Efficient Data-Driven MPC for Demand Response of Commercial Buildings

Overview

• Model predictive control (MPC) can significantly improve the energy efficiency of buildings while maintaining thermal comfort
• Data-driven approaches using neural networks have been proposed to model building systems, but these are often nonconvex and computationally complex
• This work proposes a readily implementable, data-driven, and mixed-integer convex MPC approach for small commercial buildings
• The approach uses an input convex recurrent neural network to model thermal dynamics and is solved via derivative-free optimization within a 5-minute time limit
• The method is evaluated in several demand response (DR) settings, including demand bidding, time-of-use, and critical peak rebate programs

Plain English Explanation

Improving the energy efficiency of buildings is an important goal, as buildings account for a significant portion of global energy use. Model predictive control (MPC) is a technique that has shown promise in this area, as it can optimize the building's energy use while still maintaining a comfortable indoor temperature.

However, traditional data-driven approaches to modeling the building's thermal dynamics using neural networks can be complex and computationally intensive, leading to optimization problems that are difficult to solve in a practical timeframe. This makes them challenging to implement in real-world building management systems.

In this work, the researchers developed a new method that is more readily implementable. They used an input convex recurrent neural network to model the building's thermal behavior, which allowed them to formulate a mixed-integer convex MPC problem. This type of optimization problem can be solved more efficiently, within a 5-minute time limit, using derivative-free optimization techniques.

The researchers then applied this approach to several demand response (DR) scenarios, where buildings adjust their energy use in response to signals from the electric grid. This includes things like demand bidding, time-of-use pricing, and critical peak rebate programs. By participating in these DR programs, the buildings can reduce their energy costs while still maintaining thermal comfort for the occupants.

Overall, this work provides a practical and effective way to improve the energy efficiency of small commercial buildings using MPC, which could have significant implications for reducing the environmental impact of the built environment.

Technical Explanation

The researchers designed a data-driven and mixed-integer convex MPC approach for small commercial buildings. They used an input convex recurrent neural network to model the thermal dynamics of the building, which allowed them to formulate a convex optimization problem that could be solved efficiently within a 5-minute time limit using derivative-free optimization.

The key aspects of the technical approach are:

1. Building Thermal Modeling: The researchers used an input convex recurrent neural network to model the building's thermal behavior. This type of neural network architecture ensures that the optimization problem remains convex, making it computationally tractable.

2. Mixed-Integer Convex MPC: The researchers formulated a mixed-integer convex MPC problem, which allowed them to accurately model the discrete control inputs (e.g., on/off states) of the building's heating, ventilation, and air conditioning (HVAC) system.

3. Derivative-Free Optimization: Given the limited computational time of 5 minutes, the researchers solved the MPC problem using a derivative-free optimization algorithm. This approach does not require the calculation of gradients, which can be computationally expensive for complex models.

4. Demand Response Scenarios: The researchers evaluated their approach in several demand response settings, including demand bidding, time-of-use pricing, and critical peak rebate programs. This allowed them to assess the effectiveness of their method in reducing energy consumption and costs while maintaining thermal comfort.

The researchers compared the performance of their approach to other data-driven methods and a set-point controller. Their results showed that the proposed method was able to improve thermal comfort while reducing energy consumption and costs through participation in demand response programs.

Critical Analysis

The researchers acknowledge several limitations and areas for further research in their work:

1. Scalability: The proposed approach was evaluated on a small commercial building, and the researchers note that the computational complexity may increase for larger buildings or more complex HVAC systems.

2. Model Accuracy: While the input convex recurrent neural network provided a good balance between model complexity and optimization tractability, the researchers suggest that more advanced neural network architectures could potentially improve the accuracy of the thermal modeling.

3. Uncertainty Handling: The current approach does not explicitly consider uncertainties in factors like weather, occupancy, or energy prices. Incorporating robust or stochastic optimization techniques could help the controller better handle these uncertainties.

4. Hardware Implementation: The researchers focused on the algorithmic aspects of the MPC controller, but the practical implementation of the system on building automation hardware would require additional consideration.

5. Generalization: The evaluation was limited to a specific building simulation model, and further research is needed to assess the performance of the approach across a wider range of building types and climates.

Despite these limitations, the proposed method represents a promising step towards more practical and effective data-driven approaches for building energy management. By striking a balance between model complexity and optimization tractability, the researchers have developed a readily implementable solution that could have significant real-world impact.

Conclusion

This work presents a data-driven and mixed-integer convex MPC approach for improving the energy efficiency of small commercial buildings while maintaining thermal comfort. By using an input convex recurrent neural network to model the building's thermal dynamics and solving the optimization problem via derivative-free methods, the researchers have developed a readily implementable solution that can be applied in various demand response settings.

The results demonstrate the potential of this approach to reduce energy consumption and costs through participation in demand response programs, without compromising occupant comfort. While there are some limitations that warrant further research, this work represents an important step towards more practical and effective building energy management strategies, which could have significant implications for the environmental impact of the built environment.