Optimalizace odpadového hospodářství s využitím IoT a Smart Solutions

This master's thesis explores the potential for optimizing municipal waste collection in the city of Prague using data from IoT sensors monitoring the fill levels of waste containers. The aim of the thesis is to evaluate, based on real operational data, the potential of dynamic waste collection management in comparison with the existing static model based on fixed pickup schedules. Within the scope of the thesis, a data pipeline was first designed and operated to continuously collect data from 7 705 monitored containers from the Golemio platform between November 2025 and April 2026 (5 months). As the dataset did not contain a reliable ground truth for actual pickups, a heuristic methodology was developed that combines system logs with a physical interpretation of the sensor signal; this approach made it possible to select 59 109 sufficiently credible pickups from 97 565 detected changes. An association analysis using the CleverMiner tool, based on the GUHA method, was then performed on the resulting dataset and revealed four principal patterns: structural overloading of paper containers at the beginning of the week, high inefficiency of glass pickups during weekends, a strong influence of spatial context on container fill rate and an empirical difference between the municipal district of Prague 11, where dynamic collection is already being piloted, and the rest of Prague. In Prague 11, the share of pickups with optimal fill reaches 70.8 % compared with 43.2 % in the rest of the city, while the share of overloaded pickups remains practically identical. This empirical result serves as a validation bridge to the simulation part of the thesis, in which vehicle routing problems optimizing routes using the OR-Tools library were solved and three collection scenarios were compared across a set of 3 680 glass containers, as glass had the most available data and this commodity provided the greatest room for efficiency improvement. Dynamic collection coordinated at the level of spatial clusters reduces the total distance travelled from 81 067 km to 33 593 km and the corresponding CO₂ emissions from 74.9 t to 31.0 t compared with the static regime, whereas a purely reactive threshold regime without spatial coordination yields a reduction of only approximately fourteen percent in distance travelled. Given the simplifications of the simulation model, the results should primarily be interpreted in relative terms, as an indicator of the magnitude of the potential improvement between regimes. The thesis clearly demonstrates that the benefit of dynamic waste collection management is conditional on combining data-driven collection planning with intelligent route planning and that it represents a direction with demonstrable support in the real operation of Prague's waste management system.