Application of Artificial Neural Networks in Assessing Mining Subsidence Risk

Abstract

Subsidence at abandoned mines sometimes causes destruction of local areas and casualties. This paper proposes a mine subsidence risk index and establishes a subsidence risk grade based on two separate analyses of A and B to predict the occurrence of subsidence at an abandoned mine. For the analyses, 227 locations were ultimately selected at 15 abandoned coal mines and 22 abandoned mines of other types (i.e., gold, silver, and metal mines). Analysis A predicts whether subsidence is likely using an artificial neural network. Analysis B assesses a mine subsidence risk index that indicates the extent of risk of subsidence. Results of both analyses are utilized to assign a subsidence risk grade to each ground location investigated. To check the model's reliability, a new dataset of 22 locations was selected from five other abandoned mines; the subsidence risk grade results were compared with those of the actual ground conditions. The resulting correct prediction percentage for 13 subsidence locations of the abandoned mines was 83-86%. To improve reliability of the subsidence risk, much more subsidence data with greater variations in ground conditions is required, and various types of analyses by numerical and empirical approaches, etc. need to be combined. ANN provides tools for optimising operations, equipment selection, and problems involving large amounts of information that humans cannot easily assimilate in the process of decision-making. Hence, the neural network can serve as a tool for determining the relative importance of the factors influencing the stability of underground objects [18].Since the 1990s, when a number of ANN-based systems were introduced in the mining industry [19], ANN has significantly improved the current approach towards predicting abandoned mine subsidence risk [20][21][22][23]. In particular, Oh et al. [24] evaluated the performance of predictive Bayesian, functional, and meta-ensemble machine learning models in generating land subsidence susceptibility maps. Kim et al. [25] also constructed a hazard map for possible ground subsidence around abandoned underground coal mines in Korea using an ANN together with a geographic information system (GIS). Zhao and Chen [26] used an ANN to predict ground subsidence at a metal mine because ground subsidence is influenced by many factors which are fuzzy and nonlinear, and the ANN has a great function to be able to handle such nonlinear problems, without knowing specific mine conditions. Meanwhile, Mine Reclamation Corporation (MIRECO) who is responsible for the safety and management of abandoned mines and implements mine reclamation projects in Korea has made several attempts to reliably predict subsidence by numerical, statistical, probabilistic, and/or model approaches [27-31] prior to this study. However, the prediction results were not up to par with MIRECO's standards because they were not appropriate for the assessment of many as well as extensive abandoned mines.Finally, MIRECO decided to adopt ANN as well to predict subsidence consideri...

Prediction of Utility Tunnel Performance in a Soft Foundation during an Operation Period Based on Deep Learning

“…By using the trial-and-error method, the determined hyperparameters of the DBN model are summarized in Table 5. As one typical artificial intelligence method, the ANN is generally used for prediction study [27,[35][36][37]. Therefore, it is used here for comparison.…”

Predictive Modelling for Blasting-Induced Vibrations from Open-Pit Excavations

“…The input and output layers consist of causal and result parameters, respectively. The training algorithm of the ANN used in this study was back-propagation, which is the most efficient ANN training algorithm available [16,17]. In back-propagation, the output values calculated in the forward direction through weights and biases are used to calculate training errors from the true values.…”

Prediction of Mine Subsidence Based on InSAR Technology and the LSTM Algorithm: A Case Study of the Shigouyi Coalfield, Ningxia (China)

“…As a result, an increasing number of deep-learning algorithms are being utilized to research long time series prediction, thereby making it possible to obtain mine subsidence characteristic information and dynamic forecasting in mining areas [9,10]. Deep-learning algorithms such as artificial neural networks (ANNs) [11] and recurrent neural networks (RNNs) [12] can be used to analyze long time series data and predict mine subsidence characteristics and dynamics. However, these methods need further improvement to improve the accuracy of prediction.…”