Dr. Ziwen Yu


Date: September 26, 2022

Speaker: Dr. Ziwen Yu

Affiliation: Agricultural and Biological Engineering, UF

Title: A Staged QAQC Method for IoT Based Environment Monitoring

Abstract: Environment monitoring is susceptible to various uncertainties stemming from the complexities of the underlying eco-environmental system. Typically manifesting as biases and abnormal values in data, these uncertainties—with their unexpected occurrences and magnitudes—pose significant challenges for accurately and precisely interpreting the behaviors of interest within the eco-environmental system. With the assistance of Internet of Things (IoT) sensing, near-real-time monitoring can be achieved, facilitating the collection of more data for enhanced analysis and modeling, including the study and addressment of the impacts of these uncertainties before conceptualizing the system in a presumed simple scenario.

In this study, we have developed a multi-staged method employing data-driven algorithms to identify various suspicious data points and patterns in environmental monitoring. The distinct manifestations of all uncertainties can generally be categorized as outliers, abnormal patterns, and uncommon spatial-temporal correlations. Statistical methods have been devised to identify outliers in direct readings and abrupt changes across different time intervals. The similarity between various patterns of time series events has been quantified using the Dynamic-Time-Warping (DTW) method. The results, along with their associations with other related parameters, have been explored using a rule-based learning algorithm (Apriori) to refine special patterns.

Leveraging the established monitoring networks, the correlations among data from different observation locations offer another dimension for identifying suspicious information. While the algorithm is still in development, a profound understanding of complex weather processes forms a critical foundation for the further integration of more sophisticated data models, such as Artificial Intelligence (AI). Changes in air pressure have been employed as the key indicator of weather event shifts and magnitudes based on meteorological concepts. A preliminary Graph Neural Network (GNN) has been designed to explore the evolution of weather events in a monitoring network and assess the existing correlations.

At present, as an integrated part of the Florida Automated Weather Network (FAWN), this method has been partially utilized to perform Quality Assurance and Quality Control (QAQC) for the majority of weather monitoring data collected throughout the state of Florida.