Topology-aware Piecewise Linearization of the AC Power Flow through Generative Modeling

TL;DR

This paper introduces a generative neural network-based piecewise linear approximation for AC power flow that improves accuracy and enables efficient topology optimization in large-scale power grids.

Contribution

It presents a novel generative modeling approach using neural networks to produce accurate PWL approximations of AC power flow, accounting for topology-related nonlinearities.

Findings

Accurate PWL approximation demonstrated on IEEE test systems.

Enables efficient mixed-integer linear reformulation of power flow constraints.

Achieves competitive topology optimization solutions with low computational effort.

Abstract

Effective power flow modeling critically affects the ability to efficiently solve large-scale grid optimization problems, especially those with topology-related decision variables. In this work, we put forth a generative modeling approach to obtain a piecewise linear (PWL) approximation of AC power flow by training a simple neural network model from actual data samples. By using the ReLU activation, the NN models can produce a PWL mapping from the input voltage magnitudes and angles to the output power flow and injection. Our proposed generative PWL model uniquely accounts for the nonlinear and topology-related couplings of power flow models, and thus it can greatly improve the accuracy and consistency of output power variables. Most importantly, it enables to reformulate the nonlinear power flow and line status-related constraints into mixed-integer linear ones, such that one can efficiently solve grid topology optimization tasks like the AC optimal transmission switching (OTS) problem. Numerical tests using the IEEE 14- and 118-bus test systems have demonstrated the modeling accuracy of the proposed PWL approximation using a generative approach, as well as its ability in enabling competitive OTS solutions at very low computation order.