(496d) Long-Term Planning for Energy Transition Under Endogenous Uncertainty

Growing concern for the effects of greenhouse gas emissions has led to the creation of emission-reducing policies by governments around the world. Stricter emission regulations have created an opportunity to advance underdeveloped and underutilized technologies, and to further develop new technologies that share the goal of reducing emissions. Not only is development of new technologies important in achieving sustainability goals, but the way in which new technologies are deployed can have significant impacts on the choice of systems to develop, the cost of implementation, and the ability to meet emission regulations. The decisions about which technologies should be invested in can be difficult to make since they are based on information about the future, which is uncertain. One way to study the effect of uncertainty on energy system models is through sensitivity analysis, as was done in the Net Zero America study [1]. Another method is multistage stochastic programming (MSSP), which accounts for uncertainty through scenarios that represent different outcomes of uncertain parameters [2].

When considering emerging technologies, a source of uncertainty to consider is how the costs will develop over time. Learning curves are used to model the decrease in cost as the total installed capacity of a technology increases. However, the extent to which the cost decreases is uncertain. Uncertainty is classified into two types, exogenous and endogenous. If the uncertainty is exogenous, it is known when the uncertainty will be revealed, and decisions can be made based on when this observation will occur. In the case of endogenous uncertainty, it is not known when uncertainty will be observed and, thus, in MSSP setting when scenarios will become distinguishable; rather, the timing depends on decisions. For example, a cost reduction for a technology will only be observed if that technology is chosen to be invested in; otherwise, it will remain unknown. Previous work in this area has led to an MSSP model for capacity expansion that considers uncertain learning curves and was applied to a small case study of a network of energy generation technologies [3].

In this work, we develop a linear mixed-integer MSSP model and apply it to a large-scale system, using real-world data, with the goal of reaching net-zero carbon emissions in the United States by 2050 [4]. The model includes nine technologies in the transportation sector and twelve technologies in the energy sector for the entire United States over a 30-year time horizon broken into six 5-year periods. Endogenous uncertainty is included in the learning curve of biomass technologies. Additional exogenous uncertainty in the demand of both sectors and policy related to carbon emissions is also included. The model is used to analyze how uncertainty in the cost reduction of emerging technologies will affect the investment decisions into fossil fuel-based processes and their low-carbon alternatives.

Furthermore, given that an energy transition will involve a sequential decision-making process, a rolling horizon approach is applied to both the deterministic (meaning that the problem has perfect foresight when it comes to parameters) and stochastic models, which allows for the models to incorporate feedback. We present a number of auxiliary methods that allow us to simulate this system that has a number of challenging real-world characteristics (e.g., large time delays, lagged capacity installation, impact of social acceptance). This allows for feedback to occur in the next iteration, considering any realizations of random parameters that might have occurred. After all iterations have been completed, a new energy transition (investment timeline), in the presence of feedback, is generated.

When the models are employed to generate a single (predictive) solution, it is seen that in scenarios when costs of emerging technologies decrease to competitive prices, decisions to invest in these technologies should be made earlier to allow for the decrease in costs to be taken advantage of in the future. It is also seen that a wider variety of energy and biomass technologies are expanded when uncertainty is included. Interestingly, we also see lower carbon emissions when uncertainty is considered, because the stochastic model includes scenarios of high energy demand, and in order to meet those demands, more investment must be made into different, including renewable, energy technologies. When a rolling horizon approach is used the cost of the new investment timeline of the stochastic model is slightly lower than the solution obtained using, iteratively, the deterministic model. Importantly, the transition plan generated using the stochastic model still has a wider variety of technologies installed.

We close with a short but broader discussion of energy transitions. Specifically, we discuss how the proposed model allows us to understand how real-world constraints and technology readiness will impact the implementation of different energy transition pathways and how these technologies can affect the probability of meeting certain emission goals at lower cost.

References

1. Larson, E., Greig, C., Jenkins, J., et al. (2021). Net-Zero America: Potential Pathways, Infrastructure, and Impacts, Final Report Summary. Princeton University.
2. Birge, J.R., Louveaux, F. (1997). Uncertainty and modeling issues. Introduction to Stochastic Programming, 54-61.
3. Rathi, T., Zhang, Q. (2022). Capacity planning with uncertain endogenous technology learning. Computers and Chemical Engineering, 164, 107868.
4. “U.S. Energy Information Administration - EIA - Independent Statistics and Analysis,” https://www.eia.gov/state/search/#?5=126&r=false.