Feasibility study for seismicity forecasting in seismically active underground mines
Reference number | |
Coordinator | Luleå tekniska universitet - Avdelningen för geoteknologi |
Funding from Vinnova | SEK 500 000 |
Project duration | March 2020 - September 2020 |
Status | Completed |
Venture | The strategic innovation programme for Swedish mining and metal producing industry - SIP Swedish Mining Innovation |
Call | Towards a sustainable development in the mining and metal extraction industry |
Purpose and goal
Earthquakes may cause severe damage to the mine and expose personnel working in the mine at risk and potentially loss of life. Thus, forecasting of earthquakes and increased seismic hazard are of considerable importance in order to minimise the risks to personnel working underground. Based on available data from the Kiirunavaaramine Mine, the purpose of the study was to identify specific problems in relation to earthquake forecasting and to define a way forward to a real-time forecasting procedure.
Expected results and effects
1) Review of AI methods for earthquake prediction 2) Pre-processing: normalization, data feature extraction in both space and time, noise removal 3) Testing of AI methods - deep learning’s LSTM and density-based clustering’s OPTICS, SOM 4) Decision Making approach most suitable for Kiruna Mine identified 5) Conclusions: Outliers and duplicates analysis as well as normalization are very important The most promising ML approach - integrated predictive one - clustering, deep learning and classification models The open/close decisions have to be made closely with decision makers
Planned approach and implementation
Software to estimate the performance of machine learning (ML) models for the prediction of seismic activity in underground mines was implemented. It is based on state of the art Python libraries for ML like TensorFlow and scikit-learn. The implementation design includes the following analytic processes: Data import + cleaning Spatiotemporal clustering Time-Series data preprocessing Measurement of prediction performance Each process step is an independent module, whose common basis is set by databases and configurations for data transformations and execution properties.