Use of Neural Networks to Forecast Seismic Hazard Expressed by Number of Tremors Per Unit of Surface

Abstract

The seismic and rock burst hazard should be considered as one of the most important hazards in Polish hard coal and copper ore mines. Over the last several years, Upper Silesia has witnessed a constant process of limiting the extraction of hard coal. Despite a significant decrease in production, no reduction in the number of registered high-energy tremors with energy higher than 105 J. is observed. Therefore, it is important to develop the existing methods for assessing the state of seismic hazard and seeking new ways to estimate the level of this threat. The article presents the results of research aimed at determining the possibility of using artificial neural networks to forecast the level of seismicity induced by mining operation. The data used in the conducted research came from the area of operation of a heavily seismic USCB mine. The presented research results showed that in the case of the discussed area, the use of the new tool, which is an artificial neural network, allowed to obtain good results of the forecast of the number of tremors. For the data set in question, the neural network with optimal architecture is composed of twelve neurons in the entrance layer, two neurons with bipolar characteristics in the hidden layer and one neuron with linear or bipolar characteristics in the initial layer. The results of calculations made for the highly seismic area of mining works carried out in a hard coal mine confirmed the possibility of using neural networks to estimate the changes in the size of induced seismicity associated with the deposit operation.