Source-linked AI summary
Learning-Based Power Flow Estimation with Sparse PMU Deployment in Power Systems
Van-Hai Bui, Quang Manh Hoang, Akhtar Hussain, Wencong Su
TL;DR
Dense PMU deployment is costly, and the effects of both PMU count and location on learning-based branch-flow estimation remain insufficiently addressed. The paper uses selected-bus measurements as neural-network inputs and evaluates deployment scenarios; on the IEEE 30-bus system, full deployment achieves RMSE = 0.010 and R2 = 0.999997. The findings point to informative PMU locations as a consideration in balancing deployment cost and estimation accuracy.
Problem
Dense PMU deployment is costly, and learning-based estimation methods often assume broad measurement coverage despite accuracy depending on both the number and locations of PMUs.
Source: §1 · pp. 2–3Method
A neural network uses active and reactive power measurements from selected buses to estimate active power flows across all branches, with multiple PMU scenarios evaluated.
Source: §1 · p. 3Results
RMSE = 0.010 and R2 = 0.999997 under full PMU deployment on the IEEE 30-bus system with 50,000 operating scenarios.
Source: §4 · p. 13Takeaways & Limitations
Strategically placing PMUs at electrically important buses provides higher accuracy than simply increasing PMU count, informing cost-effective deployment strategies.
Source: §4 · p. 13
Abstract
from arXiv · showhide
Accurate monitoring of power flow is fundamental to the secure and reliable operation of modern power systems. Existing data-driven power flow estimation methods typically assume that synchronized measurements are available at every bus, requiring extensive deployment of phasor measurement units (PMUs). Such dense sensing infrastructures substantially increase installation, communication, and maintenance costs. This paper proposes a learning-based framework that minimizes the number of installed PMUs while maintaining high-accuracy systemwide power flow estimation. Instead of relying on measurements from all buses, the proposed method identifies an informative subset of buses whose synchronized measurements are sufficient to accurately reconstruct branch power flow across the network. To this end, a neural network is developed to learn the nonlinear relationship between PMU measurements and system-wide power flows, and multiple PMU deployment scenarios are evaluated to quantify the trade-off between measurement availability and estimation accuracy. The results enable utilities to identify cost-effective PMU placement strategies that balance deployment cost and prediction performance. Case studies on the IEEE 30-bus test system demonstrate that the proposed framework significantly reduces the number of required PMUs while maintaining accurate power flow estimation.
1 Introduction
Dense PMU deployment is costly, while learning-based power-flow estimation has not fully addressed how both the number and locations of PMUs affect branch-flow accuracy. The paper evaluates a neural-network framework that uses measurements from selected buses to estimate system-wide branch flows and assess deployment trade-offs.
1 Introduction
- PMU installation at every bus is generally impractical because hardware, communications, data storage, cybersecurity, maintenance, and integration add cost and complexity.Utilities must balance improved observability against the cost and complexity of dense deployment.Source: §1 · p. 2
- Prior learning-based studies address related tasks, but the cited graph-learning framework focuses on heterogeneous measurement synchronization, state reconstruction, and event detection rather than PMU placement for branch-flow estimation.The paper frames placement effects on direct system-wide branch-flow estimation as a distinct question.Source: §1 · p. 2
- Existing learning-based methods often assume many measured buses or fixed sensing configurations, while the effect of PMU number and location on direct branch-flow estimation remains unresolved.Prior PMU-placement work primarily targets observability, with comparatively little attention to the trade-off between deployment and learning-based estimation accuracy.Source: §1 · pp. 2–3
- The framework estimates system-wide branch power flows from active and reactive power measurements collected at selected buses, directly linking PMU deployment to estimator performance.It evaluates how both the number and locations of monitored buses affect accuracy.Source: §1 · p. 3
- The study evaluates the framework on the IEEE 30-bus system across diverse loading and generation scenarios and analyzes how estimation accuracy changes as PMU availability is reduced.Its deployment analysis seeks informative locations and quantifies the trade-off between measurement availability and prediction performance.Source: §1 · pp. 3–4
2 Problem Formulation
The section formulates branch power-flow estimation as learning from measurements at monitored buses, with power-flow equations supplying simulated training and test data. It defines the neural-network inputs and outputs and the regression metrics used to evaluate predictions.
