| Abstract [eng] |
In this Master ‘s Final Degree Project we have planned, analyzed, designed, implemented and deployed geographic information system TAVSIS. It allows users to manage (register, edit, delete) and localize pipe breakage accidents that occur in district heating networks of Kaunas region. These networks are managed by the customer of the project – company “Kauno energija”. In the analysis phase we have researched various publications related to accident prognosis and machine learning. Also, we have collected necessary data, prepared methodology, identified possible problems and how to mitigate them. In the design phase we used requirements collected from the customer to create a detailed system project which was used for the system implementation. The implemented system is based on geographic and web technologies and works through internet as an interactive web map application. Accident localization module is based on supervised machine learning methods and allows its users to find district heating pipe segments with highest accident probability. In addition, we have performed a research to find out if we have used the best supervised machine learning model available. We were able to find a better implementation (AUC score +0,7%). Although the increase in performance was not enough to change the model we are already using. Finally, we have performed an experiment to find out if our machine learning model can localize broken pipe segments in real-life. The results have shown that model needs to be improved. We have added guidelines what should be improved in the future. |
