Speculation Of Distributed Optical Fiber Monitoring Data Based On Deep Learning

With the proposal of intelligent mining in mines,establishing a large-scale monitoring system for overburden deformation during coal seam mining plays an important role in ensuring safe and efficient mining.Distributed fiber optics are introduced for deformation monitoring of mine overburden,and much of the research is carried out in the laboratory and in the engineering field.However,both in laboratory studies and in practical engineering applications,the complexity and high cost of distributed fiber laying make it difficult to perform intensive laying.This brings difficulties to the large-scale monitoring of overburden deformation and accurate characterization of deformation.To solve this problem,this paper investigates a distributed fiber optic monitoring data inference method for mining overburden deformation based on deep learning,and provides a research basis for the application of distributed fiber optic to intelligent surrounding rock deformation monitoring in coal mines.In this paper,a physically similar simulation experiment of distributed fiber optic monitoring of mining overburden deformation is used as the research background,and the frequency shift values of vertical fiber optic measurement points are used as the data source.The idea of inferring the frequency shift data of virtual measurement points at unburied fiber locations in the overburden by the frequency shift data of existing fibers is proposed,and a hybrid CNN-GRU deep neural network inference model is established.MAE(Mean Absolute Error),RMSE(Root Mean Square Error)are used as performance evaluation metrics of the model for comparison and analysis with the inferred effects of CNN(Convolutional Neural Network)and GRU(Gated Recurrent Unit)networks.The model parameters are optimally searched by GA(Genetic Algorithm)in the simulation experiments.The simulation experimental results show that the CNN-GRU model has a MAE of 0.6818 and an RMSE of 0.8617 in the projection of the frequency shift value of the working plane gradually approaching the vertical fiber,which are 0.8611 and 1.0458 lower than the MAE and RMSE of the CNN projection method,and 0.729 and 0.7579 lower than the MAE and RMSE of the GRU projection method;the MAE of the CNN-GRU model is 9.4213 and the RMSE is 11.6665 in the projection of the working face over the fiber and gradually moving away,which are 11.2544 and 12.0161 lower than the MAE and RMSE of the CNN projection method,and 6.3011 and 6.1463 lower than the MAE and RMSE of the GRU projection method.The simulation experimental results show that the hybrid CNN-GRU model has higher accuracy for the data inference of the frequency shift value of the unburied vertical fiber virtual measurement points,which realizes the data inference of distributed fiber monitoring.Data speculation is predicated on the need to ensure the integrity of existing fiber optic measurement point data.Therefore,this project also investigates the method of filling missing data at fiber failure monitoring points,and establishes the method of missing data filling based on XGBoost.The project completes the missing data filling simulation experiment at 10%,20%,30%,40%different failure rates,and uses R2 and MAE as a performance evaluation index,compares to the filling effect of the three machine learning methods of K nearest neighbor,SVR,and LightGBM.The simulation experimental results show that the data filling by XGBoost method,during the gradual approach of the working surface to the vertical fiber,the R2 is 0.9912,0.9831,0.9683,0.9095,and the MAE is 0.7203,0.7218,0.8411,1.0505 at 10%,20%,30%,40%failure rate.In the process of working face over the fiber and gradually moving away,the R2 is 0.9934,0.9911,0.9619,0.9187 and MAE is 3.9035,6.5147,8.6042,12.8883 at different failure rates using XGBoost method.The results show that the XGBoost method has higher accuracy and better generalizability compared with the other three methods in filling missing data at fiber failure monitoring points.