An Intelligent Building Occupancy Detection System Based on Sparse Auto-Encoder

Abstract

Energy saving in buildings has attracted many researchers attention. One of the research topics is occupancy detection for each room to automatically control airconditioner, light, heating, etc. by monitoring the temperature, humidity, light, and co <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> using the corresponding sensors. For existing data analysis, traditional regression methods such as CART, RF, LDA are often used to predict the occupancy status after the data are collected. However, traditional methods have disadvantages of difficult adaptation to large data sets and bad generalization. In this paper, deep learning technology is used to perform occupancy detection. We construct sparse auto-encoder to learn features from data, then the learned features are input to Softmax, SVM and Liblinear classifiers to detect the room occupancy status. Experiments and results show the efficiency of the proposed algorithm whose detection accuracy outperforms that of benchmark project. Moreover, proposed algorithm has very low computation overhead which facilitates the real-time detection.