Contribution type: Lightning talk and Poster
Smart-Linking of Power, Gas, and Hydrogen Models via Optimization Proxies
Presenter: Emad Nematbakhsh
Description:
The investment and operation of multi-carrier energy networks is extremely complex. Hard-linking these systems is computationally intensive, whereas soft-linking struggles to fully capture their interactions. This work introduces an intermediate “smart-linking” solution that overcomes these limitations. We create a proxy of the gas and hydrogen networks and embed it as a set of optimization constraints within the electricity network model, transforming the multi-level problem into a solvable single-level optimization. To accurately capture the varying operational contexts that affect coupling variables, we apply factor selection to include both local and global features. Ultimately, this allows the electricity system operator to analyze network issues at high speed and in a scalable manner.
Background:
Supervised learning methods typically train proxies on pre-solved datasets, which imposes a massive computational burden for realistic multi-carrier systems. To address this, our methodology utilizes decomposition methods (such as Benders decomposition) to generate training data. This iterative process generates a rich trajectory of solutions rather than just a single final endpoint. By intelligently sampling operationally meaningful intermediate states and transition regions, the proxy achieves superior accuracy across a wider operating space compared to blind, pre-defined sampling strategies.
Optional links: