مدل سازی تغییرات فنی در تجزیه و تحلیل سیستم انرژی: تجزیه و تحلیل مقدمه ای از آموزش با کار در مدل های انرژی پایین به بالا

The main objective of this paper is to provide an overview and a critical analysis of the recent literature on incorporating induced technical change in energy systems models. Special emphasis is put on surveying recent studies aimed at integrating learning-by-doing into bottom-up energy systems models through so-called learning curves, and on analyzing the relevance of learning curve analysis for understanding the process of innovation and technology diffusion in the energy sector. The survey indicates that this model work represents a major advance in energy research, and embeds important policy implications, not the least concerning the cost and the timing of environmental policies (including carbon emission constraints). However, bottom-up energy models with endogenous learning are also limited in their characterization of technology diffusion and innovation. While they provide a detailed account of technical options—which is absent in many top-down models—they also lack important aspects of diffusion behavior that are captured in top-down representations. For instance, they often fail in capturing strategic technology diffusion behavior in the energy sector as well as the energy sector's endogenous responses to policy, and they neglect important general equilibrium impacts (such as the opportunity cost of redirecting R&D support to the energy sector). Some suggestions on how innovation and diffusion modeling in bottom-up analysis can be improved are put forward.

Throughout history technological development has fundamentally changed the structure of the energy sector by making possible the diffusion of new and cheaper technologies. However, the new production processes have also given rise to negative impacts on the environment, perhaps best illustrated by the current concerns about climate change, which is caused (primarily) by the burning of fossil fuels. Somewhat paradoxically, policy makers worldwide now hope that future technological development will solve the problems it has given rise to in the past. In addition, the long-term effects of global warming require policy efforts to be heavily focused on innovation and technological change in the energy sector. For the above reasons there exists a need to understand more closely the process of technical change and how different policy instruments can be used to influence this process. It is generally agreed that energy modelers and analysts do not yet possess enough knowledge about the sources of innovation and diffusion to properly inform policymaking in technology dependent domains such as energy and climate change. Even though the literature on technical change stresses the fact that technological change is not exogenous in the sense that it simply appears as manna from heaven, most energy models still rely on exogenous characterizations of innovation. Specifically, in exogenous representations technical change is reflected through autonomous assumptions about, for instance, cost developments over time or annual efficiency improvements. However, in real life new technologies require considerable development efforts, much of it by private firms. In recent years energy researchers have therefore shown an increased interest in introducing endogenous (induced) technical change into energy system models, often with the purpose of analyzing explicitly the impact of technical change on energy systems. Thus, in such representations technical change is influenced over time by energy market conditions, policies and expectations.1 The overall purpose of this paper is to provide an overview of the literature on introducing endogenous technical change in energy systems models. Special emphasis will be put on surveying recent attempts at integrating learning-by-doing impacts into ‘bottom-up’ (technology-specific) energy systems models with the use of so-called learning curves, and on analyzing the relevance of learning curve analysis for understanding innovation and technology diffusion in the energy sector. In contrast to previous surveys on induced technical change in energy system modeling—see in particular Grubb et al. (2002) and Löschel (2002)—this paper focuses in more detail on the usefulness and the limitations of learning curve applications and the incorporation of learning-by-doing in bottom-up energy models. This focus is motivated by the fact that there now exist a relatively large number of empirical applications, which make a more detailed evaluation of the method possible. In addition, the learning-by-doing hypothesis has gained plenty of empirical support, and the studies surveyed here have been pointed out as some of the most satisfactory approaches towards the characterization of induced technical change (e.g., Grubb et al., 2002). In essence, learning curve analysis in its simplest form highlights and aims at measuring the empirical relationship between the cumulative experience of a given technology and cost reductions. In other words, it is not only the case that a new technology is used because it becomes cheap; it also becomes cheap through increased deployment and learning-by-doing. Empirical studies of learning curves result in so-called “‘learning-by-doing’ rates”, which indicate the cost reduction (in percent) for a given technology resulting from each doubling of capacity (or production). Recently researchers have extended this simple representation by also incorporating R&D expenses as additional drivers of innovation in the energy field. These specifications are typically referred to as two-factor learning curves, and result in so-called ‘learning-by-searching’ rates (see Section 3 for details). The estimated learning rates for different technologies can be implemented in larger-scale ‘bottom-up’ models, which explicitly specify technological options using both technical and economic parameters. In this paper we survey these recent studies and discuss their most important policy implications and limitations. The paper proceeds as follows. In Section 2 we outline two different broad types of energy system models (top-down and bottom-up models) and discuss briefly in what way endogenous technical change can be introduced in these model approaches. The remainder of the paper focuses in detail on bottom-up models, and Section 3 provides an analysis of learning curves and explains in what way such representations of technological change can be integrated into bottom-up models. Section 4 surveys some selected recent studies that have employed bottom-up energy models with endogenous learning. In Section 5 we critically analyze the most significant policy implications as well as important limitations of these research efforts, while Section 6 outlines some concluding remarks and provides some suggestions for future research efforts in the field.

