I. Introduction
Digital transformation in buildings has brought considerable opportunities to optimize their energy performance by integrating various advanced information and communication tech-nologies [1]. Among them, digital twin technology is today a powerful tool for building management. With continuously collected data [2], a digital twin reflects the latest status of its physical counterpart in nearly real-time [3]. In addition, more advanced data analysis applications, such as energy forecasting and predictive controls, can be developed based on the virtual model and data from meters, sensors, actuators, and control systems [4]. Deep learning has shown great potential in data analytics [5], [6]. Based on large amounts of the collected data, deep learning methods can be used to develop models for identifying patterns in operational data, such as making predictions about energy use and revealing the potential for energy optimization [7].