Technological Learning in the Transition to a Low-Carbon Energy System: Conceptual Issues, Empirical Findings, and Use, in Energy Modeling

Technological Learning in the Transition to a Low-Carbon Energy System: Conceptual Issues, Empirical Findings, and Use in Energy Modeling quantifies key trends and drivers of energy technologies deployed in the energy transition. It uses the experience curve tool to show how future cost reductions and cumulative deployment of these technologies may shape the future mix of the electricity, heat and transport sectors. The book explores experience curves in detail, including possible pitfalls, and demonstrates how to quantify the ‘quality’ of experience curves. It discusses how this tool is implemented in models and addresses methodological challenges and solutions.

For each technology, current market trends, past cost reductions and underlying drivers, available experience curves, and future prospects are considered. Electricity, heat and transport sector models are explored in-depth to show how the future deployment of these technologies—and their associated costs—determine whether ambitious decarbonization climate targets can be reached - and at what costs. The book also addresses lessons and recommendations for policymakers, industry and academics, including key technologies requiring further policy support, and what scientific knowledge gaps remain for future research.

• Provides a comprehensive overview of trends and drivers for major energy technologies expected to play a role in the energy transition
• Delivers data on cost trends, helping readers gain insights on how competitive energy technologies may become, and why
• Reviews the use of learning curves in environmental impacts for lifecycle assessments and energy modeling
• Features social learning for cost modeling and technology diffusion, including where consumer preferences play a major role

• Language: English
• Publisher: Academic Press
• Release date: Nov 25, 2019
• ISBN: 9780128187630

Book preview

Germany

Part I

Introduction and methods

Outline

Chapter 1 Introduction

Chapter 2 The experience curve: concept, history, methods, and issues

Chapter 3 Implementation of experience curves in energy-system models

Chapter 4 Application of experience curves and learning to other fields

Chapter 1

Introduction

Martin Junginger¹ and Atse Louwen¹,²,    ¹Copernicus Institute of Sustainable Development, Utrecht University, Utrecht, The Netherlands,    ²Institute for Renewable Energy, Eurac Research, Bolzano, Italy

Abstract

The ongoing energy transition is driven by a need to mitigate climate change and switch from fossil to low-carbon fuels and renewable energy. However, while technologies such as onshore and offshore wind energy, solar energy, and batteries have made significant progress over the past decades, and they can increasingly compete directly with fossil fuels, their deployment is still only a fraction of what is needed to fully decarbonize our economy—a process that is going to take at least several more decades and is going to require major investments. Also, the intermittent character of especially wind and solar energy will require major changes in, for example, storage of energy (both heat and electricity). How this transition will play out, and which technologies will ultimately become winners and losers are highly relevant questions which this book will help to answer by providing both the individual market deployment and cost reduction trends per technology and the results of modeling a portfolio of energy technologies in various sector models and overall energy models.

Keywords

Experience curve; energy transition; energy technologies; renewables; fossil fuels; cost reductions

Chapter outline

Outline

1.1 Introduction 3

1.1.1 Background and rationale 3

1.1.2 Objectives and structure 6

1.1 Introduction

1.1.1 Background and rationale

It is clear that the further development of various energy technologies is crucial to reduce the emission of greenhouse gases (GHGs), achieve other environmental targets, limit growing global energy demand, and ultimately enable the transition to a low-carbon society—preferably at low costs. These aims can only be achieved when a large number of technologies to supply renewable energy and to save energy become commercially available and thus are at the core of most energy and climate policies worldwide. Important scenario analyses of the world’s future energy system and climate change mitigation scenarios illustrate that technological progress is key to minimizing costs of such development pathways. Given the need for drastic decarbonization, and related substantial investment needs, the political and public debate about the societal costs of this transition is increasing, making it even more important to point out possible cost reductions of novel energy technologies and ultimately the benefits of a low-carbon energy system. Furthermore, the speed of development is essential in order to meet required reductions and supply contributions on time. Many scenarios also highlight the positive economic and security impacts of strong support for research, development, demonstration and deployment of such technologies. Lastly, developing and deploying such energy technologies is seen as a major opportunity for development, (sustainable) industrial activity, and (high-quality) employment. Many (national) policies support both research and development (R&D) and market deployment of promising new energy technologies.

