Use-Inspired Power Flow Modeling for Optimization and Dynamics Applications

Managing modern power grids involves solving computationally challenging tasks at short intervals. Linearly approximating the governing complex physical laws is standard practice to lighten the computational burden, albeit with some inaccuracies. Traditionally, linearization design is agnostic of the end use.

However, increased variations in operating conditions highlight the limitations of this one-size-fits-all approach. This NSF project aims to bridge the modeling and application gap by developing novel use-inspired approaches that yield linear models tailored for specific tasks and anticipated operating conditions. Reaching beyond power systems, the project will transform how scientists and engineers linearize complex physical laws to enable real-world applications.

The intellectual merits of the project include the development of application-suited linearizations that enable robust decision-making and accurate simulation studies at scale. The broader impacts of the project include the development of tractable grid operation tools to increase the deployment of distributed energy resources. The fundamental aspects of this project will be used to increase the research participation of undergraduate students.

Existing power-flow (PF) linearizations are typically developed and assessed based on their accuracy compared to the AC PF equations. However, when using PF linearizations for downstream applications, focusing on the application output accuracy is crucial. Closing the PF linearization loop around the end-use, this project will span three intertwined thrusts, developing certifiably optimal PF linearizations tailor-made for i) deterministic optimization, ii) stochastic optimization, and iii) dynamic modeling.

The targeted canonical problems represent high-impact applications such as bulk system dispatch, distributed energy resource management, voltage control, uncertainty management, power system planning, and time-domain simulations. To tackle such diverse challenges, this project will source ideas from bilevel (stochastic) optimization, systems theory, sensitivity analysis, automatic differentiation, system identification, and neural networks.

This project will develop comprehensive, rigorous, and tractable approaches allowing explicit inclusion of input data distributions and end-use application structure into the PF linearization process. Due to the generalizable structure of the proposed research, the project findings are envisaged to contribute towards use-inspired modeling in broader science and engineering disciplines.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.