Exploration on coal mining-induced rockburst prediction using Internet of things and deep neural network

Jinrui Zhang
DOI: https://doi.org/10.1007/s11227-022-04424-4
IF: 3.3
2022-03-27
The Journal of Supercomputing
Abstract:With the increase in mining depth and intensity, the threat of rockburst caused by high stress in stope to coal mining (CM) safety production is becoming ever more serious. However, it is more difficult to predict and prevent rockburst in deep mining coal seams under complex environments. Firstly, this paper analyzes the detection requirements in the tested environment and designs a new multi-parameter monitoring platform for CM-induced rockburst through wireless sensor network, communication protocol and Internet of things (IoT) technology. Then, two improvements are made to the original faster region-convolutional neural network (Faster R-CNN). The first point is to replace Faster R-CNN's common feature extraction network. The second point is to fuse the region proposal network structure of Faster R-CNN with feature pyramid network and combine Faster R-CNN with probabilistic neural network (PNN). Finally, the proposed system is tested. The outcomes corroborate that under the complex external geographical environment, the communication distance of the proposed CM-induced rockburst-oriented multi-parameter monitoring platform is far, and the proposed system meets the actual needs. The prediction results of the original PNN show that the prediction error rate of the training is 0.999%, and the prediction error rate of the test set is 0.995%. The prediction results based on PNN + optimized Faster R-CNN show that the prediction error rate of training is 0.623%, and the prediction error rate of the test set is 0.409%. Therefore, the prediction effect of PNN + optimized Faster R-CNN is better than PNN. The proposed CM-induced rockburst-oriented dynamic pressure prediction system provides some ideas for applying IoT and deep neural network technology in the CM industry.
What problem does this paper attempt to address?