Calibration method of particulate matter sensor based on density peaks clustering combined with stacking algorithm

More and more low-cost sensors (LCS) were used to obtain monitoring data with higher temporal-spatial resolution, as air quality has become more concerned in recent decades. However, due to its working principle, relatively simple internal structure, and complex external environmental conditions, the accuracy of LCS's measurement data has always been questioned. Therefore, it is necessary to develop calibration method for LCS to improve the reliability of the data. This study proposed a calibration method for particulate matter LCS using a K-nearest Neighbor Fuzzy Density Peaks Clustering combined with Stacking Ensemble Learning (KFDPC-Stacking). Experiments were conducted using the LCS network in Zhengzhou, China to verify the effectiveness of the developed calibration model. The results show that the developed model had higher accuracy in data calibration compared to other calibration models based on Machine Learning techniques. The transferability of the model had been validated in multiple areas of Zhengzhou, and the results indicated that the calibration model could be applied directly in comparable environments. Additionally, the data of LCS in Zhengzhou City within one year after calibration were analyzed, and the suggested calibration interval for such sensors was determined, which could provide information and supports for the subsequent LCS research and calibration.