Project No: 16211924

Title: Dynamic k factor prediction from correlogram of PM optical sensor measurement for PM mass prediction using machine learning algorithms

Principal Investigator: Prof. Zhi NING

Abstract:

Particulate matter (PM), specifically PM2.5, is a key indicator of ambient air quality due to its significant health and environmental impacts. Accurate measurement of PM concentration is vital, given its diverse sources and varied physicochemical properties. A central challenge in PM monitoring is the dynamic nature of the 'correction k factor', which is essential for converting raw PM number concentration measurements from sensors into mass concentration. The k factor encapsulates the particle's refractive index and density, and any variation in this factor can lead to inaccuracies in PM mass concentration estimations. Recent advancements in sensor technology, especially multichannel optical PM sensors, offer a promising avenue for fine-tuned PM measurement. Our preliminary research has revealed a strong correlation between the correlogram of PM optical sensor measurements and particle chemical composition. This correlation suggests the potential of using the correlogram as an indicator for determining the k factor dynamically, enhancing the accuracy of PM mass concentration predictions. To address the challenge of dynamically adjusting the k factor in real-world scenarios characterized by mixed aerosols, this proposal introduces a novel approach. By harnessing machine learning algorithms, we aim to process correlogram images from the AQP sensor to output empirical k factor values. The training of this algorithm will utilize laboratory phase tests that establish benchmark k values for different particle types, while field operations will tackle mixed aerosol scenarios. The overarching goal is to match mixed correlogram images with benchmark k values, optimizing PM2.5 mass concentration predictions. This integrated sensor-machine learning approach seeks to reconcile the AQP optical responses with dynamic PM2.5 size and composition changes, facilitating more precise, real-time PM2.5 mass concentration forecasting. Through this innovative methodology, we aim to revolutionize PM monitoring, offering a more reliable and accurate tool for air quality assessments.