GeoKinesia’s predictive solution is a unique, InSAR based early warning tool, which allows detecting early and generating an alarm for a deformation, which may become a threat to an object or an operation.
The solution makes use of the Long Short-Term Memory (LTSM) recurrent neural network and improved coherent points selection technique. This is one of the first successful applications of the Artificial Intelligence (AI) and LSTM model, in particular, for InSAR processing and early warnings.
Cadia mine, Australia, March 2018 collapse and early warning model prediction using Sentinel-1 data. The collapse area is circled
The main underlying principle is time series forecasting. The model is trained, using univariate or multivariate time series (i.e., previously observed values of single or multiple parameters) to forecast future values. In addition to the accurately predicted results, the model could predict linear trends of the settlements on reclaimed lands and the buildings’ seasonal pattern. To ensure the quality of the forecast, the system utilizes adjacent time series to decrease the possibility of falsely detected safe areas. Once a difference between the forecasted and actual value is detected and checked, a warning is generated.
The model has been trained and then tested on a number of large-scale accidents, including Cadia, Australia, Brumandinho, Brazil, Cobre Las Cruses, Spain etc. and demonstrated consistently reliable results.
This is a highly effective tool to improve the risk management capabilities and get early warnings for a broad range of practitioners in a number of sectors such as civil protection, mining, civil engineering, construction, infrastructure monitoring as well as for the managers responsible for objects stability and general safety.
Cobre Las Cruces, Spain, January 2019 collapse and early warning model prediction using Sentinel-1 data. The collapse area is circled.