Anna Jacobson

Runtime and memory requirements for typical formulations of energy system models increase non-linearly with resolution, computationally constraining large-scale models despite state-of-the-art solvers and hardware. This scaling paradigm requires omission of detail which can affect key outputs to an unknown degree. Recent algorithmic innovations employing decomposition have enabled linear increases in runtime and memory use as temporal resolution increases. Newly tractable, higher resolution systems can be compared with lower resolution configurations commonly employed today in academic research and industry practice, providing a better understanding of the potential biases or inaccuracies introduced by these abstractions. We employ a state-of-the art electricity system planning model and new high-resolution systems to quantify the impact of varying degrees of spatial, temporal, and operational resolution on results salient to policymakers and planners. We find models with high spatial and temporal resolution result in more realistic siting decisions and improved emissions, reliability, and price outcomes. Errors are generally larger in systems with low spatial resolution, which omit key transmission constraints. We demonstrate that high temporal resolution cannot overcome biases introduced by low spatial resolution, and vice versa. While we see asymptotic improvements to total system cost and reliability with increased resolution, other salient outcomes such as siting accuracy and emissions exhibit continued improvement across the range of model resolutions considered. We conclude that modelers should carefully balance resolution on spatial, temporal, and operational dimensions and that novel computational methods enabling higher resolution modeling are valuable and can further improve the decision support provided by this class of models.

The paper addresses the inefficiencies in large-scale energy systems planning models, traditionally based on linear programming (LP) or mixed integer linear programming (MILP). These models often struggle with tractability due to necessary abstractions that compromise result quality. To overcome this, the authors introduce a novel Benders decomposition approach. This method separates investment from operational decisions and decouples operational time steps using budgeting variables, enabling parallel processing of subproblems and accommodating policy constraints over time. This new approach significantly improves runtime, scaling linearly with temporal resolution, and shows marked runtime reductions for all MILP and some LP formulations, varying with problem size. Beyond energy, this algorithm is applicable to planning in domains like water, transportation, and production processes. Notably, it can tackle large-scale problems otherwise intractable. The enhanced resolution achieved through this method reduces structural uncertainty, thereby improving the accuracy of planning and investment recommendations.