TL;DR
This paper introduces a novel approach to demand response (DR) for natural gas networks, adapting concepts from electrical DR to manage peak residential heating demand. By modeling smart thermostat behavior and using optimal control, the authors demonstrate significant reductions in peak gas consumption, crucial for grid resilience during extreme weather events.
What Problem Does This Paper Address?
The electrical power industry has mature demand response programs that effectively manage peak loads, leading to efficiency and resilience benefits. However, natural gas systems, increasingly operating at capacity, especially during winter, lack similar DR strategies. This is problematic because extreme weather events, like polar vortices, can stress gas infrastructure, leading to potential disruptions. While electrical DR can shift usage, natural gas DR for heating cannot easily do so. The paper addresses this gap by proposing analogues to electrical DR for residential natural gas consumption, aiming to reduce peak demand and enhance system resilience without compromising customer comfort.
Methods
The core of this paper is the development of mathematical formulations for natural gas demand response using Optimal Control Problems (OCPs). The authors model residential heating dynamics using physics-based thermodynamic equations, similar to those used for Thermostatically Controlled Loads (TCLs) in electrical systems. They propose two types of smart thermostats:
- Type 1 (Decentralized): These thermostats have limited computational power, use local temperature data and forecasts to optimize their own operation, and aim to balance temperature comfort with gas savings. The OCP for these is a Mixed-Integer Linear Program (MILP) solved individually for each house.
- Type 2 (Centralized): These thermostats send data to an aggregator, which then computes an optimal on-off control policy for all participating households. This centralized approach allows for system-wide optimization to meet specific peak reduction targets. The OCP for this model is also formulated as an MILP, but solved for the entire ensemble of houses.
To handle the computational complexity of these OCPs, especially for larger systems or longer time horizons, the authors employ a Receding Horizon (RH) algorithm. This algorithm breaks down the overall time horizon into smaller, sequential segments, solving each segment and using the end-state of one segment as the initial condition for the next.
Datasets
The case studies utilize historical ambient temperature data for Chicago:
- A typical winter day: January 2, 2020.
- An extreme weather event: The polar vortex of January 26-27, 2019.
The paper also references a GitHub repository for house-specific data, including thermal coefficients, furnace burn rates, and physical characteristics, which were used for the simulations.
Key Insights and Ideas
- Analogues for Gas DR are Necessary: The paper highlights that direct translation of electrical DR strategies (like price-based shifting) isn't always suitable for natural gas due to factors like social equity concerns during critical heating periods. This necessitates developing analogous approaches.
- Smart Thermostats as Key Enablers: The research leverages the capabilities of current and anticipated smart thermostats, differentiating between decentralized and centralized control models, to implement DR.
- Physics-Based Modeling is Crucial: Using thermodynamic models to capture house temperature dynamics is essential for accurately formulating the OCPs and understanding the trade-offs between comfort and consumption.
- Trade-offs Exist: The decentralized model explicitly formulates a trade-off between maintaining temperature comfort (minimizing deviation) and reducing gas consumption, controlled by a parameter $\lambda_h$.
- Centralized Control Offers Greater Peak Reduction: While decentralized control provides benefits, the centralized model demonstrates a greater capacity for peak shaving, achieving up to 15% reduction in gas consumption during extreme events.
- Receding Horizon Algorithm is Practical: The use of an RH algorithm is key to making these OCPs computationally tractable for real-world applications.
Key Results
- Peak Demand Reduction: The decentralized model achieved reductions in gas consumption for a typical winter day, with savings amplified by aggregation. For the "typical-day" case study, reductions of up to 0.07 kg per house were observed, translating to significant savings for aggregated households.
- Extreme Weather Performance: Even during the severe "polar-vortex" scenario, the decentralized model showed potential for gas savings with good deviation performance, though less substantial than on a typical day.
- Centralized Model Efficacy: The centralized model demonstrated the ability to achieve up to 15% peak reduction in gas consumption during extreme events, with temperature deviation comparable to or better than the baseline.
- Computational Efficiency: The decentralized model's computation time was generally low (seconds), while the centralized model's computation time varied, with the "min. max." objective function being more computationally efficient than the "min." objective function due to reduced degeneracy.
Limitations and Future Work
- Pipeline Network Ignored: The current models do not account for the natural gas pipeline network's dynamics or its impact on DR.
- Uncertainty Not Fully Addressed: While temperature forecasts are used, uncertainty in these forecasts and in the estimation of thermal coefficients is not fully incorporated.
- Incentive Mechanism Not Detailed: The paper focuses on the technical formulation and does not deeply explore the design of economic incentives to encourage customer participation, beyond stating that costs should be offset.
Future work suggested includes:
- Incorporating network effects of the pipeline.
- Modeling uncertainty in temperature forecasts and thermal coefficients.
- Quantifying the link between time-based rates/incentives and direct load control.
- Developing more sophisticated price-based DR mechanisms that might require real-time gas rates.
Summary updated