Water quality online modeling using multi-objective and multi-agent Bayesian Optimization with region partitioning

Abstract

Monitoring water resources and their quality is an activity that is gaining more importance through the years. Efficient and intelligent monitoring systems must be developed by taking advantage of cutting edge technologies like robotic agents. The utilization of autonomous surface vehicles equipped with water quality sensors is a promising approach to continuously measure physico-chemical parameters related to water quality. However, most of current related works do not acknowledge for the increasing availability and affordability of these systems. Therefore, the current approaches do not generalize well to account for multiple objectives and the involvement of multiple agents. The present work provides one of the first approaches considering the usage of multiple agents equipped with multiple water quality sensors so that online modeling of water bodies is done. Furthermore, the measurements are done considering a Voronoi Region Partitioning system using an underlying Bayesian Optimization with multiple objectives. Results show that the system can robustly obtain very accurate surrogate models despite the limited available information and energy autonomy constraint of the vehicles. When compared with coverage and patrolling-based approaches, the proposed system outperforms these approaches on average by 23.6% and 43.5%, respectively, regarding the error of modeling. The performance of this approach is also enhanced by its robustness and scalability when compared to offline monitoring missions.