Abstract:

This paper presents the results of assessing the artificial neural network usability to predict surface subsidence, caused by underground coal mining. In this paper, a 2-layer feedforward network are used. Training and testing data are taken from the subsidence forecast model that has been demonstrated to fit with geological - mining conditions in Quang Ninh coal seams. Assessment of predictability of the neural network after training period was conducted in 3 geological - mining conditions which are absolutely different from the training conditions. The largest differences between predicted and real values, corresponding to 3 cases of prediction, are 0.127m, 0.212m and 0.019m respectively. The largest RMS of 3 cases is 0.106, equivalent to 5% of maximum subsidence. This result is a premise to propose a neural network model for prediction of subsidence due to underground mining in Quang Ninh coal basin.