Experimental study on the application of mobile sensing in wireless sensor networks development: Node placement planning and on-site calibration

Abstract

Wireless sensor networks (WSN) are a crucial means of obtaining real-time indoor air quality (IAQ) information in large public buildings. Improper placement of WSN nodes may lead to a significant waste of sensing resources and the overlooking of critical IAQ information. In this study, mobile sensing was used to predict indoor CO₂ concentration in order to guide the placement of WSN nodes by a spatiotemporal processing method. Experimental studies were conducted in a well-controlled chamber with the consideration of different ventilation systems, indoor temperatures, and CO₂ source positions and numbers. Mobile sensing was also used for on-site calibration of low-cost NDIR CO₂ sensors under 20 experimental conditions, and a genetic algorithm-optimized back propagation (GA-PB) neural network was used for calibration fitting. The results indicated that the raw voltage signals of low-cost sensors can be effectively denoised with the use of four-level wavelet decomposition. Mobile sensing performed well in predicting CO₂ concentration; its poorest prediction performance was found at the measurement points closest to the source, with a determination coefficient (R²) of 0.682 and a root mean square error (RMSE) of 6.5 ppm, which met the accuracy requirements. The K-means algorithm yielded the best clustering results, with average CH index values of 11.81, 13.90, 13.89 and 14.05 for cluster numbers of 3, 4, 5 and 6, respectively. For on-site calibration of low-cost CO₂ sensors, the GA-PB neural network achieved an R² above 0.97, an RMSE ranging from 19.2 to 27.6 ppm, and a MAPE in the range of 2.05–2.69 %.