To manage an electrical grid, operators must solve complex mathematical problems in real time. Whether calculating the most efficient way to route electricity (Optimal Power Flow), mapping the physical layout of the grid (Topology Learning), or determining the current state of the network (State Estimation), the underlying physics are governed by alternating current (AC) power flow equations.
In a standard three-phase power system, electricity is transmitted via three separate conductors, each carrying an alternating current wave shifted by 120 degrees. For decades, the foundational literature in power systems engineering has relied on a powerful simplifying assumption: the balanced network assumption.
The Intuition of the Balanced Assumption
Why do engineers assume balance? If the electrical loads (demands) and generation sources are perfectly identical across all three phases, the currents cancel each other out at the neutral point. Mathematically, this means the entire three-phase system can be modeled as a single, isolated phase.
This single-phase reduction transforms a highly coupled, multi-dimensional matrix problem into a far simpler scalar problem. For example, Steven Low demonstrated in Distributed Optimal Power Flow Algorithm for Radial Networks, I: Balanced Single Phase Case that under a balanced single-phase assumption, the non-convex Optimal Power Flow (OPF) problem for radial networks can be solved using Second-Order Cone Programming (SOCP) relaxations and decentralized optimization. Similarly, Swaroop S. Guggilam et al. leveraged single-phase configurations in Scalable Optimization Methods for Distribution Networks With High PV Integration to linearize power-flow equations, turning a computationally daunting task into a fast, convex optimization capable of handling thousands of nodes.
By ignoring the multi-phase nature of the grid, algorithms for feedback control (Emiliano Dall'Anese et al. in Regulation of dynamical systems to optimal solutions of semidefinite programs: Algorithms and applications to AC optimal power flow) and topology estimation (Yuxiao Liu et al. in Topology and Admittance Estimation: Precision Limits and Algorithms) achieve the mathematical tractability required for real-time operation.
Why the Assumption Breaks Down at the Grid Edge
While the balanced assumption holds reasonably well for high-voltage transmission lines, it falls apart in low-voltage distribution networks—the "grid edge" that connects directly to residential homes.
Residential loads are inherently single-phase; a single house draws power from only one of the three phases. More importantly, the rapid adoption of distributed energy resources (DERs), such as rooftop solar photovoltaics (PVs) and household battery storage, introduces severe spatial and temporal volatility. If one household installs a massive solar array on Phase A, while its neighbor on Phase B has no solar and high consumption, the phases become deeply unbalanced.
When algorithms designed for balanced systems encounter this reality, their accuracy and safety guarantees degrade significantly across several domains:
- Voltage Control and Dispatch Failures: As noted by Emiliano Dall'Anese et al. in Optimal Dispatch of Residential Photovoltaic Inverters Under Forecasting Uncertainties, assuming balance fails to represent the true physics of distribution networks. Advanced control frameworks, such as the distributed model predictive control proposed by Chao Ma et al. in Distributed MPC-Based Voltage Control for Active Distribution Networks Considering Uncertainty of Distributed Energy Resources and A Hierarchical Voltage Control Strategy for Distribution Networks Using Distributed Energy Storage, or the zonal strategies developed by Chuanliang Xiao et al. in Cluster Partition-Based Zonal Voltage Control for Distribution Networks With High Penetrated PVs, face an open research challenge in adapting to phase-specific fluctuations. Similarly, decentralized Volt-Var control models often rely on single-phase simplifications that cannot address real-world three-phase complexities (Geunyeong Byeon et al. in Distributionally Robust Decentralized Volt-Var Control with Network Reconfiguration).
- Inaccurate State Estimation and Topology Learning: Power System State Estimation (PSSE) and topology mapping rely on measuring voltages and currents to infer the grid's state or structure. If an algorithm assumes a balanced system, it cannot account for the unequal phase loads common in actual networks, leading to reduced estimation accuracy (Shaomei Zhang et al. in A KRR-UKF robust state estimation method for distribution networks) and a failure to capture multi-phase infrastructure complexities (Vassilis Kekatos et al. in PSSE Redux: Convex Relaxation, Decentralized, Robust, and Dynamic Approaches). Physics-informed topology learning models (Deepjyoti Deka et al. in Tractable Learning in Underexcited Power Grids) and graph-based connectivity identification frameworks (Qiuling Yang et al. in Learning Connectivity and Higher-order Interactions in Radial Distribution Grids) remain limited by their reliance on single-phase representations.
- Miscalculated Grid Capacity and Reconfiguration: When estimating how much reactive power support a local grid can offer the main transmission network, single-phase equations compromise the validity of the results (Zhengshuo Li et al. in Robust Estimation of Reactive Power for an Active Distribution System). Even modern machine learning approaches, such as the large language model framework introduced by Panayiotis Christou et al. in LLM4DistReconfig: A Fine-tuned Large Language Model for Power Distribution Network Reconfiguration, face the necessity of transitioning to complex multi-phase simulation environments to accurately model real-world electrical scenarios.
The Modeling Gap: Inter-Phase Coupling
Why not simply run three separate single-phase algorithms side-by-side?
The core reason this strategy fails is inter-phase coupling. In a real three-phase distribution line, the three physical wires run parallel to each other on the same utility poles. Because they are in close proximity, the alternating current flowing through Phase A induces a magnetic field that physically alters the voltage and current in Phase B and Phase C.
This interaction is captured mathematically by the off-diagonal elements of the nodal admittance matrix. As Ye Yuan et al. point out in Inverse Power Flow Problem, extending network reconstruction frameworks to three-phase systems requires explicitly modeling these inter-phase couplings, which dramatically increases mathematical complexity.
When a system is unbalanced, you can no longer assume these cross-terms cancel out or remain constant. The state of Phase A directly depends on the state of Phase B and C through mutual impedances. Modeling this accurately requires moving from scalar equations to full multi-phase matrix equations, significantly increasing the computational burden and breaking many of the neat convex relaxations that power systems engineers have relied on for decades. Bridging this phase imbalance modeling gap remains one of the most critical frontiers in power systems research.
Key Foundational Work
Optimal Dispatch of Residential Photovoltaic Inverters Under Forecasting Uncertainties (2014) — contributed 1 concept elements to this direction.
Regulation of dynamical systems to optimal solutions of semidefinite programs: Algorithms and applications to AC optimal power flow (2015) — contributed 1 concept elements to this direction.