Selecting Sensors when Forecasting Temperature in Smart Buildings

Forecasts of temperature in a “smart” building, i.e. one that is outfitted with sensors, are computed from data gathered by these sensors. Model predictive controllers can use accurate temperature forecasts to save energy by optimally using Heating, Ventilation and Air Conditioners while achieving comfort. We report on experiments from such a house, in which we select different sets of sensors, build a temperature model from each set, and then compare the accuracy of these models. While a primary goal of this research area is to reduce costs by reducing energy consumption, in this paper, besides the cost of energy, we consider the cost of data collection and management. Each sensor employed in the forecast calculation incurs costs for installation and maintenance and an incremental cost for computation. Some sensors, however, may contribute little or no improvement to the forecast accuracy. We incrementally construct sets of sensors until we arrive at a set for which no superset produces a better forecast. Then we construct a successive series of subsets, such that forecast accuracy degrades slowly. As each sensor is removed, on the one hand, the forecast error increases, so the energy costs may increase for a given controller. On the other hand, the costs for installing sensors and for computing models are reduced. By considering this tradeoff over the the series of sets, an optimal set of sensors can be found to be used with that controller.