Maximum likelihood estimation of distribution grid topology and parameters from smart meter data

Lisa Laurent, Jean-Sébastien Brouillon, Giancarlo Ferrari-TrecatearXiv 2022 · OpenReview

TL;DR

This paper adapts a powerful Maximum Likelihood Estimation (MLE) technique to estimate the electrical network topology and parameters of distribution grids using only readily available smart meter data (voltage magnitudes and power injections). The key innovation is handling the absence of phase information, a common limitation of smart meters, showing it introduces a manageable ~30% error increase in the estimated network parameters.

What Problem Does This Paper Address?

Modern power grids are becoming increasingly complex with the integration of distributed energy resources like solar and wind. This shift from centralized to decentralized power flow necessitates a detailed understanding of the grid's topology and the electrical characteristics of its components, often represented by the admittance matrix. Traditionally, accurate grid parameter estimation relied on synchronized measurements from specialized devices like Phasor Measurement Units (PMUs). However, the widespread deployment of smart meters, which offer voltage magnitude and power measurements but lack precise time synchronization and phase information, presents a significant challenge. This paper addresses the gap of estimating grid parameters using this more accessible, yet less informative, smart meter data.

Methods

The authors adapt a Maximum Likelihood Estimator (MLE) previously developed for PMU data. The core challenge is to reformulate the MLE for smart meter data, which lacks voltage and current phase information. They achieve this by:

  1. Linearizing Power Flow Equations: They use approximations for small voltage angle differences and small voltage magnitudes to linearize the power flow equations.
  2. Adapting the MLE Constraint: The original MLE constraint, which relies on precise phasor measurements, is modified to accommodate the linearized power flow equations derived from phase-less smart meter data. This involves making assumptions about the noise characteristics of the measurements.
  3. Numerical Implementation: The adapted MLE is implemented and tested on a standard IEEE 33-bus benchmark network. They also compare their method against a linear regression approach and analyze the impact of noise.

Datasets

The paper uses the IEEE 33-bus benchmark network for evaluation. This is a commonly used test case in power systems research, representing a typical distribution grid with 33 buses and 32 branches. The simulations involve generating load profiles with Gaussian noise and adding Gaussian noise to voltage and power measurements to mimic real-world conditions.

Key Insights and Ideas

  • Phase Information is Valuable, But Not Essential: The absence of voltage phase information from smart meters significantly impacts the accuracy of admittance matrix estimation, leading to an approximate 30% increase in error. However, the MLE framework can still yield reasonable results.
  • MLE Outperforms Simpler Methods: The adapted MLE approach demonstrates superior performance compared to simpler linear regression methods, especially under varying noise levels.
  • Approximations are Manageable: The linearization of power flow equations and the assumptions made to handle phase-less data, while introducing some error, are shown to be acceptable for practical estimation. The error introduced by these approximations is comparable to or even better than that of linearized power flow equations themselves.
  • Sparsity Loss: When phase information is missing, the sparsity of the admittance matrix (meaning most connections are zero) is lost in the estimation, requiring potential regularization techniques (like Maximum A Posteriori estimates) to recover it.

Key Results

  • ~30% Error Increase: Estimating the admittance matrix without voltage phase information results in approximately a 30% increase in the Relative Root Mean Square Error (RRMSE) compared to methods using phase information.
  • Robustness to Noise: The sensitivity of the estimation to measurement noise is found to be similar whether phase information is available or not.
  • Power Flow Accuracy: The adapted MLE's power output approximations are shown to be close to those of linearized power flow equations, suggesting the method is reliable for estimating power flow quantities.
  • Admittance Matrix Comparison: Visualizations of the admittance matrices show that the estimation without phase information loses its distinct sparsity pattern, with many small-magnitude off-diagonal elements appearing.

Limitations and Future Work

  • Sparsity Recovery: The paper acknowledges that the loss of sparsity in the estimated admittance matrix when phase is absent is a limitation. Future work could involve using techniques like Maximum A Posteriori estimation with L1-norm penalties to recover sparsity.
  • Approximating Voltage Phase: The authors suggest that approximating the voltage phase with its average value over a longer period (if available) could improve accuracy.
  • Hybrid Measurement Systems: Extending the method to combine measurements from both smart meters and PMUs in the same grid is a promising future direction.
  • Assumption 2 Impact: While Assumption 2 (small voltage angle differences) is crucial for linearization, its practical impact on the accuracy of the estimation could be further investigated.

Summary updated

Abstract

This paper defines a Maximum Likelihood Estimator (MLE) for the admittance matrix estimation of distribution grids, utilising voltage magnitude and power measurements collected only from common, unsychronised measuring devices (Smart Meters). First, we present a model of the grid, as well as the existing MLE based on voltage and current phasor measurements. Then, this problem formulation is adjusted for phase-less measurements using common assumptions. The effect of these assumptions is compared to the initial problem in various scenarios. Finally, numerical experiments on a popular IEEE benchmark network indicate promising results. Missing data can greatly disrupt estimation methods. Not measuring the voltage phase only adds 30\% of error to the admittance matrix estimate in realistic conditions. Moreover, the sensitivity to measurement noise is similar with and without the phase.
BibTeX
@misc{laurent2022maximum,
  author = {Lisa Laurent and Jean-Sébastien Brouillon and Giancarlo Ferrari-Trecate},
  title = {{Maximum likelihood estimation of distribution grid topology and parameters from smart meter data}},
  year = {2022},
  eprint = {2210.02217},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2210.02217},
}
Figure