Measuring Exploration: Review and Systematic Evaluation of Modelling to Generate Alternatives Methods in Macro-Energy Systems Planning Models

TL;DR

This paper reviews and systematically evaluates various Modeling to Generate Alternatives (MGA) methods in macro-energy systems planning, proposing a hybrid approach that balances exploration breadth and computational efficiency.

Contribution

It provides the first comprehensive review of MGA vector selection methods and introduces a hybrid approach combining the strengths of Random Vector and Variable Min/Max methods.

Findings

Random Vector offers the broadest exploration of the feasible region.

Variable Min/Max yields the most extreme solutions.

The hybrid approach combines exploration breadth and computational speed.

Abstract

As decarbonization agendas mature, macro-energy systems modelling studies have increasingly focused on enhanced decision support methods that move beyond least-cost modelling to improve consideration of additional objectives and tradeoffs. One candidate is Modeling to Generate Alternatives (MGA), which systematically explores new objectives without explicit stakeholder elicitation. Previous literature lacks both a comprehensive review of MGA vector selection methods in large-scale energy system models and comparative testing of their relative efficacies in this setting. To fill this gap, this paper provides a comprehensive review of the MGA literature, identifying at least seven MGA vector selection methodologies and carrying out a systematic evaluation of four: Hop-Skip-Jump, Random Vector, Variable Min/Max, and Modelling All Alternatives. We examine each method's runtime, parallelizability, new solution discovery efficiency, and spatial exploration in lower dimensional (N <= 100) spaces, as well as spatial exploration in a three-zone, 8760-hour capacity expansion model case. Through these tests, we find Random Vector provides the broadest exploration of the near-optimal feasible region and Variable Min/Max provides the most extreme results, while the two tie on computational speed. We thus propose a new Hybrid vector selection approach combining the two methods to take advantage of the strengths of each. Additional analysis is provided on MGA variable selection, in which we demonstrate MGA problems formulated over generation variables fail to retain cost-optimal dispatch and are thus not reflective of real operations of equivalent hypothetical capacity choices. As such, we recommend future studies utilize a parallelized combined vector approach over the set of capacity variables for best results in computational speed and spatial exploration while retaining optimal dispatch.