Data-driven modelling for flood defence structure analysis

We present a data-driven modelling approach for detection of anomalies in flood defences (levees, dykes, dams, embankments) equipped with sensors. An auto-regressive linear model and feed-forward neural network were applied for modelling a transfer function between the sensors. This approach has been validated on a dike in Boston, UK—one of the pilot sites of the
UrbanFlood project— that showed both normal and abnormal sensor behaviour. Comparison of the linear and non-linear mod- els is presented. The suggested model-based anomaly detection approach will extend functionality of the developed Artificial Intelligence component of the UrbanFlood Early Warning System.