Seismology for rockburst prevention control and prediction 2

This final report on the SIMRAC project GAP409 presents a method (SOOTHSAY) for predicting larger mining induced seismic events in gold mines, as well as a pattern recognition algorithm (INDICATOR) for characterising the seismic response of rock to mining and inferring future behaviour. The pattern recognition algorithm codifies and quantifies previous intuitive, qualitative analyses. The predictor, based on accelerating seismic release, depends on the existence of sufficient data, a past history in an area, and power law accelerating behaviour of a seismic time series.
 
Literature surveys form a crucial part of the research throughout the life of the project since prediction is a fast-developing and changing field. Several experts were consulted throughout the project, either in person or via e-mail in order to ensure that the latest techniques are taken into account.
 
The pattern recognition algorithm has been applied to several mines in the Carletonville area, as well as in the Free State, with clear patterns shown to emerge when input boundary conditions and thresholds are tuned to an area.
 
SOOTHSAY, the prediction algorithm, was applied to the deepest shaft pillar extraction in world in the West Wits region, as well as to a novel shaft pillar extraction in the Free State. Defining the time series of a specific function on a catalogue as a prediction strategy, the algorithm currently has a success rate of 53% and 65%, respectively, of large events claimed as being predicted in these two cases, with uncertainties in the predicted time of occurrence of a large instability of less than a week. In both cases, the prediction strategies are very far from random, and present a significant improvement over a pessimist strategy of trying to randomly forecast a large event daily by factors of 5.7 and 3.26, respectively.
 
As one way of unearthing analysis artefacts, the algorithm SOOTHSAY was applied to a pseudo-random synthetic catalogue. In this case only one prediction was made, for one function, with a 0% success rate for all functions.
 
The pattern recognition algorithm was developed in such a way that it can be extended to a fully-fledged prediction strategy in its own right. While further rigorous, statistical testing has to be performed on the SOOTHSAY algorithm, it is recommended that the current version be implemented on an experimental basis.
AttachmentSize
PDF icon GAP409REP00501.pdf1.24 MB