Optimization of sensor nodes deployment for drive-by sensing and k-hop calibration

Abstract

Air pollution presents significant health related risks, requiring detailed and accurate monitoring of air quality with high spatio-temporal resolution. Drive-by sensing is a popular air quality monitoring technique due to its high spatio-temporal coverage, lower deployment costs, and high-resolution data compared to traditional approaches. In this thesis we address the optimization of mobile or bus-based low-cost sensors deployment to maximize air quality monitoring coverage and reference-grade stations placement to make sure that the low-cost sensors mobile sensors deployed on selected routes are continuously calibrated for reliable air quality measurements. In this thesis, we have proposed an optimized approach to select the bus routes for deployment of bus-based sensors on for maximizing the total unique sensing area and deployment of reference-grade sensors on geolocations, for maximizing the probability of k-hop calibration. A metric-based system developed in this model, using geospatial operations which includes both spatial and temporal joins to quantify the contribution of each bus route and rank them accordingly. We have formulated the optimization problem by considering it as a mixed integer linear program (MILP) for the selection of the best possible subset of bus routes required for bus-based sensors deployment while maintaining the level of sensing and temporal coverage contribution similar to the total unique area. We have adopted greedy heuristic approach to solve this problem and use real-world bus-transit data from two cities, Toronto, Canada, and Manchester, UK, for ease of implementation of our approach. Using bus transit datasets in two different cities gives us an opportunity to analyze our proposed model's reliability and adaptability in two different environments. For performance comparison, the greedy approach outperformed the random approach, where the results showed that to achieve 50% area coverage for Toronto bus transit dataset using greedy approach proposed 39 bus routes at 100m buffer distance while for the random algorithm 64 bus routes at buffer distances of 100m. Moreover, we have developed alternative methods to analyze the performance of our proposed weighted approach where the alternate method where the alternative non-weighted approach use metrics (area and trip frequency) separately. We use this comparative analysis of our model in different environment, algorithms and buffer distances, for highlighting the relevance of the proposed metric-based system. The results of the alternative non-weight approach from the Toronto and Manchester bus-transit dataset indicate that our weighted method performed better in certain aspects in terms of spatio-temporal coverage when compared to the alternative approaches with raw or absolute data of area coverage and trip frequency.

Date of Award	3 Oct 2024
Original language	English
Awarding Institution	University of Bedfordshire
Supervisor	Vladimir Dyo (Supervisor) & Vitaly Schetinin (Second supervisor)