The latter, in particular, will require substantial investment. However, designing such policies effectively (e.g., timing and amount of incentives) has proved to be a challenge. The energy sector and manufacturing industry need strategic planning of their R&D portfolio and have to identify key market niches for new technologies (with or without policy support). Taken together, this situation makes an improved understanding of technological learning pivotal. Currently, most strategies and policies are only based to a limited extent on a rational and detailed understanding of learning mechanisms and technology development pathways. The conditions that provide efficient development routes are subject to much research, for example, in the innovation sciences. However, in addition to what may provide the optimal conditions and settings to achieve technological progress and rapid market deployment, it is clear that a detailed understanding of specific technologies, their performance, and factors influencing their performance are essential in order to design and implement effective policies and strategies. Historically, technological learning has resulted in the improvement of many technologies available to mankind, subsequent efficiency improvements and reduction of production costs, and has been an engine of economic development as a whole. Many of the conventional technologies in use today have already been continually improved over several decades, sometimes even over a century (e.g., most bulk chemical processes, cars, ships, and airplanes). Specifically for the electricity sector, coal-fired power plants have been built (and improved) for nearly a century now, while nuclear plants and gas-fired power plants have been built and developed since the 1960s and 1970s on a large commercial scale. Note that these well-established technologies are also still continuously improved, though this mainly leads to incremental improvements and concomitant cost reductions. Due to this long-term development, the established fossil fuel technologies have relatively low production costs. However, they also have a number of negative externalities, especially the emission of GHGs.

In contrast, many renewable/clean fossil fuel–energy technologies and energy-saving technologies used to have higher production costs, but lower fuel demands and GHG emissions. A few examples are electricity from biomass, offshore wind, and photovoltaics (PVs), and energy-efficient lighting and space-heating technologies. For many of these new technologies the potential for further technological development and resulting production cost reductions is deemed substantial, and relatively high-speed cost reduction occurs compared to the conventional technologies. In the past 10 years, the gap between conventional and new technologies has been (largely) closed, and in some cases breakeven points have been reached. Electricity from onshore and offshore wind parks and large amounts of PV systems already today push out fossil generation units in Germany, as these technologies have no fuel costs. Crucial questions are, however, what will happen when intermittent electricity technologies will gain an even larger market share, making backup capacity and various forms of electricity storage (and associated additional investments) a necessity. Also, many renewable heating and transport technologies cannot yet compete with their fossil counterparts, so for these technologies, further technological progress is essential.

Thus the past and future development in time of production costs of (renewable) energy technologies (and the linked cost of CO2 equivalent emission reduction) are of great interest, as the information allows policy makers to develop strategies for cost-effective implementation of these new technologies.

One approach to analyzing the reduction in production costs employs the so-called experience curve. It has been empirically observed for many different technologies that production costs tend to decline by a fixed percentage with every doubling of the cumulative production. As a rule of thumb, this cost reduction lies between 10% and 30%. To date, the experience curve concept has been applied to (renewable) energy technologies with a varying degree of detail.

The importance of progress in technological development of energy technologies is evident. Many (national) policies support R&D and provide the usually costly incentives for market deployment of targeted energy technologies. However, timing of incentives, the specific design of policy measures, and the amount of support that may be effective for success are very hard to determine. The resulting situation makes an improved understanding of technological learning extremely important. The relevance is clear from the urgency to achieve significant changes in the energy system (both in efficiency and in supply) at a rapid pace, to minimize costs and at the same time achieve competitive performance as soon as possible.

In recent years, much more insight has been gained into how learning regarding energy technologies has been acquired and also how their vital, further improvement can continue in the future. Many of these insights are derived from studies that have employed the experience curve approach. In 2009, Junginger, van Sark, and Faaij compiled a first comprehensive overview of experience curves for various energy technologies (both fossil and renewable). Since then, however, the energy transition has further progressed, and technologies have for the first time been commercially deployed on a significant scale (e.g., LED lamps and electric vehicles), which further matured (e.g., offshore wind and heat pumps) or regained new interest (such as green hydrogen production). In this book, less emphasis than previously has been put on the incumbent fossil energy technologies, and the focus has been put on these new technologies, which are likely to play an important role in the coming decades.

