Graph Convolutional Networks: Application to Database Completion of Wastewater Networks

Wastewater networks are mandatory for urbanisation. Their management, including the prediction and planning of repairs expansion operations, requires precise information on their underground components (manhole covers, equipment, nodes, pipes). However, due to years service increasing number maintenance operations they may have undergone over time, attributes characteristics associated with various objects constituting a network not all available at given time. This is partly because (i) multiple actors that carry out extensions necessarily operators who ensure continuous functioning network, (ii) undertaken changes properly tracked reported. Therefore, databases related wastewater suffer from missing data. To overcome this problem, we aim exploit structure in learning process machine approaches, using topology relationship between components, complete values pipes. Our results show Graph Convolutional Network (GCN) models yield better than classical methods represent useful tool data completion.