A New Case-based Reasoning Method for Prediction of Fractured Height of Longwall Panels

When longwall mining involves total extraction, it includes the overlying strata movements. In order to better control these movements, the height of fracturing (HoF) must be determined. HoF includes both the caved and continuous fractured zones, and represents the region of the broken ground whereby a hydraulic connection to the mined seam occurs. Among the various empirical models for predicting HoF, the Ditton's geometry and geology models are widely used in the Australian coalfields. This work uses a case-based reasoning (CBR) method in order to predict HoF. The model's variables, including the panel width (W), cover depth (H), mining height (T), key stratum thickness (t), and its distance from the mined seam (y), are selected via the Buckingham's p-theorem. The data set consisting of 31 longwall panels is partitioned into the training and test subsets using the W/H ratio as the primary classifier of a semi-random partitioning method. This partitioning method overcomes the class imbalance and sample representativeness problems. A new CBR model presents a linear mathematical equation to predict HoF. The results obtained show that the presented model has a high coefficient of determination (R^2 = 0.99) and a low average error (AE = 8.44 m). The coefficient of determination for the CBR model is higher than that for the Ditton’s geometry (R^2 = 0.93) and geology (R^2 = 0.97) models. Contrary to the Ditton's models, the performance of the CBR model is consistent regarding the average and standard errors (AE and SE) of the training and test stages. The proposed model has an acceptable performance for all the width to depth ratios to predict HoF.