Probabilistic Dynamic Line Rating with Line Graph Convolutional LSTM

Min‐Soo Kim, Vladimir Dvorkin, Jip KimarXiv 2025 · DOI · arXiv

Abstract

Dynamic line rating (DLR) is an effective approach to enhancing the utilization of existing transmission line infrastructure by adapting line ratings according to real-time weather conditions. Accurate DLR forecasts are essential for grid operators to effectively schedule generation, manage transmission congestion, and lower operating costs. As renewable generation becomes increasingly variable and weather-dependent, accurate DLR forecasts are also crucial for improving renewable utilization and reducing curtailment during congested periods. Deterministic forecasts, however, often inadequately represent actual line capacities under uncertain weather conditions, leading to operational risks and costly real-time adjustments. To overcome these limitations, we propose a novel network-wide probabilistic DLR forecasting model that leverages both spatial and temporal information, significantly reducing the operational risks and inefficiencies inherent in deterministic methods. Case studies on a synthetic Texas 123-bus system demonstrate that the proposed method not only enhances grid reliability by effectively capturing true DLR values, but also substantially reduces operational costs.

Introduction

Power grid operators traditionally rely on Static Line Ratings (SLR)—fixed limits on how much current a transmission line can carry. To ensure safety, these limits are set using "worst-case" weather assumptions, such as a scorching, windless day. While safe, this conservative approach leaves a massive amount of transmission capacity unused.

Dynamic Line Rating (DLR) offers a solution by adjusting these limits in real-time based on actual weather conditions, such as wind cooling the lines. However, DLR is notoriously difficult to predict. Most existing forecasting methods are either deterministic, providing a single number that lacks uncertainty information, or local, ignoring the spatial correlations of weather patterns moving across the grid. If a forecast is too optimistic, the grid faces reliability risks; if it's too pessimistic, economic efficiency is lost, and renewable energy is curtailed unnecessarily.

Methodology: The LGCLSTM Framework

To address these challenges, we propose the Line Graph Convolutional LSTM (LGCLSTM). This novel architecture integrates spatial graph convolutions with temporal LSTMs to provide network-wide probabilistic forecasts.

The proposed LGCLSTM architecture consists of four main stages: graph representation, line graph conversion, probabilistic forecasting using LGCLSTM and quantile layers, and integration into grid operation models.

The framework is built on four key components:

  1. Line Graph Conversion: Standard Graph Convolutional Networks (GCNs) learn features of nodes (buses). However, line ratings are properties of edges (lines). We convert the power grid into a "line graph" where every transmission line becomes a node. This allows the model to capture spatial correlations between connected lines using standard GCN mathematics.
  2. Multi-hop LGCN: Instead of looking only at immediate neighbors, the model uses a "multi-hop" approach to capture broader weather patterns, such as a storm front moving across a region, without the memory overhead of stacking many deep layers.
  3. Spatio-Temporal Integration: The LGCN is embedded directly into a Long Short-Term Memory (LSTM) cell. The GCN component handles the spatial layout of the grid, while the LSTM processes the time-series nature of weather data.
  4. Quantile Regression: Instead of a single point estimate, the model outputs multiple quantiles. This provides operators with a "prediction interval," allowing them to select a safety margin based on specific risk preferences.

Key Insights and Results

The model was evaluated using the Texas 123-bus backbone transmission (TX-123BT) system with five years of historical weather data.

Efficiency and Spatial Awareness

By transforming the grid topology into a line graph, the model achieves superior performance with significantly fewer parameters—roughly 93% fewer than competing graph-based models. This efficiency is crucial for large-scale deployment. Furthermore, the LGCLSTM performs best in regions where buses are densely clustered, confirming its ability to learn how neighboring lines influence each other's thermal states.

Comparative Performance

The LGCLSTM was compared against several baselines, including Quantile Regression Forests (QRF) and standard LSTMs.

Comparison of probabilistic DLR forecasting using QRF (top) and the proposed LGCLSTM (bottom). The LGCLSTM provides sharper prediction intervals that more accurately track the true DLR values.

As shown in the comparative analysis, LGCLSTM achieved the lowest Quantile Score (QS) and the most precise prediction intervals (PINAW). In grid simulations, this translated to the lowest total operating costs ($1231.33k hourly average) and a significant reduction in renewable energy curtailment compared to static ratings.

The Safety Margin Paradox

An interesting finding was that using a conservative probabilistic forecast (e.g., the 5th percentile) can sometimes lead to lower redispatch costs than having "perfect" DLR information. This conservative buffer helps absorb unexpected errors in load forecasting, acting as a multi-purpose safety net for grid operations.

Conclusion

The LGCLSTM framework represents a significant step forward in transmission capacity forecasting. By providing accurate, network-wide probabilistic ratings, it allows grid operators to move beyond conservative static limits safely. While future work will focus on integrating these forecasts into more complex Security-Constrained Unit Commitment (SCUC) models and testing against extreme weather events, the current results demonstrate clear benefits for grid reliability, economic efficiency, and renewable energy integration.

Summary updated

Abstract

Dynamic line rating (DLR) is an effective approach to enhancing the utilization of existing transmission line infrastructure by adapting line ratings according to real-time weather conditions. Accurate DLR forecasts are essential for grid operators to effectively schedule generation, manage transmission congestion, and lower operating costs. As renewable generation becomes increasingly variable and weather-dependent, accurate DLR forecasts are also crucial for improving renewable utilization and reducing curtailment during congested periods. Deterministic forecasts, however, often inadequately represent actual line capacities under uncertain weather conditions, leading to operational risks and costly real-time adjustments. To overcome these limitations, we propose a novel network-wide probabilistic DLR forecasting model that leverages both spatial and temporal information, significantly reducing the operational risks and inefficiencies inherent in deterministic methods. Case studies on a synthetic Texas 123-bus system demonstrate that the proposed method not only enhances grid reliability by effectively capturing true DLR values, but also substantially reduces operational costs.
BibTeX
@misc{kim2025probabilistic,
  author = {Min‐Soo Kim and Vladimir Dvorkin and Jip Kim},
  title = {{Probabilistic Dynamic Line Rating with Line Graph Convolutional LSTM}},
  year = {2025},
  eprint = {2512.04369},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.04369},
  howpublished = {arXiv},
}
Figure