(383aj) Long-Term Capacity Planning Under Uncertainty

Long-term capacity planning is a strategic process used to ensure future demand is effectively and efficiently met. Long-term capacity planning involves setting goals and making decisions over an extended period, often years into the future. These decisions focus on determining the production capacity needed to meet changing demands for a final product while still satisfying other concerns of the decision maker, such as costs. Optimization models are often used in these processes to make data-driven decisions that balance multiple constraints, such as costs, resource availability, and emission levels. These models help decision makers understand trade-offs within the defined system enabling a more efficient use of resources and improved long-term outcomes. Long term capacity planning models include forecasting future needs, allocating resources, and setting priorities to achieve the desired goals of the decision maker. As these models are based on uncertain future parameters, such as demand for the final product, accounting for uncertainty within our models can lead to reduced costs.

We developed a deterministic (perfect foresight) and a multistage stochastic programming linear mixed integer capacity expansion model for the energy system of the United States. The model contains constraints on capacity expansion, demand and supply, mass/energy flows through technologies, and conversions of each technology. The model also includes constraints on transportation between and within the regions, the material allowed to be transported by different transportation modes, and the direction material is allowed to flow between the regions. The model is applied to a large-scale problem, using real-world data for the United States. We then developed a sequential decision-making approach to better represent how real-world decision are made when predictions about critical parameters, like energy demand and capital costs, may be incorrect and a decision-maker will need to react to the actual observed value. With this approach, we solve multiple iterative optimization models with updated information at each iteration. This allows for feedback to occur in each iteration, considering any observations of random parameters as well as new information about the system as the horizon moves forward.

We then built upon our model by considering more geographical granularity, estimating congestion in the transmission of electricity through increases in transmission costs, and accounting for real-world characteristics through the sequential decision-making approach. We consider two real-world characteristics: social acceptance of low-carbon and renewable technologies and delays in the construction of a facility that can lead to only a portion of the capacity of a technology to be installed on time. These real-world characteristics can affect the timeline of the deployment of low-carbon and renewable technologies, which can affect our ability to meet the energy demand profile while simultaneously reducing carbon emissions.

Lastly, we focused our study on liquid fuel production in the United States, specifically sustainable aviation fuel (SAF). We developed two large-scale spatially explicit integrated landscape and supply chain design models for SAF based on field level biomass availability data across the United States. The first model is a static model which only considers the planting of fields and the installation of biorefineries at a single point in time. Static models are the most common landscape and supply chain design models for biomass systems. The solutions give a snapshot of what fields should be developed and what types of biorefineries should be built and their location, but the solutions do not account for how parameters change over time. The second model is a temporal model which includes the planting of fields and the installation of biorefineries as well as possible expansions to the fields or biorefineries based on changes in parameters over time, such as the demand of SAF or the yield of biomass.