Summary

The Global Energy 2050 Project relies on a ‘model of models’—a series of overlapping population, GDP, energy supply/demand, and consumption/production sub-models initialized at the country level and aggregated at sub-regional, regional, and global levels. In these sub-models, variables such as resource extraction, resource demand, and resource price elasticity interact through a series of non-linear feedback loops called System Dynamics (SD) models.

Using these feedback loops, the Global Energy 2050 model conducts Exploratory Modeling Analysis (EMA) under Deep Uncertainty (DU) to generate hundreds of thousands of futures for an exogenously-designed scenario (referred to as an ensemble of future worlds). As a result, while the model is non-deterministic in its results, it is more capable of identifying actions, policies, or other variables which may be driving the results of particular scenarios by capturing the breadth of possible futures. 

These futures can be examined through a series of Key Performance Indicators (KPIs) produced under each run of the scenario. Depending on the interests of the viewer, each KPI can then be analyzed by the high, low, and median-bounding trendlines of the scenario runs across the examined timeline.
​
The current generation of the Energy 2050 model (V1.1) will improve as additional capabilities to include new feedback loops are developed and adopted, to include rare earth mineral scarcity and certain welfare effects.

Methodology

Exploratory Modeling and Analysis and Deep Uncertainty
The Global Energy 2050 Project uses Exploratory Modeling Analysis (EMA), which, unlike traditional consolidative modeling, is not focused on predicting one or even a handful of futures. Rather, it encompasses those futures along with hundreds of thousands of others, in order to allow analysts to more closely examine the human interventions—policy moves by multiple actors, from government to business, and changing social mores—that nobody can predict with certainty across any discernable energy futures scenario, particularly when the timeline spans multiple decades.

For this reason, Energy 2050 operates under a modeling process called Deep Uncertainty to investigate different possibilities for situations where analysts do not know:
​
(a) the appropriate conceptual models that describe the relationships among the key driving forces shaping the long-term future (e.g. different drivers and underlying structures than today);

(b) the probability distributions used to represent uncertainty about key variables and parameters in the mathematical representations of these conceptual models; and/or

(c) how to value the desirability of alternative outcomes.
​
Traditionally, System Dynamics (SD) is the simulation methodology used to model and simulate dynamically complex issues, like those assessed under Deep Uncertainty, and analyze the resulting non-linear behaviors over time, in order to develop and test the effectiveness (and robustness) of structural policies. The combination of SD and EMA is very useful for broadly and systematically exploring how crises might plausibly develop, analyzing their patterns and impacts, and testing the effectiveness of preventive and curative policies, without neglecting Deep Uncertainty and dynamic complexity.  A more in-depth explanation of SD models can be found below.

EMA, with its algorithm-based workbench, uses exploratory models to generate tens of thousands to millions of scenarios (called an ensemble of future worlds), in order to analyze and test the robustness of policy options—in other words, whether the outcomes are acceptable over the entire scenario space. As such, it can be used to generate insights into both how systems function and the robustness of certain policies, by taking Deep Uncertainty into account and examining the upper, lower, and median bounds of these systems over the examined timeline. In doing so, EMA allows analysts to explore which policies effectively and robustly improve system behavior under Deep Uncertainty.

The strength of EMA and SD is that, despite their inability to tell you the exact future of the geopolitics of energy, they can highlight human actions that would generate desired outcomes across all possible futures. Exploratory modeling can also identify factors that are critical to any future possibility. If, for example, an analyst wishes to avoid one of the multiple futures generated by the model, this exploratory method can explain the critical factors underlying that future. Finally, exploratory modeling can be used to model complex phenomena even with data gaps. For example, when considering the geopolitics of energy, the policy preferences of specific government leaders are important for a true predictive outcome, but while these preferences are unknown thirty or forty years in advance, the ensemble of future worlds can capture, within reason, the range of possible outcomes leading to that long-term future for analysts to examine.
System Dynamics Simulation Models
To complete the Energy 2050 model, System Dynamics (SD) models were developed. SD models are used to simulate non-linear dynamic systems characterized by important feedback effects, accumulation effects, and delays. Feedback effects, accumulations, and delays cause non-linear dynamics over time.

Demand-supply dynamics is a typical example of dynamics caused by (asynchronous) feedback effects: an increase in demand will result, all else remaining the same, in higher prices, which may in turn cause the demand to decrease. On the other hand, higher prices may incentivize producers to increase their supply capacity, which results in more supply, falling prices, or higher demand, though often with a delay. Investment in electric power generation capacity provides an alternative example of a delay and accumulation effect: there is quite a time delay between the planning of new electric power plants and their commissioning, though markets may begin to adjust in the interim, and after installation, electric power plants are part of the installed capacity (i.e., the stock of installed power plant capacity) until they are decommissioned.

