Building Occupancy Estimation with Environmental Sensors via CDBLSTM

摘要: Buildings consume quite a lot of energy; hence, the issue building energy efficiency has attracted great deal attention in recent years. A key factor achieving this objective is occupancy information that directly impacts on energy-related control systems. In paper, we leverage environmental sensors are nonintrusive and cost-effective for estimation. Our result relies feature engineering learning. The conventional requires one to manually extract relevant features without clear guideline. This blind extraction labor intensive may miss some significant implicit features. To address issue, propose convolutional deep bidirectional long short-term memory (CDBLSTM) approach contains network structure automatically learn from sensory data human intervention. Moreover, networks able capture temporal dependencies can take past future contexts into consideration final identification occupancy. We have conducted real experiments evaluate performance our proposed CDBLSTM approach. Instead estimating exact number occupants, attempt identify range i.e., zero, low, medium, high, which adequate most experimental results indicate effectiveness compared with state-of-the-art methods.