Also, the future energy system is challenged by the intermittent nature of renewables and requires therefore several flexibility options. Still, the interaction between different options, the optimal portfolio, and the impact on environment and society are unknown. It was the core objective of the H2020 REFLEX project to analyze and evaluate the development toward a low-carbon energy system with focus on flexibility options in the EU. The analysis was based on a modeling environment that considered the full extent to which current and future energy technologies and policies interfere and how they affect the environment and society while considering technological learning of low-carbon and flexibility technologies. By assessing the competitiveness of technologies and their interrelation, the cost-effectiveness of the whole system for future years as well as the systemic context demands for a well-founded energy system analysis including an appraisal of technological learning. Within REFLEX, this challenge was addressed by the integration of experience in an integrated energy models system. The main findings and lessons of integrating experience curves in the various sector models are of importance for modelers, but also policy makers and industry.

The renewed need for comprehensive overview of the technological learning progress of various technologies needed for the energy transition and the increasing interlinkage of the electricity, heat and transport sectors and the need to jointly model are the rationales for this book.

1.1.2 Objectives and structure

This book aims to provide an overview of the technological development and cost reductions achieved by the major energy technologies that are expected to be deployed as part of the ongoing energy transition. At the same time, it shows how future cost reductions and subsequent deployment of these technologies may shape the future mix of the electricity, heat and transport sectors. The central concept in this book is the experience curve, which quantifies the (past) cost reductions that occur together with the cumulative deployment of a technology.

The book first explains the concept in detail, including possible pitfalls and a new pedigree approach of mapping the quality of experience curves, discusses how this tool is currently implemented in models (and associated methodological challenges and solutions), and explores new applications of the concept, such as the ex ante assessment of environmental impacts of energy technologies (e.g., reduction of energy use of associated GHG emissions with increasing deployment). In a second part, nine chapters will focus on specific technologies that are relevant for the energy transition (e.g., PV and concentrated solar power, onshore and offshore wind, batteries and electric vehicles, heat pumps, power-to-hydrogen technologies, and space-heating technologies). For each technology the current market trends, past cost reductions and underlying drivers, available experience curves, and future prospects are treated in detail. In the third part of the book the results of various electricity supply, energy demand, and transport sector models play a key role. They also show how the future deployment of these technologies (and their associated costs) will determine whether ambitious decarbonization climate targets can be reached (and at what costs). The final chapter focuses on general lessons and recommendations for policy makers, industry, and academics, focusing on among others what technologies may require further policy support on the short term to have a major impact later on, which investments will be needed, and what scientific knowledge gaps remain for future research.

Chapter 2

The experience curve: concept, history, methods, and issues

Atse Louwen¹,² and Juliana Subtil Lacerda¹,    ¹Copernicus Institute of Sustainable Development, Utrecht University, Utrecht, The Netherlands,    ²Institute for Renewable Energy, Eurac Research, Bolzano, Italy

Abstract

The experience curve is a tool that has a long history and has been broadly applied to a large selection of technologies over almost a century. Building on knowledge of learning processes, such as learning-by-doing, learning-by-searching, and upscaling, it represents technological progress and the resulting decline in technology cost in a simple mathematical form. In this chapter, we discuss the history and origins of the experience curve concept and deliberate on data collection and estimation of experience-curve parameters. Subsequently, we show some main applications of experience curves and discuss key issues that are linked to experience curves.

