This report describes how Learning-by-Doing (LBD) is implemented endogenously in the National Energy Modeling System (NEMS) for generating plants. LBD is experiential learning that correlates to a generating technology's capacity growth. The annual amount of Learning-by-Doing affects the annual overnight cost reduction. Currently, there is no straightforward way to integrate and make sense of all the diffuse information related to the endogenous learning calculation in NEMS. This paper organizes the relevant information from the NEMS documentation, source code, input files, and output files, in order to make the model's logic more accessible. The end results are shown in three ways: in a simple spreadsheet containing all the parameters related to endogenous learning; by an algorithm that traces how the parameters lead to cost reductions; and by examples showing how AEO 2004 forecasts the reduction of overnight costs for generating technologies over time.

For at least the last decade, evaluation of the benefits of research, development, demonstration, and deployment (RD3) by the U.S. Department of Energy has been conducted using deterministic forecasts that unrealistically presume we can precisely foresee our future 10, 25, or even 50 years hence. This effort tries, in a modest way, to begin a process of recognition that the reality of our energy future is rather one rife with uncertainty. The National Energy Modeling System (NEMS) is used by the Department of Energy s Office of Energy Efficiency and Renewable Energy (EE) and Fossil Energy (FE) for their RD3 benefits evaluation. In order to begin scoping out the uncertainty in these deterministic forecasts, EE and FE designed two futures that differ significantly from the basic NEMS forecast. A High Fuel Price Scenario and a Carbon Cap Scenario were envisioned to forecast alternative futures and the associated benefits. Ernest Orlando Lawrence Berkeley National Laboratory (LBNL) implemented these scenarios into its version of NEMS, NEMS-LBNL, in late 2004, and the Energy Information Agency created six scenarios for FE in early 2005. The creation and implementation of the EE-FE scenarios are explained in this report. Both a Carbon Cap Scenario and a High Fuel Price Scenarios were implemented into the NEMS-LBNL. EIA subsequently modeled similar scenarios using NEMS. While the EIA and LBNL implementations were in some ways rather different, their forecasts do not significantly diverge. Compared to the Reference Scenario, the High Fuel Price Scenario reduces energy consumption by 4 percent in 2025, while in the EIA fuel price scenario (known as Scenario 4) reduction from its corresponding reference scenario (known as Scenario 0) in 2025 is marginal. Nonetheless, the 4 percent demand reduction does not lead to other cascading effects that would significantly differentiate the two scenarios. The LBNL and EIA carbon scenarios were mostly identical. The only major difference was that LBNL started working with the AEO 2004 NEMS code and EIA was using AEO 2005 NEMS code. Unlike the High Price Scenario the Carbon Cap scenario gives a radically different forecast than the Reference Scenario. NEMS-LBNL proved that it can handle these alternative scenarios. However, results are price inelastic (for both oil and natural gas prices) within the price range evaluated. Perhaps even higher price paths would lead to a distinctly different forecast than the Reference Scenario. On the other hand, the Carbon Cap Scenario behaves more like an alternative future. The future in the Carbon Cap Scenario has higher electricity prices, reduced driving, more renewable capacity, and reduced energy consumption. The next step for this work is to evaluate the EE benefits under each of the three scenarios. Comparing those three sets of predicted benefits will indicate how much uncertainty is inherent within this sort of deterministic forecasting.

Small-scale (100 kW-5 MW) on-site distributed generation (DG) economically driven by combined heat and power (CHP) applications and, in some cases, reliability concerns will likely emerge as a common feature of commercial building energy systems in developed countries over the next two decades. In the U.S., private and public expectations for this technology are heavily influenced by forecasts published by the Energy Information Administration (EIA), most notably the Annual Energy Outlook (AEO). EIA's forecasts are typically made using the National Energy Modeling System (NEMS), which has a forecasting module that predicts the penetration of several possible commercial building DG technologies over the period 2005-2025. Annual penetration is forecast by estimating the payback period for each technology, for each of a limited number of representative building types, for each of nine regions. This process results in an AEO2004 forecast deployment of about a total 3 GW of DG electrical generating capacity by 2025, which is only 0.25 percent of total forecast U.S. capacity. Analyses conducted using both the AEO2003 and AEO2004 versions of NEMS changes the baseline costs and performance characteristics of DG to reflect a world without U.S. Department of Energy (DOE) research into several thermal DG technologies, which is then compared to a case with enhanced technology representative of the successful achievement of DOE research goals. The net difference in 2025 DG penetration is dramatic using the AEO2003 version of NEMS, but much smaller in the AEO2004 version. The significance and validity of these contradictory results are discussed, and possibilities for improving estimates of commercial U.S. DG potential are explored.

This report describes LBNL's approach for assessing uncertainty in any National Energy Modeling System (NEMS)-related analysis. Based on years of experience using LBNL-NEMS for various analyses, LBNL developed an alternative approach that aims to provide a simple yet comprehensive perspective of how the results behave under a given set of what we believe to be some of the issues important to large-scale energy modeling. This project has established a standard set of eight sensitivity cases that can be run overnight and are highly likely to produce stable and interesting results. The goal was to establish a limited number of interesting sensitivity cases that would routinely produce adjunct results to LBNL-NEMS reporting that will be of value to our readers. These cases will be routinely reported together with future LBNL-NEMS results in the form of a standard output table. As an example, this work uses a Government Performance and Results Act (GPRA) analysis run as the baseline, but the goal is to establish a standardized set of cases that would change little over time and be applicable to other analyses in addition to GPRA. The approach developed here cannot serve as a substitute for a sensitivity analysis tailored to the question at hand, but it can provide a fast review of some areas that have proven to be of interest in the past.

This memo explains what Berkeley Lab has learned about how the residential central air-conditioning (CAC) end use is represented in the National Energy Modeling System (NEMS). NEMS is an energy model maintained by the Energy Information Administration (EIA) that is routinely used in analysis of energy efficiency standards for residential appliances. As part of analyzing utility and environmental impacts related to the federal rulemaking for residential CAC, lower-than-expected peak utility results prompted Berkeley Lab to investigate the input load shapes that characterize the peaky CAC end use and the submodule that treats load demand response. Investigations enabled a through understanding of the methodology by which hourly load profiles are input to the model and how the model is structured to respond to peak demand. Notably, it was discovered that NEMS was using an October-peaking load shape to represent residential space cooling, which suppressed peak effects to levels lower than expected. An apparent scaling down of the annual load within the load-demand submodule was found, another significant suppressor of the peak impacts. EIA promptly responded to Berkeley Lab's discoveries by updating numerous load shapes for the AEO2002 version of NEMS; EIA is still studying the scaling issue. As a result of this work, it was concluded that Berkeley Lab's customary end-use decrement approach was the most defensible way for Berkeley Lab to perform the recent CAC utility impact analysis. This approach was applied in conjunction with the updated AEO2002 load shapes to perform last year's published rulemaking analysis. Berkeley Lab experimented with several alternative approaches, including modifying the CAC efficiency level, but determined that these did not sufficiently improve the robustness of the method or results to warrant their implementation. Work in this area will continue in preparation for upcoming rulemakings for the other peak coincident end uses, commercial air conditioning and distribution transformers.