Deployment Support for Sensor Networks in Indoor Climate Monitoring

Automatic recognition of personal comfort is critical in realizing autonomous control of building utilities. We can infer human comfort level based on indoor environmental conditions, such as temperature and humidity, collected through sensor networks. However, the majority of methods for optimally deploying sensor networks in indoor climate monitoring mainly focused on achieving accurate measurements such as temperature distribution map with the minimum cost. Indeed, for automatic recognition of comfort using machine learning, we need to collect datasets preserving as much of the discriminatory information for inferring personal comfort with the minimum cost. In this paper, we present a novel method of placing and minimizing sensor nodes for sensor networks in smart energy systems. We have developed ZigBee-based sensor nodes and collected temperature, humidity, and illumination dataset from 13 nodes for a week. Using the dataset, we group the sensor nodes into coherent clusters, and then select a representative node which has the maximum value of RSSI for each cluster and remove the other redundant sensors, reducing the number of sensor nodes deployed. To show the feasibility of the proposed method, we perform a classification analysis of building environment. The recognition accuracy decreased by 13 percent with 6 selected sensor nodes, compared to the result with all 13 sensor nodes.