TL;DR
This paper introduces a novel "Switching Rate" actuation mechanism for controlling populations of thermostatically controlled loads (TCLs), like refrigerators, for demand response. They develop a stochastic hybrid system model for individual TCLs and derive a partial differential equation (PDE) model for the aggregate population behavior, which is then validated through simulations. This work offers a more flexible and potentially more stable approach to demand response compared to synchronized switching methods.
What Problem Does This Paper Address?
In modern power grids, especially with the rise of intermittent renewable energy sources, managing electricity demand is crucial. Thermostatically controlled loads (TCLs) – devices that maintain a temperature setpoint, such as refrigerators, air conditioners, and heat pumps – represent a significant portion of electricity consumption. These TCLs can act as distributed energy storage, allowing grid operators to subtly adjust their operation to balance supply and demand. However, individual TCLs have very small storage capacities, meaning that effective demand response requires coordinating large populations of these devices.
Previous approaches often involved synchronized switching of TCLs, which can lead to undesirable power peaks that destabilize the grid. The challenge is to develop control strategies that can influence TCL populations for demand response without causing these disruptive synchronized peaks, while still maintaining the devices within their operational temperature bounds and without breaking other constraints.
Methods
The authors model an individual TCL as a Stochastic Hybrid System (SHS). This means it has both continuous states (temperature) governed by Stochastic Differential Equations (SDEs) and discrete states (on/off mode) that change based on deterministic rules (thermostat) and probabilistic events.
The core innovation is the "Switching Rate" actuation mechanism. Unlike previous methods that might synchronize switches, this mechanism uses an external signal to control the rate at which TCLs probabilistically switch between on and off states. This rate is parameterized by an external signal, allowing for flexible control. To avoid issues like equipment damage from rapid switching or inefficient operation, two heuristics are introduced:
- Safe Distance: Rate-switching is only allowed when the temperature is a certain distance away from the thermostat boundaries.
- Minimum Dwell Time: A minimum time is enforced in each mode (on/off) to prevent very short cycles.
This SHS model for individual TCLs is then used to derive a population-level model in the form of a Partial Differential Equation (PDE), specifically a generalized Fokker-Planck equation. This PDE describes the evolution of the probability density function (pdf) of the TCL population's states (temperature and mode). The paper presents the PDE for the unactuated TCL and then extends it to include the Switching Rate actuation.
For validation, they employ two simulation methods:
- Monte Carlo Simulation: Simulating a large number (10,000) of individual SHS TCL models.
- Finite Volume Method (FVM): A numerical technique to approximate the solution of the derived PDE.
Datasets
Not discussed in the paper. The paper focuses on modeling and simulation rather than training on specific datasets.
Key Insights and Ideas
- Decoupled Switching is Key: The "Switching Rate" actuation, by allowing asynchronous probabilistic switches, avoids the synchronized power peaks associated with earlier "Switching Fraction" methods. This is crucial for grid stability.
- PDE as a Population Model: The derivation of a Fokker-Planck-type PDE from the SHS model provides a powerful analytical tool for understanding and predicting the aggregate behavior of large TCL populations.
- Heuristics for Practicality: The inclusion of "safe distance" and "minimum dwell time" heuristics addresses real-world constraints, making the model more robust and practical for deployment.
- Model Equivalence: The paper demonstrates a strong agreement between the Monte Carlo simulations of individual TCLs and the FVM solution of the population PDE, validating the modeling approach.
Key Results
The paper successfully derives a PDE model for TCL populations under the proposed Switching Rate actuation. Numerical simulations (Monte Carlo and FVM) show that the model accurately captures the behavior of the TCL population. Figure 1 shows that the power consumption predicted by the PDE model closely matches the output of the Monte Carlo simulation for two different input signals. Figure 2 illustrates that the temperature distributions predicted by the PDE model also align well with the empirical distributions from the Monte Carlo simulation.
Limitations and Future Work
Limitations: * The paper acknowledges that the formal description of the Generalized Stochastic Hybrid System (GSHS) definition does not explicitly include dependencies on external control elements or time, and this is deferred for future work. * The current PDE model for Switching Rate actuation does not include the minimum dwell time feature, which is a simplification.
Future Work: * Rigorous GSHS Formalism: Consolidate the modeling results with a more rigorous mathematical framework, completing the GSHS description. * Incorporate Minimum Dwell Time: Develop and implement a two-dimensional FVM scheme to accurately model the minimum dwell time feature in the Switching Rate actuation. * Control Analysis: Analyze the derived bilinear model (Equation 15) for control design purposes, aiming to optimize demand response strategies.
Summary updated