

m2)

Experience Curve 1992-2000

10

100

1 10 100 1000Cumulative Sales (Mm2)

Re

lativ

e P

rod

uct

ion

Co

st

(Pe

rce

nt) Uncoated glass

Compound learning system

Announce-ment

Examples of regulation stimulating Technology Learning and measured by Experience Curves

(Wene, 2008a)

Germany 1992-2000: Coated Glass for Selective Windows (Data from Blessing 2002)

TechnologyLearning System

(manufacturing ind.++)

Energy System

The Technology Learning System and the Energy Systemare coupled to each other

Structural coupling: “interlocked history of structural transformation, selecting each other’s trajectories”

(Varela, 1979)

Technology Path with Fossil Fuels, Nuclear and Wind

hydroconv.coal

conv.gas

conv.oil

nuclear

adv.coal

ngcc

wind

0

5000

10000

15000

20000

25000

30000

35000

40000

1995 2005 2015 2025 2035 2045

TWh

Total system cost: 9117 billion US$

Technology Path with Fuel Cells, PV and Wind

hydroconv.coal

nuclear

adv.coal

ngcc

fuel cell

wind

pv

pv-h2

0

5000

10000

15000

20000

25000

30000

35000

40000

1995 2005 2015 2025 2035 2045

TWh

Total system cost:9106 billion US$

Modelling experiment showing effective but alternative paths(Results from Genie model 1997)

The structural coupling between ETLSs and energy system expressed in Experience Curves

have created two very different Least-Cost solutions from identical starting points and assumptions

Critical assessment of Experience/Learning Curves:High-level Reports positive but important caveats

IEA Energy Technology Perspectives ● Key phenomenon for determining future cost of renewable ● State-of-the-art does not permit reliable extrapolations

UK Stern Report ● Can be used to justify deployment support ● Very different learning rates from causes uncertain

Empirical studies: ● Analyse and verify learning rates ● Features, Events, Processes (FEPs) causing technology learning

Theoretical basis: ● Cybernetic Approach proposed - FEP do not explain learning rates

● Modelling the technology production system

Theory: Logic of the argument

► Operational closure: The technology learning system is an operationally closed system.

► Fundamental Cybernetic Theorem: All operationally closed systems develop Eigenbehaviour (von Förster, Varela)

► Operators: Define operators working on the internal state function and compatible with the EC&LC equation

► Eigenvalues: Use the operators to calculate eigenvalues for the system

► Experience Parameter: Interpret the eigenvalues in terms of the experience parameter in the EC&LC equation

A formal view of the present theory

Lim CSRL 0

0 C+k ∞

k

=(2n+1)π 0

0 1

i

1

n = 0, 1, 2, …

i

1

Result from the Theory

Value of E and Learning Rates: Eigenvalue analysis provides

E(n) = 1/[(2n+1)π] for n= 0, 1, 2, 3, …

LR(n) = 20%, 7%, 4%, … for n = 0, 1, 2

Theory reformulates the research question:From “Why is the learning rate X%?” to

“Why are not all learning rates 20%?”

Frequency Distribution of Learning Rates(In firms and by cost; Dutton and Thomas 1984)

0

2

4

6

8

10

12

14

-12-

-11

-8--7 -4

-3

00-0

1

04-0

5

08-0

9

12-1

3

16-1

7

20-2

1

24-2

5

28-2

9

32-3

3

36-3

7

40-4

1

44-4

5

Learning Rate

Fre

quen

cy

Frequency distribution of Learning Rates: 108 cases from individual firms and by cost

Theory predictsLR0 = 20%

Distribution of E valuesDutton and Thomas (1984)

0

20

40

60

80

100

120

-0.30 -0.10 0.10 0.30 0.50 0.70 0.90

E values

Cum

ulat

ive

num

ber

of c

ases

Emean (DT) = 0.3110Etheory (0) = 0.3183

Comparison theoretical and measured distribution:108 measurements in individual firms and by cost

Distribution of E valuesMcDonald & Schrattenholzer (2001)

0

5

10

15

20

25

30

35

40

45

-0.40 -0.20 0.00 0.20 0.40 0.60 0.80 1.00

Cumulative number of cases

E v

alu

es

Comparison theoretical and measured distribution:42 Energy technologies on industry level and by price

Energy technologies/(McDonald & Schrattenholzer, 2001)

0

1

2

3

4

5

6

7

8

-14

-10 -6 -2 2 6 10 14 18 22 26 30 34 38 42 46

Learning Rate

Fre

quen

cy

Higher modes of learning (75%) - Insufficient closure - External perturbations

Price/cost cycle (but only 25% of total cases at E≈0.10)

Developing the cybernetic approach withinthe AES project

► Closure and Eigenvalue - Matrix formulation to include double closure - Phenomena of radical innovation, technology drift, grafted technologies, compound systems, dispersion

► Modelling the Technology Learning System - Feasibility of using Beer’s Viable System Model

► Applications - Cooperation to apply the theoretic approach to a few key technologies (renewables and energy efficiency)

Thank you!

Radical innovation for PV modules

0

1

2

3

4

5

6

7

8

9

10

1980 1990 2000 2010 2020 2030 2040 2050Year

Pric

e (

US

D(2

00

1)/

Wp

)

Radical Innovation

Incumbent in radical innovation scenario

No radical innovation

Competitvein mass markets

Effect of Radical InnovationResetting the cumulative sales (resetting feedback)