494. HVAC System Management Using an Optimization with Constraint Learning Framework
Invited abstract in session WB-48: New Approaches in Explainable Optimization, stream Transparent and Fair Decision Making with Mathematical Optimization.
Wednesday, 10:30-12:00Room: Parkinson B09
Authors (first author is the speaker)
| 1. | Donato Maragno
|
| Amazon | |
| 2. | Marco Caserta
|
| Amazon | |
| 3. | Abhinav Pradhan
|
| Amazon |
Abstract
Heating, Ventilation, and Air Conditioning (HVAC) systems account for a significant portion of energy consumption in buildings, making their optimization crucial for reducing costs and environmental impact. This work presents a novel optimization framework for minimizing energy costs of HVAC systems while maintaining indoor temperature within the comfort zone. We formulate the problem as a mixed-integer optimization (MIO) model to determine optimal temperature setpoints over a daily planning horizon. To overcome the challenge of modeling non-explicit, system-dependent, and highly non-linear functions like power consumption and indoor temperature, we propose an optimization with constraint learning (OCL) approach. Machine learning models are trained on historical data to predict the outcomes of interest and are embedded as constraints and objective functions within the optimization model. The power of ML models such as decision trees, gradient boosting, and neural networks lies in their ability to capture non-linear functions while being themselves MIO-representable. This is why the OCL framework allows us to find globally optimal setpoints efficiently using out-of-shelf MIO solvers. A case study on an Amazon delivery station demonstrates the effectiveness of our approach in reducing energy consumption by 5.2% while maintaining indoor temperatures within comfort bounds.
Keywords
- Machine Learning
- OR in Energy
- Programming, Mixed-Integer
Status: accepted
Back to the list of papers