Research On Geological Structure Identification Of Channel Wave Based On Deep Learning

The hidden geological structure in the mining face is the main factor affecting the safety production of coal mines.It is extremely important to accurately predict the geological conditions of the excavation area,to lay out support and anti-outburst measures in advance,and to ensure the safe and efficient production of coal mines.This paper analyzes the limitations of the current geological structure identification and focuses on deep learning and machine learning algorithms.The paper proposes a classification and recognition model of channel wave geological structure based on convolutional neural network and a prediction model of small fault location boundary.The model construction process is mainly divided into three stages:the establishment of a slot wave data set,model training,and identification and predictionDuring the establishment of the channel wave data set,the geological features of the four geological structures of the goaf,scour zone,fault and subsided column were first studied,as well as their impact on the surrounding coal seams and surrounding rocks.Four geology were constructed in the COMSOL software to construct the model,the small fault has constructed two sets of models with different positions and intervals.A 19 × 3 phased-transmitting receiver array was set up,using the Riker as the transmitting source.By controlling the location of each source,the phase of each source was controlled indirectly,and the total emission angle was changed.A channel wave echo signal data set was established.Taking into full consideration that the collected original data has many differences and the data dimension is large,the data is pre-processed before the model training,such as standardization and dimensionality reduction,and the data set is divided.In the model training stage,the classification and recognition model of channel wave geological structure based on convolutional neural network proposed in this paper is built on the basis of the classic model LeNet-5.First,find the optimal number of layers and nodes of the model through cross-validation.After determining the initial network structure of the model,the effects of batch size,maximum iterations and learning rate on classification accuracy were analyzed.Then optimize the network parameters by gradient descent method,combined with L2 parameter regularization strategy and Dropout technology to prevent over-fitting of the model.Finally,CNN-based geological structure classification and recognition model was determined.Small fault location recognition model fine-tuning composition on classification recognition model.In the classification prediction stage,input the channel wave data of the test set that did not participate in the model training into the constructed model,and then the structure category corresponding to the channel wave signal vector can be output,so as to realize the prediction of the geological structure type and position.Finally,the paper verifies the validity and reliability of the classification recognition model and location recognition model.In 400 random channel wave signal samples,the classification prediction accuracy rate reached 94.0%,and the model loss error was 0.47.The accuracy of the small fault location interval between 10m and 20m reached 91.0%,and the accuracy of the location interval ?5m reached 89.2%.Then it compares the classification accuracy and position recognition results of the four machine learning algorithms of K nearest neighbor algorithm,support vector machine,random forest,and naive Bayes with the algorithm model constructed in this paper.The results show that the algorithm model constructed in this paper is significantly better than the other four algorithms in identifying and predicting results.Comparison with traditional methods in prediction accuracy and processing complexity further proves the advantages of the algorithm model.It can be seen that the prediction model of channel wave geological structure based on convolutional neural network has high prediction accuracy and can be used for the identification and prediction of groove wave geological structure.