In modern blasting engineering research,the matching model of explosive and rock provides a scientific basis for revealing the internal mechanism of blasting process and predicting the economic benefits of blasting system,which has become an irreplaceable important tool.However,due to the diversity and complexity of soil-rock medium and the uncertainty of explosion process,the interaction between explosive and rock is more complex and uncertain,and it is difficult to study the matching of explosive and rock from their interaction process.Earlier studies mainly relied on empirical formulas and field tests for calculation and summary,which often had high eigenvalues and harsh application environment.However,the feature of machine learning is that it only considers the beginning and the result,and does not care about the middle process,which ensures its universality in the study of explosive-rock matching model.The XGBoost algorithm,together with multi-threading,data compression and fragmentation method,has the advantages of high efficiency in the case of largedata amount,and is suitable for training of a large amount of field data.In view of this,a field test was carried out in a mine in Guizhou province,and XGBoost algorithm was used to establish a matching system between explosives and rocks.The network was trained through successful examples,and the trained neural network was applied to practical projects.The results show that the performance of the explosives selected by the matching system based on this method is similar to that of the industrial explosives used at present,and the error is within ±10%,which has a high reliability,and further verifies the rationality of the explosive-rock matching system based on XGBoost algorithm.