Application of the autoregressive integrated moving average (ARIMA) model in prediction of mining ground surface displacement

Underground mining, regardless of the excavation method used, has an impact on the terrain surface. For this reason, continuous monitoring of the ground surface above the excavations is necessary. Deformations on the ground surface occur with a time delay in relation to the mining works, which poses a risk of significant deformations in built-up areas, leading to building disasters. In addition to monitoring, it is therefore necessary to forecast displacements, which at present is usually done using the empirical integral models, which describes the shape of a fully formed subsidence basin and require detailed knowledge of the geological situation and parameters of the deposit. However, insufficiently precise determination of coefficients may lead to significant errors in calculations. Machine learning can be an interesting alternative to predict ground displacement in mining areas. Machine learning algorithms fit a model to a set of input data so that it best represents all the correlations and trends detected in the set. However, the fitting process must be controlled to avoid overfitting. The validated model can then be used to detect new deformations on the ground surface, categorize the resulting displacements, or predict the value of subsidence. In this case ARIMA model (Auto-Regressive Integrated Moving Average) was used to predict deformation values for single points placed in the centers of the subsidence basins in the LGCB (Legnica-Głogów Copper Belt) area. The InSAR time series calculated using the SBAS method for the years 2016-2021 was used as input data. The results were compared with the persistence model, against which there was an improvement in accuracy of several percentage points.