MLRA2021 groundwater time-series forecasting
Machine learning prediction event for the international conference in "Machine learning & Risk assessment in geoengineering"
Machine learning as a technique is increasingly used as a "tool" to handle challenging problems in geotechnical engineering. The number of successful use-cases is growing rapidly, which can be seen from the number of related research papers and real-life applications.
The machine learning competition is organized as an event at the MLRA2021 (Machine Learning and Risk Assessment in geoengineering) Conference in Wroclaw, Poland, in October 2021 (conference website). The results from the contest will be presented at the conference. Geotechnical engineers (students and practitioners) are invited to take part in the competition. Participants are encouraged to share knowledge in the Discussion forums while participating.
The task is to perform time-series forecasting on 2 groundwater sensors of 3 types, in total 6 sensors, for a total period of 90 days:
- pore pressure, unit - KPa
- water level in core-drilled hole, unit - MOBT (meter below surface level of drillhole)
- water level in lake, unit - mVs (water level in meter relative to a reference level)
Example plots from the 3 types of sensors can be found under "Code" fane.
The sensor-data comes from the combined railway and road project called FRE16, located in the southeastern part of Norway: Link to project. See the map in the figure below.
Construction start is planned to 2022 for the 40 km long megaproject, containing long tunnels, open pits, bridges and railway-stations. Environmental concerns is an important issue in the project. One of these issues is to preserve the natural ground water level. A decreasing ground water level due to constructing activities will have detrimental effects on nature and buildings in a broad zone along the project.
In order to carry out the correct actions to mitigate a decreasing ground water table during construction, it is vital to have information about what will be the natural ground water level, un-affected by construction activities. The actual measured values in the construction period can then be compared with the forecasted values in order to highlight construction influence on the local ground water table.
Let's hope machine learning can help us to do a realistic forecasting that is valuable for making decisions.:-)