Generating Occupancy Profiles for Building Simulations Using a Hybrid GNN and LSTM Framework

Building occupancy profiles are critical in thermal and energy simulations. However, determining an accurate profile is difficult due to its stochastic nature. In most simulations, the occupant activities usually represented by fixed yearly schedules, which often derived from guides other similar sources may not represent simulated building accurately. Therefore, inaccuracy defining can be a source of error Over past few years machine learning has become very popular ability reveal hidden patterns relationships between data this makes it suitable for investigating data. This study proposes novel hybrid model combining Graph Neural Network Long Short-term Memory neural network (LSTM) predict individual rooms on typical office floor. The proposed LSTM produce high-resolution that good agreement with reference same office. were agent-based using AnyLogic used training network. outperformed networks tested such as Recurrent (RNN), Gated Unit (GRU) LSTM. When compared tested, there range improvement 13.5 14.6% index agreement, 38.3 46.8% mean absolute 34.4 40.0% root square error, when averaging differences over whole