Prof. Zhe Wang and Wanfu Zheng Win Second Place in CityLearn Challenge 2023
Prof. Zhe Wang and Ph.D. student Wanfu Zheng from the Civil Department have won second place in the NeurIPS CityLearn Challenge 2023, a global competition that leverages the potential of AI and distributed control systems to address important problems in the built-environment of our societies.
The NeurIPS CityLearn Challenge has been held since 2020. In each challenge, the contestants are asked to develop advanced controllers to enhance the energy-efficiency, grid-interactivity, and resilience of urban energy systems. This year, the CityLearn Challenge was co-hosted by NeurIPS 2023, AIcrowd and the University of Texas at Austin, and was sponsored by Amazon, Climate Change AI and Mitsubishi Electric. The 2023 Challenge attracted 647 participants (forming 105 teams), 2568 submissions, and 43.2k views worldwide.
CityLearn Challenge 2023 required participants to create smart control agents using either reinforcement learning or model predictive control to manage electrical and domestic hot water storage systems as well as heat pumps in buildings. The primary objectives were to improve thermal comfort, to enhance energy efficiency, to reduce carbon emissions, and to ensure that the systems would remain resilient during power outages. To achieve the goal, the HKUST Smart Building team applied linear regression, k-nearest neighbors (k-NN) and LightGBM to develop the forecaster, and employed data-driven model predictive control as well as model-based model predictive control to develop the controller (details of their solution can be found in the figure below). Their solution ranked first in five sub-categories (emissions, ramp average, average load, peak load, and average peak load), and second in one sub-category (discomfort).
Solution developed by the HKUST Smart Building team
Congratulations to Prof. Zhe Wang and Wanfu Zheng.
