TL;DR
This paper introduces a novel nonlinear optimal control framework for managing natural gas pipeline networks that includes underground storage. By developing a physics-based, reduced-order model for storage facilities and integrating it with existing pipeline dynamics, the authors enable more efficient and profitable day-to-day operations, especially crucial given the increasing variability of natural gas demand.
What Problem Does This Paper Address?
The increasing reliance on natural gas for electricity generation, particularly to balance the intermittency of renewable energy sources, has led to more variable and less predictable gas demand. This poses a significant challenge for managing natural gas pipeline networks, which traditionally relied on steady-state models. Furthermore, the effective utilization of underground gas storage facilities, essential for meeting peak winter demand, is complicated by the complex interplay between pressure gradients and flow dynamics within the pipeline system. Existing models for storage are often too simplistic (treating them as mere sources/sinks) or too complex for real-time operational optimization. This paper addresses the need for a unified, physics-informed model that can optimize the joint operation of pipeline flows and storage facilities under transient conditions to maximize economic profit and network efficiency.
Methods
The core of this work lies in developing a nonlinear optimal control problem for intra-day operation. The authors achieve this by:
- Developing Reduced-Order Models: They create simplified, yet physically representative, differential-algebraic equation (DAE) models for key components of a natural gas pipeline network. This includes:
- Rectilinear Pipes: Extending existing models to account for non-horizontal pipes and incorporating the Darcy-Wiesbach friction approximation.
- Compressors: Modeling their function as density multipliers, with specific models for unidirectional and bidirectional flow.
- Storage Facilities: This is a key contribution. They model underground reservoirs as rigid containers with uniform density and isothermal conditions, and wells as vertical pipes. Crucially, their model captures the physics of gas storage, moving beyond simplistic "battery" models where withdrawal capacity is independent of the state of charge.
- Synthesizing a Network Model: These component models are spatially discretized and assembled into a single, reduced-order DAE system that represents the entire network, including storage.
- Formulating an Optimal Control Problem: An objective function is defined to maximize economic profit (considering gas sale/purchase prices) and network efficiency (minimizing compressor power). This is subject to the derived flow equations and operational constraints (pressure limits, flow rates, storage capacity, etc.).
- Solving the Problem: The nonlinear optimal control problem is solved using a primal-dual interior point solver.
Datasets
The paper uses synthesized test cases for evaluation, including:
- A 6-junction network without storage (used as a baseline).
- A 6-junction network with a single storage facility.
- A larger, more realistic test network synthesized from the Questar pipeline system in the US, featuring multiple storage units.
The authors note that data for the Questar network was synthesized from openly available data.
Key Insights and Ideas
- Physics-Informed Storage Models are Crucial: The paper strongly argues against simplistic "battery" models for gas storage. Their results show that storage withdrawal capacity is not constant but decreases as the reservoir pressure drops, a phenomenon accurately captured by their physics-based model. This has significant implications for realistic operational planning.
- Joint Optimization is Key: Effectively managing pipeline capacity and storage utilization requires a unified approach. Optimizing flows and storage simultaneously allows for better responsiveness to fluctuating demand and market prices.
- Reduced-Order Models Enable Scalability: By developing computationally tractable reduced-order models for complex components like storage, the authors make it feasible to apply optimal control to large-scale, real-world pipeline networks.
- Transient Dynamics Matter: The increasing variability of demand means that steady-state assumptions are no longer sufficient. The paper demonstrates the importance of modeling transient flow conditions for accurate operational optimization.
Key Results
- The developed optimal control framework successfully integrates pipeline and storage dynamics.
- Simulations on a 6-junction network showed that the storage facility could meet demand, reducing unmet off-take values to zero, and significantly decreasing the need for gas from the intake point due to its lower cost.
- The spatial discretization study indicated that a spatial discretization of 10 km for pipes provided acceptable accuracy, with mean relative errors below 10⁻⁴.
- The Questar network case study demonstrated the scalability of the approach, with the optimal control problem solved in just 3 minutes for a large network with multiple storage units. The results showed that two of the four storage units were actively injecting gas.
Limitations and Future Work
- Simplifications in Storage Model: While an improvement over battery models, the storage model still makes simplifying assumptions (e.g., uniform density, isothermal conditions). Future work could incorporate more complex reservoir physics.
- Adiabatic Compression: The paper assumes adiabatic compression for compressors. Future work could explore isothermal or polytropic compression.
- Limited Scope of "Intra-day" Operation: The optimization horizon is set to 24 hours. Extending this to longer time horizons or incorporating more detailed market dynamics could be beneficial.
- Validation with Real-World Data: While the Questar network is based on real data, the specific operational parameters and results are from a synthesized model. Further validation with actual operational data would strengthen the findings.
- Uncertainty: The current model does not explicitly handle uncertainties in demand, supply, or prices. Incorporating robust or stochastic optimization techniques would be a valuable next step.
Summary updated