Machine Learning based Floor-level Sensor Positioning in Smart Buildings

With the emergence of new technologies, such as 5G, WiFi6, and NB-IoT, sensors are easier to be deployed in smart buildings than before, and more buildings have being built or transformed into smart buildings. Thus, how to track and locate these devices has become a problem. The traditional sensors positioning is achieved by manually recording the position of the sensor or using RSSI measurement. However, these traditional methods suffer from high costs, inaccurate positioning accuracy and high complexity. Different from the traditional sensor positioning algorithm, our design locates the position of sensors by learning the pattern of atmospheric pressure data collected by the sensor. Firstly, we designed the IoT data collection networks which can collect indoor environment data through WiFi, such as light, atmospheric pressure, temperature, humidity, etc. Secondly, the original data is preprocessed by removing abnormal points, adding missing points and filtering noise. Finally, using the processed data, a restricted Boltzmann machine is applied to obtain the pattern between the atmospheric pressure and floors. Experimental results indicated that our sensor positioning algorithm provides high-precision floor positioning.