In spite of the fact that, mathematically speaking, they are composed of systems of integral equations (or differential equations), SD models are built to be understandable glass-box models: every variable in a model has a real world meaning/counterpart and all relations between variables are causal. The structure of the model should correspond largely to the real system and overall behavior should correspond largely to overall behavior in the real system. SD models are also built following particular representational conventions. They contain four types of variables: (1) state variables, mostly referred to as stock variables, (2) flow variables linked into or linked out of stock variables, (3) auxiliary variables, and (4) parameters or constants. Stock variables are represented by box-like symbols and flow variables by valve-like symbols.

Stock variables are integral equations: they accumulate (i.e., they sum everything that goes in, minus everything that goes out over time). Flow variables are differential equations: they describe the dynamics of the stock variables they are connected to. Auxiliary variables change instantly unless a delay equation is used. Parameters and constants do not change during simulation time but can assume different (initial) values for each simulation experiment.
Variable Organization
Version 1.1 (V1.1) of the model consists, for each of the 217 countries, of a population sub-model, a GDP sub-model that calculates GDP with and without energy demand elasticity effects, economic sectors with their respective sectoral energy demands initialized with sectoral energy data from the IEA , energy supply from fossil fuel extraction (oil, gas, coal) initialized with data from the IEA and BP, electricity generation (oil, gas, coal, nuclear, hydro, wind, solar, biomass) initialized with data from the IEA and BP. Beyond these country-related sub-models, there are sub-models to assign energy demand (oil, gas, coal) from all countries and energy extraction (oil, gas and coal) by producing countries; as well as calculate long-term price effects, emissions,  Key Performance Indicators (KPIs); and activate policies.

The big feedback loops in this model are similar to the aforementioned supply/demand dynamics: more demand results in higher prices, which, without extension of the supply leads to less demand then would have been the case without the price rise. This is the elasticity of energy demand effect. Long-term elasticity of supply is also included in the model. Increases in demand and higher prices both incentivize supply additions, which, if profitable, happen after a delay. In the model, supply-demand dynamics are more complicated though, since gradual shifts between energy sources are possible too. However, choices lead to legacy, not only in terms of installed capacity, but also in terms of the impact on learning effects. Experience curve effects (i.e., learning effects) for power plants are also included in the model, both on a country and world level: each doubling of cumulative installed capacity of a particular type of power plant results in a corresponding percentage decrease in investment cost.

The fossil fuel extraction sub-models and electrical power generation sub-model are detailed. The extraction sub-models contain structures for estimating current and near future extraction needs, planned extraction and unexpected extraction needs within the installed extraction capacity, extending/mothballing/decommissioning extraction capacity, extraction and the effects on reserves, the development of (shale) resources into reserves, and the supply to internal and/or external markets. The electrical power generation sub-model is also very detailed: it captures both plant capacity investments (for eight different types of power plants) and the fraction of power generation with these different power plants. This version of the core energy model is still relatively simple in many respects. Power plants are, for example, not de/commissioned in chunks, nor is a distinction made between base load and peak load.

Another example of simplification is GDP broken down according to the shares of the different sectors in each country’s economy. That is, detailed sub-models of these sectors have not been included in the model, only their respective total energy demands. Sectoral energy GDP elasticities of the seven regional blocks (i.e., East Asia and Pacific, Europe and Central Asia, etc…) have been used to translate GDP growth into sectoral energy demand growth. Agricultural energy demand is currently not dealt with separately. Passenger transport and goods transport are split out, but the current vehicle transition sub-models included here remains very simple (typical S-shaped transitions). A separate model was developed to calculate vehicle transition in more detail. International bunkers are kept constant at current levels. Energy demand from other sectors (households, services, industries, etc…) only shifts gradually—after a normal refurbishment time—between energy sources (i.e., between coal, oil, gas, electricity, and distributed renewables) depending on shifts in relative preferences for these energy sources. These preferences for energy sources are assumed to depend on the income group (high, higher middle, lower middle, low) to which countries belongs. Sectors in low income countries have been assumed to have higher relative preferences for affordability than for reliability and health. Sectors in high income countries have been assumed to have extremely high relative preferences for health and reliability, favoring electricity. Using price elasticity of demand functions, energy demand of each of these sectors also in/decrease with de/increasing prices. Energy demand changes due to price elasticity effects are used to calculate the corresponding “GDP with energy price elasticity effects.”
​
Each of the previous remarks point to suggestions for future extensions and refinements. The model and results would also be improved by additional sub-models that properly capture issues such as urbanization, electrification policies, extraction logics, polities, structures to account for efficiency gains and (un)expected innovations, real energy scarcity in non-OECD countries in terms of the physical availability of energy, welfare effects that are not accounted for in GDP calculations, and potential material scarcity (e.g., rare earths).
 

Explore a Scenario