Indoor Temperature Forecast based on the Lattice Boltzmann method and Data Assimilation

Control of the indoor environment in buildings is reactive and usually based on sensor data at a single location in a room without any consideration of the air flows within the space. The ability to accurately simulate indoor conditions would enable predictive control offering improved environmental conditions and better energy efficiency. This study investigates the temporal prediction of temperature and its coupling with sensor data for improving thermal comfort forecasting in an indoor environment. We present a real-time implementation of a computational fluid dynamics (CFD) Lattice Boltzmann method (LBM) model coupled with data assimilation (DA) to make periodic updates to the state of the model. Variational and sequential DA techniques are evaluated and two novel methods for real-time accurate temperature prediction are presented. The models are demonstrated for prediction of temperature in an idealized room, with simulated temperature sensor readings used to update LBM-based flow predictions. The LBM-DA approach overcomes the need for accurate boundary conditions while considerably improving the transient and spatial prediction of temperature in terms of root mean squares error, thereby avoiding large deviation from the room’s true flow state. The accuracy of prediction is shown to depend on number of sensors with poor prediction using just a single sensor. Results also depend on quality of sensor data, with highly variable data yielding poorer results. The approach has potential for application in indoor environments to provide more accurate and faster response of control systems to changing environmental conditions.