This paper has surveyed and critically analyzed the modeling of endogenous technical change through learning-by-doing in bottom-up energy systems models. Technical progress is likely to be the single most important factor determining the future of the energy systems world wide, and the notion of technical progress as occurring through learning-by-doing has gained considerable empirical support. For the above reasons bottom-up models with endogenized technical change represent important advances in the energy modeling field, and they provide important policy implications about, for instance, the costs and the timing of environmental policy. The most recent applications even permit the analysis of the optimal allocation of R&D expenses among competing technologies as well. However, bottom-up energy models with endogenous learning are also limited in their characterization of technology diffusion and innovation. While they provide a detailed account of technical options and their associated costs—something that is absent in many top-down models—they also lack representation of significant aspects of diffusion behavior (that, however, are captured in many top-down models). For instance, they fail in capturing strategic considerations in the energy sector and in particular the energy sector's endogenous responses to policy instruments. They also neglect important general equilibrium impacts (such as the opportunity cost of redirecting R&D support to the energy sector). This does not imply, though, that we should re-direct research efforts towards top-down models. It suggests instead that bottom-up and top-down models with induced technical change should not be viewed as substitutes but rather as complements.14 It is possible to identify a number of issues that should be addressed to improve modeling efforts aiming at introducing technology learning in bottom-up energy system models. We have noted that many bottom-up models build on the assumption of perfect foresight and thus no uncertainties—about policy, future technical change or economic conditions—in the investment process. In many ways this is a very restrictive assumption, not the least for such a capital intensive industry as energy. As was noted above, this calls for the use of alternative models as well as qualitative analyses that complement (but not replace) bottom-up model results. There exist, however, also ways in which existing bottom-up models could be improved to address some of the issues raised in this paper. First, an important step in the bottom-up modeling process is the identification of the number of technologies to be included. So far (greenfield) investments in new plants dominate these technology portfolios. As was noted above, though, it may be useful to include here also a number of alternatives that represent investment in existing capacity, such as incremental capacity expansion and lifetime extension in nuclear and hydropower, and the conversion of fuels in thermal power. This will add some amount of (additional) path-dependence in the energy system model, and this approach can be used to analyze, for instance, how new investment in renewable energy technologies can be affected by the presence of such incremental investment alternatives. The potential for making use of existing capacity in this way will of course depend on the present structure of the energy system. However, just as technology learning is a crucial parameter for new energy technologies, it is likely to be important for these alternatives as well. Second, the empirical validity of technological learning is clearly established but it has remained difficult to quantify with any precision the exact influence of learning on cost reductions. For this reason there exists a clear need to extend the so far limited research on introducing uncertainty in the learning rates used in bottom-up modeling. One alternative is also to link the exogenous assumptions about learning rates to different policy scenarios, and thus not only to cumulative capacity and investment in R&D. For instance, in a policy scenario characterized by continued confusion about future climate policy (in terms of both stringency and implementation design),15 it may be valid to assume that the learning rates for investment in existing capacity are higher than in a scenario characterized by less uncertainties. Such and other policy-induced assumptions about technology learning could be a welcome addition to the present research on technical change in energy system modeling.