1Regularized Deconvolution-based Approaches for Estimating Room Occupancies

Abstract—We address the problem of estimating the number of people in a room by means of commonly available information in standard HVAC systems. We propose to use classical system identification concepts: first identify linear dynamics relating oc-cupancy levels, CO2 concentration levels and room temperature signals, and then formulate the occupancy estimation problem as a deconvolution one. The estimated occupancy is thus the input that best trades off explaining the currently measured CO2 levels and respecting a regularity condition. This condition corresponds to fused lasso priors (i.e., since rooms occupancy levels are assumed piecewise constant, we thus promote piecewise constant solutions by adding an `1 norm-dependent term to the associated cost function). We propose various estimation algorithms, each one tailored for different levels of available information (namely, the availability of measurements of actuation levels of the ventin