Rockburst is one of the key scientific problems that must be solved in large-scale underground geotechnical engineering and deep mineral resource mining.The safety of personnel and equipment on site was directly threatened by rockburst.Rockburst could be effectively avoided and controlled in time by scientific and accurate rockburst prediction of intensity classification.Through the analysis of six rockburst engineering examples，on the basis of the factors，characteristics and causes of rockburst，a rockburst prediction index system composed of four evaluation indices，i.e.，tunnel-wall surrounding rock’s maximum tangential stress，rock uniaxial compressive strength，rock uniaxial tensile strength，and rock elastic energy index was established.With reference to other rockburst intensity classification schemes，considering the intensity of rockburst occurrence and the main influencing factors，the rockburst intensity was divided into four levels：None rockburst（Ⅰ），slight rockburst（Ⅱ），intermediate rockburst（Ⅲ） and strong rockburst（Ⅳ）.According to the selected rockburst evaluation index and rockburst intensity grade，a literature survey method was used to establish a database containing 301 groups of rockburst engineering examples，which would be used as the sample data for rockburst prediction.In order to accurately and reliably predicted rockburst disasters，machine learning technology was introduced.First，a random forest-based rockburst evaluation index importance analysis model was established，a new index weight calculation method of random forest-analytic hierarchy processs was proposed，and the rockburst prediction model based on the RF-AHP-cloud model was constructed.Then，the firefly algorithm based on good point set variable step strategy was introduced to optimize the penalty parameters and radial basis function parameters of the support vector machine，and the rockburst prediction model based on ⅠGSO-SVM was constructed.Finally，the Dropout method was used to regularize the model，and the improved Adam algorithm was used to update weight，and the rockburst prediction model based on DA-DNN was constructed.The effectiveness and correctness of the three models were validated by the prediction results of 60 groups of rockburst engineering examples.The research results show that：The DA-DNN，ⅠGSO-SVM，and RF-AHP-cloud model have prediction accuracy rates of 98.3%，90.0% and 85.0%.The core of rockburst intensity classification prediction based on cloud model is weight determination，and the RF-AHP weight calculation method proposed in this paper has a good effect.The data-driven ⅠGSO-SVM and DA-DNN models are based on rockburst engineering instance data.Through data mining，the rockburst intensity level can be effectively predicted，and higher prediction accuracy can be achieved by improvement.The theory of DA-DNN model is easy to understand，the coding is relatively simple and it is easy to implement.As various underground geotechnical engineering develops deeper，rockburst disasters occur frequently，the amount of rockburst data is increasing，and the DA-DNN model has a wider application prospect.

http://www.goldsci.ac.cn/article/2020/1005-2518/1005-2518-2020-28-6-920.shtml