Keywords

Technological learning; experience curve; learning curve; learning-by-doing; learning-by-searching; one-factor experience curve; two-factor experience curve; multifactor experience curve

Chapter outline

Outline

2.1 Introduction 9

2.2 Learning and experience curves 10

2.2.1 The single factor experience curve 11

2.2.2 Two and multifactor experience curves 12

2.3 Empirical data collection for experience curves 15

2.3.1 Cost versus price data 15

2.3.2 Functional unit 17

2.3.3 Data harmonization 17

2.3.4 Common issues with data collection 18

2.4 Estimation of experience curve parameters 20

2.4.1 Regression method: linear or nonlinear 20

2.4.2 Determining fit accuracy and experience curve parameter errors 20

2.5 Applications of experience curves 21

2.5.1 Direct applications of experience curves for policy makers 23

2.5.2 Indirect application of experience curves for policy: energy and integrated modeling 26

2.6 Main issues and drawbacks of experience curves 27

2.6.1 Nonconstant learning rates and learning rate uncertainty 27

2.6.2 Technology systems and components 27

2.6.3 No explanation for cost reductions and causality 28

2.6.4 Radical innovations 28

2.6.5 Technology quality 29

References 29

2.1 Introduction

Especially since the Industrial Revolution, impressive technological progress has been made. This progress has resulted in the development of many novel technologies, but there are also abundant examples of technologies that have remained in essence the same (at least performing the same function) but which have seen gradual but strong improvements over time. Examples that come to mind are the airplane and passenger car and, more recently, technologies in the computer industry, such as processors and memory chips. Learning is considered a key driver in these examples of endogenous technological change (Junginger et al., 2010). In this chapter, we introduce the concepts of the learning curve and experience curve, mathematical relationships that describe the technological progress for a technology—measured in unit cost reductions—as a result of increases in cumulative production of this technology. We discuss their origins, definitions, key applications, and finally present some main methodological issues and drawbacks.

2.2 Learning and experience curves

Within the context of technological learning, it is important to distinguish two concepts: (1) learning and (2) experience curves. Both refer in some ways to the same phenomenon that, as producers, gain more experience with manufacturing of a product, the costs of production will decrease. However, the exact parameters that the curves describe differ between the two concepts.

The phenomenon of the learning curve was first observed and documented in the 19th century by a German psychologist Hermann Ebbinghaus. He described that learning is an exponential process, meaning that the fastest learning occurs in the beginning and that exponentially more effort is required for subsequent increases in learning (Ebbinghaus, 1885). Ebbinghaus was the first researcher to mathematically document the learning process in an experiment he conducted upon himself. He measured the number of repetitions it took to memorize lists of words and found that those declined in an exponential manner (Ebbinghaus, 1885), as shown in Fig. 2.1.

Figure 2.1 Learning curve made from Hermann Ebbinghaus’ experiments on the number of repetitions required to memorize certain lists of words. Data from: Ebbinghaus (1885).

The first well-documented quantified example of the learning curve in the context of technology costs was published by Wright (1936). When examining the manufacturing of airplanes, Wright stated that the time required (measured in unit labor costs) for each airplane built decreased with a constant percentage every time the cumulative number of airplanes produced doubled. The relation was described by Wright with the equation:

(2.1)

. Wright attributed the unit labor cost reductions to a well-known theory that states that as assembly line workers gain more experience, they become more efficient in their work.

Wright also examined the relation between quantities of airplanes produced and unit material costs, and observed different mechanisms that led to cost decreases. First, by increasing the production quantity, relative amounts of waste decrease. Second, higher quantities allow for more economical purchasing of materials from external suppliers. By combining the curves for labor and materials, and also including overhead costs, Wright stated that the total airplane costs follow a curve that has a steeper slope in the beginning due to the higher contribution of labor costs and which gradually has a less steep slope, as the proportion of material costs to overall costs increases. He saw the value in this concept and the mathematical representation of the cost decrease resulting from the cumulated production experience to be able to assess the cost developments for very large numbers of production.

