Physics-Aware Heterogeneous GNN Architecture for Real-Time BESS Optimization in Unbalanced Distribution Systems

Battery energy storage systems (BESS) have become increasingly vital in three-phase unbalanced distribution grids for maintaining voltage stability and enabling optimal dispatch. However, existing deep learning approaches often lack explicit three-phase representation, making it difficult to accurately model phase-specific dynamics and enforce operational constraints–leading to infeasible dispatch solutions. This paper demonstrates that by embedding detailed three-phase grid information–including phase voltages, unbalanced loads, and BESS states–into heterogeneous graph nodes, diverse GNN architectures (GCN, GAT, GraphSAGE, GPS) can jointly predict network state variables with high accuracy. Moreover, a physics-informed loss function incorporates critical battery constraints–SoC and C-rate limits–via soft penalties during training. Experimental validation on the CIGRE 18-bus distribution system shows that this embedding-loss approach achieves low prediction errors, with bus voltage MSEs of 6.92e-07 (GCN), 1.21e-06 (GAT), 3.29e-05 (GPS), and 9.04e-07 (SAGE). Importantly, the physics-informed method ensures nearly zero SoC and C-rate constraint violations, confirming its effectiveness for reliable, constraint-compliant dispatch.

💡 Research Summary

The paper addresses the pressing need for real‑time, constraint‑aware dispatch of battery energy storage systems (BESS) in three‑phase unbalanced distribution networks. Conventional deep‑learning approaches for optimal power flow (OPF) or dispatch typically assume balanced conditions or treat each phase independently, which fails to capture the strong coupling caused by mutual line impedances and uneven load distribution. Consequently, such models can produce infeasible dispatch commands that violate state‑of‑charge (SoC) limits, C‑rate constraints, or voltage bounds.

To overcome these limitations, the authors propose a physics‑informed heterogeneous graph neural network (HGNN) framework that explicitly embeds three‑phase information into the graph representation. The distribution network is modeled as a heterogeneous graph G = (N, E) where node types include buses, lines, loads, storage units, and external grid connections. Each node type carries a dedicated feature vector: bus nodes store rated voltage and voltage limits; line nodes contain full three‑phase series impedances (R_a, X_a, R_b, X_b, R_c, X_c) as well as mutual admittances (G_ab, G_bc, G_ca); load nodes hold per‑phase active and reactive power demands; storage nodes encode current SoC, capacity, charge/discharge power limits, and C‑rate; external grid nodes store connection limits. By treating lines as nodes rather than edges, the graph directly captures phase‑coupling effects without resorting to graph‑edge tricks.

The HGNN employs type‑specific message‑passing: for each layer ℓ, a neighbor message is computed as m_i^{(ℓ)} = ∑{j∈N(i)} W^{τ_j→τ_i}{(ℓ)} x_j^{(ℓ‑1)} + AGG(e_{ij}), where W^{τ_j→τ_i} is a weight matrix dedicated to the source‑target node‑type pair, and AGG aggregates edge attributes (e.g., via sum, mean, or attention). Node updates combine self‑transformation and aggregated messages with a ReLU activation. After K layers, each node possesses a multi‑hop embedding that reflects both topological proximity and phase‑specific electrical interactions.

A multi‑task output head architecture is built on top of the final embeddings: (1) a bus head predicts three‑phase voltage magnitudes and angles, (2) an external‑grid head predicts per‑phase active and reactive power at the point of common coupling, and (3) a storage head predicts per‑phase active and reactive power for the BESS. This simultaneous prediction enables the model to learn cross‑task correlations, such as how a voltage deviation influences optimal battery dispatch.

Crucially, the loss function blends standard mean‑squared‑error (MSE) terms with a physics‑informed penalty:

L(Θ) = ∑_{g∈G}