Visiting Scholar from Tongji University
Tong Xiao is a PhD candidate at the School of Mechanical Engineering, Tongji University. Her research focuses on energy forecasting, optimal operation and control for HVAC systems in commercial buildings using data science technologies, particularly causal science and causal machine learning. She is also interested in automating energy management and optimization tasks with large language models, primarily focusing on automating energy efficiency diagnosis using multi-source building data. Her experience includes building energy simulation, flexible demand-side control of energy systems considering occupant behavior, and informative and automatic design of air-conditioning systems.
Her PhD thesis focuses on enhancing the predictive accuracy of data-driven energy prediction models across various buildings and usage scenarios, thereby increasing the value of using real-world data in building energy modeling. Existing data-driven approaches often struggle with generalizability, particularly when making out-of-distribution predictions, such as intervention predictions and counterfactual inference. These types of predictions are essential for tasks like building benchmarking, optimal control, demand response, and retrofit decisions. This research addresses this challenge by exploring methods to improve generalizability, focusing on energy models built from measured data and considering three key aspects: variables, models, and the data itself.