Compaction property prediction of mixed gangue backfill materials using hybrid intelligence models: A new approach

Abstract Solid backfill coal mining has become one of the main means of green mining in coal mines because of its ability to control surface subsidence and disposal of surface gangue. Mixed gangue backfill material (MGBM) is the key factor for stratum control in solid backfill mining, and its compact mechanical performance directly affects the efficiency of backfill mining. In order to better carry out backfill mining design and backfill effect evaluation, a new hybrid artificial intelligence model integrating support vector machines (SVM), differential evolution algorithm (DE) and gray wolf optimization algorithm (GWO) is proposed to predict the compaction property of MGBM. The cement, lime and fly ash materials are selected to be mixed with gangue backfill materials and a large number of compaction tests using a self-made circular cylinder barrel are carried out to provide the dataset for the DGWO-SVM hybrid model. The input variables of this model include cement content, lime content, fly ash content and overburden stress, and the output variable of the model is the compaction property of MGBM. The performance of the DGWO-SVM model is evaluated by R2, MAE and RMSE. The predictive results indicate that the DGWO-SVM hybrid model can accurately predict the compaction property of MGBM, and the R2 of the training set and the testing set are 0.9518, 0.9137. Meanwhile, the relative importance of each input variable is implemented using the MIV method, and the importance scores of cement content, lime content, fly ash content and overburden stress are 0.3266, 0.0738, 0.2448, 0.3548, respectively. The research results can provide guidance for the optimization design of solid backfill mining.