Dynamic State and Parameter Estimation for Natural Gas Networks using Real Pipeline System Data

Kaarthik Sundar, Anatoly ZlotnikarXiv 2019 · arXiv

Introduction

Modern natural gas networks are facing unprecedented operational challenges. Traditionally, these systems operated under relatively stable, "steady-state" conditions. However, the rapid expansion of intermittent renewable energy sources like wind and solar has shifted the role of gas-fired power plants. These plants now frequently cycle to balance the grid, causing large-scale, rapid fluctuations in gas demand. Existing steady-state or linearized models struggle to capture these transient phenomena. This paper introduces a scalable method for the joint estimation of gas pressures, flows, and pipe friction factors using real-world SCADA data, enabling operators to track network dynamics in real-time.

Methodology

The researchers propose a framework that treats state and parameter estimation as an optimal control problem constrained by the underlying physics of gas flow.

  1. Physics-Based Modeling: The system starts with 1D Euler equations (partial differential equations) describing gas flow. These are simplified using a "lumped element" approximation into a system of Differential Algebraic Equations (DAEs).
  2. Nonlinear Programming (NLP): The DAE system is discretized in time, transforming the continuous estimation problem into a large-scale nonlinear least-squares problem.
  3. Joint Estimation: The approach simultaneously estimates both the system state (pressure and flow) and critical physical parameters, specifically the friction factors ($\lambda$) of the pipes.
  4. Computational Efficiency: The resulting NLP is solved using the IPOPT interior-point solver, utilizing automatic differentiation via Julia/JuMP to handle complex network gradients efficiently.

Dataset and Network Configuration

The study validates this approach using a capacity planning model of a real-world pipeline system. The network consists of 78 nodes and 95 pipes spanning approximately 444 miles, supported by 4 compressor stations.

Layout of the real-world natural gas network used for validation, showing meter nodes, pipes, and compressor stations.

The researchers utilized one month of hourly SCADA measurements (pressure, temperature, and flow). A significant challenge addressed is sensor sparsity: data was only available for 31 of the 78 nodes (roughly 40% coverage), reflecting the practical limitations of industrial monitoring.

Key Results and Discussion

The proposed method demonstrated high scalability, successfully processing the medium-scale real-world network on standard consumer hardware. A critical finding was the discrepancy between estimated friction factors and the "ground truth" values used in planning models.

Comparison between friction factor (λ) estimates and the original planning model values across different pipe lengths.

As shown in the figure above, the estimated friction factors often deviate significantly from the static model values. This suggests that real-time parameter estimation is essential for maintaining model accuracy as infrastructure ages or internal conditions change. Furthermore, the study successfully converted raw SCADA data into transient gas density and withdrawal profiles, providing a proof-of-concept for high-fidelity state estimation in sparsely instrumented networks.

Conclusion and Future Work

While the method proves effective, sensor sparsity remains an open challenge for full model identification. Future research will focus on:

  • Robustness: Incorporating algorithms to handle measurement outliers and specifically targeting leak detection.
  • Uncertainty Quantification: Moving toward Bayesian filtering techniques (such as Extended Kalman Filters) to better manage uncertainty and provide continuous real-time updates.
  • Operational Safety: Leveraging these estimates to ensure the stability of gas networks as they become increasingly coupled with volatile power grids.

Summary updated

Abstract

We present a method for joint state and parameter estimation for natural gas networks where gas pressures and flows through a network of pipes depend on time-varying injections, withdrawals, and compression controls. The estimation is posed as an optimal control problem constrained by coupled partial differential equations on each pipe that describe space- and time-dependent density and mass flux. These are discretized and combined with nodal conditions to give dynamic constraints for posing the estimation as nonlinear least squares problems. We develop a rapid, scalable computational method for performing the estimation in the presence of measurement and process noise. Finally, we evaluate its effectiveness using a data set from a capacity planning model for an actual pipeline system and a month of time-series data from its supervisory control and data acquisition (SCADA) system.
BibTeX
@misc{sundar2019dynamic,
  author = {Kaarthik Sundar and Anatoly Zlotnik},
  title = {{Dynamic State and Parameter Estimation for Natural Gas Networks using Real Pipeline System Data}},
  year = {2019},
  eprint = {1912.05644},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1912.05644},
}
Figure