2.1 Power flow model
- The IEEE 30-bus system and 50,000 simulated operating scenarios provide the test setting and ground-truth branch flows.The power-flow equations are solved for each scenario.Source: §2.1 · p. 4
2.2 Dataset generation and visualization
- Scenarios vary active and reactive loads and generator outputs within ±50% of base-case values, with all branch flows as targets.The dataset is generated by repeatedly solving the power-flow equations.Source: §2.2 · p. 5
- PMU configurations are formed by removing unmonitored-bus measurements from the all-bus dataset to assess the effects of sensor number and location.Figures 3–5 show selected demand and flow distributions and the relationship between total system load and first-line power flow.Source: §2.2 · p. 5
2.3 Learning-based power flow estimation problem
- A fully connected neural network learns the nonlinear mapping from monitored-bus measurements to branch power flows.The inputs include active and reactive power measurements, with generator active-power outputs included and slack-bus active injection excluded.Source: §2.3 · pp. 6–7
- The model predicts active and reactive power flows for all 41 transmission branches simultaneously.The dataset is divided into 32,000 training, 8,000 validation, and 10,000 testing samples.Source: §2.3 · pp. 7–8
2.4 Performance metrics
- MSE measures squared prediction error, RMSE measures average prediction error, and R2 measures the variance in actual flows explained by predictions.R2 = 1 indicates perfect agreement, while R2 = 0 means performance no better than predicting the observations’ mean.Source: §2.4 · pp. 8–9
3 Results and Discussion
Full-PMU deployment provides a high-accuracy benchmark, but reduced-PMU results vary substantially with bus selection. The most informative locations can matter more than the number of installed PMUs.
3.1 Baseline performance using full PMU deployment
- Training and validation losses show stable convergence without noticeable overfitting under full-PMU deployment.The benchmark model was trained for 200 epochs.Source: §3.1 · p. 9, Figure 7 · p. 10
- With all 30 buses monitored, the model reaches MSE 0.001253, RMSE 0.010080, and R2 0.999997; predicted flows closely match actual values.Figures 8 and 9 illustrate the close agreement, including across all 41 branches in the first testing scenario.Source: §3.1 · pp. 9–10, Figure 8 · p. 10, Figure 9 · p. 11
3.2 Impact of PMU quantity
- Case 3 achieves RMSE 1.889 and R2 0.8802, while Case 1 has RMSE 4.490 and R2 0.3232 despite using the same 25 PMUs.Case 5, with 20 PMUs, also outperforms Case 1.Source: §3.2 · p. 11, Table 2 · p. 12
3.3 Impact of PMU locations
- RMSE-increase analysis identifies buses 13, 2, 8, 27, and 22 as the most influential measurement locations.Their exclusion produces the largest RMSE increases; Table 3 lists the top 10 buses.Source: Table 3 · p. 12, §3.3 · p. 12
- Case 1 performs worst among the 25-PMU cases because it omits influential buses, whereas Case 3 retains more informative buses and achieves the highest accuracy.The cited analysis names buses 2, 13, 22, 23, and 27 among those omitted in Case 1.Source: §3.3 · pp. 12–13
4 Conclusions
The study presents a neural-network framework for estimating system-wide branch power flows from limited PMU measurements. On the IEEE 30-bus test system, it reports high accuracy under full deployment and finds that strategically located PMUs can outperform simply increasing PMU count.
4 Conclusions
- On 50,000 IEEE 30-bus operating scenarios, the estimator achieves RMSE = 0.010 and R2 = 0.999997 under full PMU deployment, while strategic PMU placement improves accuracy over simply increasing PMU count.The conclusion identifies informative buses as practical guidance for balancing installation cost and estimation performance.Source: §4 · p. 13