A prescriptive tree-based model for energy-efficient room scheduling: Considering uncertainty in energy generation and consumption

This paper investigates the energy-efficient room scheduling (ERS) problem by considering uncertainties in energy consumption and renewable energy generation in buildings. Rather than the conventional ‘predict, then optimise’ approach, we propose an improved prescriptive tree-based (IPTB) model that directly ‘prescribes’ scheduling solutions. Our model utilises contextual information on energy consumption (e.g., temperature and humidity) and renewable energies (e.g., wind speeds and sunlight) to generate direct ERS solutions. It is trained using a novel optimisation loss function that aligns historical ERS solutions with current conditions, ensuring robustness and tractability by exploiting problem-specific properties. To evaluate the proposed model’s performance, experiments on randomly generated ERS instances demonstrate that the IPTB model is trained efficiently across various problem sizes and consistently outperforms advanced data-driven optimisation methods in prescriptive accuracy. Moreover, the IPTB model achieves more balanced energy consumption, particularly under practical scenarios emphasising on energy demand charges. A case study using real-world datasets from six buildings at Monash University, Australia, validates the model’s effectiveness in addressing complex practical constraints inherent in ERS problems. • We study energy-efficient room scheduling by considering related uncertainties. • We propose an improved prescriptive tree-based model for this purpose. • The model utilises contextual information on energy consumption and renewable energy. • It outperforms other advanced data-driven optimisation methods. • It also excels in balancing peak-time energy consumption in practical settings.