BOOST-RPF: Boosted Sequential Trees for Radial Power Flow
This paper introduces BOOST-RPF, which tackles power flow analysis in distribution grids by reformulating voltage prediction from a global graph regression task into a sequential path-based learning problem. The key insight is leveraging the radial (tree) topology of distribution networks to decompose them into root-to-leaf paths, then using XGBoost to predict local voltage drops between parent-child bus pairs. This approach aims to combine the speed of machine learning with the size-agnostic, recursive inductive bias of classical solvers like DistFlow.
The paper presents a well-motivated architectural shift that aligns machine learning with the recursive physics of radial power flow. The core hypothesis—that local voltage-drop regularities are more generalizable than global mappings—is supported by the results. The XGB-Parent variant consistently outperforms neural baselines in both accuracy and training efficiency (11-14 seconds vs 216-1420 seconds). However, the paper’s claims around "superior out-of-distribution robustness" require qualification: while BOOST-RPF degrades less severely than MLPs and GNNs, its OOD errors (mean 0.0134 p.u. VM) are still roughly 10-20× worse than in-distribution performance (7.30×10⁻⁴), which is only briefly acknowledged.
The physics-aligned inductive bias is soundly argued and validated. Decomposing grids into paths provides substantial sample efficiency gains—extracting 138,000 training paths from 1,200 grid samples. The comparison between the three XGBoost variants (Absolute, Parent Residual, Physics-Informed) is illuminating: Parent Residual performs best, suggesting that predicting direct voltage drops from parents is a more "learnable" task than correcting LinDistFlow approximations. The method achieves genuine linear $O(N)$ scaling and trains orders of magnitude faster than neural alternatives, making it practical for real-time applications.
The catastrophic failure of the GlobalMLP in OOD testing (RMSE 1.4872 p.u. vs 0.0056 in-distribution) appears suspiciously extreme and may indicate poor baseline implementation rather than inherent architectural limitations. This near-three-orders-of-magnitude degradation risks creating a strawman comparison. Additionally, while BOOST-RPF generalizes better than neural methods, the absolute OOD errors remain substantial for operational use (0.0134 p.u. mean, up to 0.0215 max). The method is strictly limited to radial networks—mesh grids require cycle-breaking heuristics. Finally, the $O(N)$ inference scaling is partly an artifact of Python overhead; the paper acknowledges a C++ implementation could reduce this but hasn't been demonstrated.
The evidence supports the claim that sequential path-based learning outperforms global neural approaches for radial grids. However, comparisons to analytical baselines are selective: DistFlow performs inconsistently across experiments (3.42×10⁻² in Exp 1 vs 2.83×10⁻² in Exp 2), and the comparison to Newton-Raphson regarding computational complexity ($O(N^3)$) is somewhat misleading since sparse factorization reduces this to $O(N^{1.3-1.6})$ in practice—a nuance mentioned but buried in the text. The neural baseline hyperparameters were tuned (Appendix B), but the GlobalMLP’s dimension was fixed at 256 to "prevent parameter count from ballooning," which may have constrained its capacity for generalization unfairly.
The paper demonstrates strong reproducibility practices. Code is openly available at github.com/EOkoyomon/boost-rpf, including experimental pipelines and analysis scripts. Datasets use established benchmarks (Kerber Dorfnetz, ENGAGE suite). Hyperparameters are exhaustively documented in Appendix B, with tuning procedures described for all models. The only barrier to reproduction is the inherent stochasticity of the datasets (Powerdata-gen for Kerber), though parameters are specified. The distinction between decision nodes in XGBoost versus neural weights is helpfully clarified regarding parameter counts.
Accurate power flow analysis is critical for modern distribution systems, yet classical solvers face scalability issues, and current machine learning models often struggle with generalization. We introduce BOOST-RPF, a novel method that reformulates voltage prediction from a global graph regression task into a sequential path-based learning problem. By decomposing radial networks into root-to-leaf paths, we leverage gradient-boosted decision trees (XGBoost) to model local voltage-drop regularities. We evaluate three architectural variants: Absolute Voltage, Parent Residual, and Physics-Informed Residual. This approach aligns the model architecture with the recursive physics of power flow, ensuring size-agnostic application and superior out-of-distribution robustness. Benchmarked against the Kerber Dorfnetz grid and the ENGAGE suite, BOOST-RPF achieves state-of-the-art results with its Parent Residual variant which consistently outperforms both analytical and neural baselines in standard accuracy and generalization tasks. While global Multi-Layer Perceptrons (MLPs) and Graph Neural Networks (GNNs) often suffer from performance degradation under topological shifts, BOOST-RPF maintains high precision across unseen feeders. Furthermore, the framework displays linear $O(N)$ computational scaling and significantly increased sample efficiency through per-edge supervision, offering a scalable and generalizable alternative for real-time distribution system operator (DSO) applications.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.