Robust co-design framework for buildings operated by predictive control

Paola Falugi, Edward O’Dwyer, Marta Zagórowska, Eric C. Kerrigan, Ying Nie, Goran Štrbac, Nilay ShahEnergy 2024

TL;DR

This paper introduces a robust co-design framework for residential buildings that simultaneously optimizes design, control, and operation. By decomposing the problem and intelligently selecting operating conditions, the framework significantly reduces computational complexity, achieving up to 30% cost reduction and a 10x speedup with minimal performance loss.

What Problem Does This Paper Address?

Achieving net-zero carbon emissions is a critical global challenge, and buildings are a major contributor to energy-related CO2 emissions. To decarbonize effectively, we need to optimize building design, control systems, and operational strategies in an integrated manner. However, considering all these aspects simultaneously, especially under uncertain conditions like varying weather and electricity prices, leads to computationally intractable problems. Existing approaches often neglect the interplay between design and control or simplify the problem by making assumptions that can lead to suboptimal designs. This paper aims to bridge this gap by proposing a framework that handles these complexities in a computationally efficient way.

Methods

The core of the proposed solution lies in a decomposition-based co-design framework. Instead of solving one massive optimization problem, the authors break it down into smaller, manageable sub-problems. Key techniques include:

  • Importance Subsampling: This method intelligently selects a representative subset of operating conditions (e.g., weather and electricity price scenarios) that are most critical for the design decision, rather than using all available data. This significantly reduces the computational burden.
  • Decomposition into MPC Tuning and Sizing: The problem is further decoupled into two main stages:
  • MPC Tuning: Optimizing the parameters of the Model Predictive Control (MPC) controller to achieve good closed-loop performance and computational efficiency. This involves a trade-off between performance and computational resources.
  • System Sizing: Determining the optimal physical design (e.g., PV panel area, battery capacity) based on the tuned MPC controller and economic objectives.
  • Robust Optimization: The framework explicitly accounts for uncertainties in weather and electricity prices, aiming for a design that performs well across a range of conditions, not just the average.
  • Derivative-Free Optimization: Given the potential for discontinuities in the problem formulation (e.g., due to discrete design choices or piecewise constant prices), derivative-free optimization algorithms like Mesh Adaptive Direct Search (MADS) are employed.

Datasets

The study utilizes real-world data for its case study:

  • Weather Data: Obtained from the Centre for Environmental Analysis (CEDA) archive.
  • Electricity Price Data: Assumes the Octopus Agile tariff pricing mechanism, using Market Index Price data.
  • Carbon Intensity Data: From the ESO National Grid's Carbon Intensity API.

These datasets cover a temporal resolution from years down to minutes, reflecting the dynamic nature of the problem.

Key Insights and Ideas

  • Interdependence of Design and Control: The paper strongly emphasizes that optimal building design is intrinsically linked to its control strategy. Ignoring this co-dependence leads to suboptimal outcomes.
  • Importance of Operating Conditions: The choice of significant operating conditions (subsamples) is crucial. A naive selection can lead to solutions far from optimal, while intelligent selection (importance subsampling) drastically reduces computation without sacrificing accuracy.
  • Trade-offs are Inevitable: The co-design problem involves balancing multiple objectives: performance, cost, and computational complexity. The framework provides a structured way to navigate these trade-offs.
  • Robustness Pays Off: Considering uncertainties in weather and electricity prices leads to designs that are not only more cost-effective (up to 30% reduction) but also more reliable and flexible.
  • Computational Efficiency is Key: By decomposing the problem and using smart data selection, the framework achieves significant computational speedups (10x faster) while maintaining high performance.

Key Results

  • Cost Reduction: The robust co-design framework achieved up to a 30% reduction in costs compared to deterministic approaches.
  • Computational Speedup: The proposed approach resulted in a computational time reduction of at least 10 times compared to the original, un-decomposed problem.
  • Minimal Performance Loss: This significant speedup was achieved with a deterioration in performance of only 0.6%.
  • Technology Sizing: The case study demonstrated how the framework determines optimal sizing for PV panels and batteries, showing variability based on year and operating conditions.
  • Clustering Insights: Analysis of clustering revealed that the number of clusters significantly impacts the robustness of the solution, with more clusters generally reducing uncertainty.

Limitations and Future Work

  • Sensitivity to Initial Conditions: The paper notes that the optimal solution for short-period closed-loop simulations can be sensitive to the initial state (e.g., battery charge, temperature). While they use worst-case initial conditions to enhance robustness, this remains an area for further investigation.
  • MPC Tuning Refinement: The importance subsampling approach and MPC tuning algorithm require further analysis and improvements in accuracy and computational performance.
  • Generalizability: While the framework is presented as generalizable to other contexts (power systems, robotics, etc.), further validation in these domains would be beneficial.
  • Ancillary Services: The paper suggests that incorporating revenue from ancillary services could make the combination of technologies more convenient, especially in less dynamic electricity markets. This could be a direction for future work.

Summary updated

Abstract

• Successful decarbonization of buildings needs integrated design, control and operation • Decomposition-based co-design for predictive control reduces computational complexity • Robust control co-design cuts costs by up to 30% compared to deterministic models • Proposed framework achieves 10 × faster computation with only 0.6% performance loss Cost-effective decarbonisation of the built environment is a stepping stone to achieving net-zero carbon emissions since buildings are globally responsible for more than a quarter of global energy-related CO 2 emissions. Improving energy utilisation and decreasing costs requires considering multiple domain-specific performance criteria. The resulting problem is often computationally infeasible. The paper proposes an approach based on decomposition and selection of significant operating conditions to achieve a formulation with reduced computational complexity. We present a robust framework to optimise the physical design, the controller, and the operation of residential buildings in an integrated fashion, considering external weather conditions and time-varying electricity prices. The framework explicitly includes operational constraints and increases the utilisation of the energy generated by intermittent resources. A case study illustrates the potential of co-design in enhancing the reliability, flexibility and self-sufficiency of a system operating under different conditions. Specifically, numerical results demonstrate reductions in costs up to 30% compared to a deterministic formulation. Furthermore, the proposed approach achieves a computational time reduction of at least 10 times lower compared to the original problem with a deterioration in the performance of only 0.6%.
BibTeX
@misc{falugi2024robust,
  author = {Paola Falugi and Edward O’Dwyer and Marta Zagórowska and Eric C. Kerrigan and Ying Nie and Goran Štrbac and Nilay Shah},
  title = {{Robust co-design framework for buildings operated by predictive control}},
  year = {2024},
  eprint = {2410.10446},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2410.10446},
}
Figure