CANOS: A Fast and Scalable Neural AC-OPF Solver Robust To N-1 Perturbations

TL;DR

A deep learning system (CANOS) is trained to predict near-optimal solutions of the AC-OPF problem without compromising speed, paving the way for more efficient optimization of more complex OPF problems which alter grid connectivity such as unit commitment, topology optimization and security-constrained OPF.

Abstract

Optimal Power Flow (OPF) refers to a wide range of related optimization problems with the goal of operating power systems efficiently and securely. In the simplest setting, OPF determines how much power to generate in order to minimize costs while meeting demand for power and satisfying physical and operational constraints. In even the simplest case, power grid operators use approximations of the AC-OPF problem because solving the exact problem is prohibitively slow with state-of-the-art solvers. These approximations sacrifice accuracy and operational feasibility in favor of speed. This trade-off leads to costly "uplift payments" and increased carbon emissions, especially for large power grids. In the present work, we train a deep learning system (CANOS) to predict near-optimal solutions (within 1% of the true AC-OPF cost) without compromising speed (running in as little as 33--65 ms). Importantly, CANOS scales to realistic grid sizes with promising empirical results on grids containing as many as 10,000 buses. Finally, because CANOS is a Graph Neural Network, it is robust to changes in topology. We show that CANOS is accurate across N-1 topological perturbations of a base grid typically used in security-constrained analysis. This paves the way for more efficient optimization of more complex OPF problems which alter grid connectivity such as unit commitment, topology optimization and security-constrained OPF.

Figures (5)

• Figure 1: CANOS input (left) and output (right) graph structures. Different node and edge types are represented in different colors. The output graph only contains entities with predicted features.
• Figure 2: CANOS architecture. The colored layers have trainable weights, while the striped module is a non-parameterized graph function. The decoder outputs the voltage angle $\texttt{va} = \angle (V_i V^*_j)$ and voltage magnitude $\texttt{vm} = |V_i|$ as bus node features, as well as quantities relating to generator dispatched power (real power $\texttt{pg} = \Re(S^g_k)$ and reactive power $\texttt{qg} =\Im(S^g_k)$) as generator node features. The branch flow derivation module uses these predicted values to compute the branch complex power in the two directions (real and reactive powers $\texttt{pf}$, $\texttt{qf}$, $\texttt{pt}$, $\texttt{qt}$) according to Equations \ref{['eq:constr5']} and \ref{['eq:constr6']}.
• Figure 3: Speed comparison of AC-IPOPT, DC-IPOPT and CANOS.
• Figure 4: Supervised performance as a function of Number of Message Passing Steps: Both the TRMAE (left) and MSE (right) decrease as the number of message passing steps (and consequently number of parameters) increases in the model. These metrics on the validation split of the data throughout training.
• Figure 5: Feasibility as a function of Number of Message Passing Steps: Feasibility substantially improves as we increase the number of message passing steps.