2.2.1 The single factor experience curve

After Wright, it took quite some time for this theory to become more mainstream. As discussed in Junginger et al. (2010), it was not until the RAND Corporation revisited the subject to study the possibility of cost reductions in production of war materials that gained more prominence. In the 1960s, the concept was broadened and introduced into the field of economics by Arrow (1962), and it was developed further by the Boston Consulting Group (1970) into the concept of the experience curve. Boston Consulting Group (BCG) expanded Wright’s learning curve concept to describe the total cost of products and used it to describe unit cost of a product across a whole industry, rather than within a single company, and called this concept the experience curve to distinguish it from the previous learning curve. BCG included in this theory the combined effects of learning-by-doing, learning-by-researching (more commonly research and development, R&D), scale, and investment. Taking all this into account, the experience curve got the following form:

(2.2)

, and so

(2.3)

is as in Eq. (2.1). As Wright already showed, this power law shows a straight line when plotted on a double-logarithmic scale. With this in mind the equation can also be expressed as a linear equation by expressing it in a logarithmic form:

(2.4)

):

(2.5)

(2.6)

For an LR . An example of an experience curve, for solar photovoltaic (PV) modules, is given in Fig. 2.1. Shown in Fig. 2.2 is the raw, empirical data collected, the derived experience curve, and an example of plotting this data on normal, linear scales and on double-logarithmic or log–log scales. For this dataset, an LR of 23.9% was derived, indicating a decline in price of 23.9% for every doubling of cumulative production of PV modules.

Figure 2.2 Example of experience curves on two different graph scales: normal, linear scales (left), and double-logarithmic or log–log scales (right). Source: Data from Louwen et al. (2018).

2.2.2 Two and multifactor experience curves

As discussed in the previous section, the unit cost reductions observed for a technology, are the result of a combination of learning drivers. In addition, developments in input material prices can also have a large effect on the cost development of a technology. The experience curve as discussed above, which we from now refer to as the one-factor experience curve (OFEC), treats all different drivers and developments essentially as a black box, and thus only describes the observed empirical trend of decreasing unit costs but gives little to no insight in the underlying mechanisms driving these cost reductions. By trying to separate the different drivers of cost reductions, invaluable insight can be obtained on how to influence cost reductions (e.g., from a policy maker’s point of view) and how to explain cost developments that seem to deviate from the long-term trend that is given by the OFEC.

To be able to include these considerations in the experience curve concept, several extensions have been made to the OFEC. The first example, we give here, is the extension of the OFEC to include separately the effects of learning-by-doing and learning-by-researching (R&D), in a two-factor experience curve (TFEC). By taking into account R&D expenditures, either directly or using some proxy metric, these drivers can theoretically be separated and measured:

(2.7)

(2.8)

(2.9)

(2.10)

Here, we obtain two separate LR, being the LR being the LR takes into account the fact that knowledge gained from R&D expenditures generally does not directly result in cost reductions (there is a certain time lag), and that the knowledge stock depreciates over time when no further research is conducted:

(2.11)

describes the depreciation of the knowledge stock, based on the underlying assumption that without further research, the value of knowledge gained from R&D efforts gradually declines, along with its ability to drive down unit costs.

An example of application of the TFEC in the currently developing technology is the study by Kittner et al. (2017), who studied the development of energy storage technology. By taking into account both production volume (as opposed to cumulative production) and innovation activity, they analyzed the effect of economies of scale and learning-by-researching on declining lithium-ion battery prices and found that this TFEC is better able to describe the observed price trends than an OFEC based on either production volume or cumulative production (Kittner et al., 2017). Further detail on this study is given in Chapter 8.

Further examples of extensions of the OFEC were presented by ) in the following generalized equation:

(2.12)

, respectively.

In addition to the TFEC approach discussed earlier, Kittner et al. also studied a MFEC for lithium technology, incorporating raw material prices in addition to economies-of-scale and innovation activity. Although a high correlation was observed for this MFEC, no significant improvement compared to the TFEC was observed.

2.3 Empirical data collection for experience curves

The value of the experience curve concepts stems for a large part from the use of empirical data, giving us an evidence-based description of technology cost reduction as a function of cumulated production experience. In the following sections, we describe guidelines for data collection and harmonization, experience curve parameter derivation, and assessment of data quality.

2.3.1 Cost versus price data

To be able to derive the experience curve parameters as shown in Eqs. (1.2)–(1.6), at least two datasets are required: the development of unit technology costs over time and the development of cumulative production of this technology over time. Until now in our discussions, both the learning and experience curves refer to the development of production cost. However, for most technologies, cost data is not readily available. Hence, most studies of experience curves rather make use of the unit price of