Research on artificial intelligence inversion methods for advanced detection of mine transient electromagnetic method

Abstract

During coal mining process, water inrush accidents primarily occur in the stage of roadway excavation and it accounts for approximately 70 %. At present, the most effective method is the mine transient electromagnetic advanced sounding for detecting water rich zones ahead of roadway excavation. Currently, the primary data processing technique for mine transient electromagnetic detected data is the fan shaped apparent resistivity method. However, it cannot meet the demands of coal mine geological transparency and precise mining. Besides, coal mine roadways are located in the three dimensional space and traditional transient electromagnetic 3D inversion involves massive computational quantity and it consumes a long computation time. Hence, it also cannot meet the requirement of rapid processing and real-time monitoring. Machine learning inversion algorithms can learn from large datasets and it can quickly identify target bodies. Based on coal mine hydrogeological characteristics, 1500 coal mine hydrogeological models were established to train sample data. Besides, finite volume method is employed to calculate mine transient electromagnetic responses to form sample database. During the 3D forward modeling of coal mine transient electromagnetic, three scenarios of coal mine hydrogeological geology are considered and they are disaster geological bodies located in the coal seam, floor and roof, respectively. Moreover, a 3D convolutional neural network was developed to conduct three-dimensional inversion calculations for transient electromagnetic data in coal mines. Furthermore, the comparison between the forward model and the inversion results have been done and it demonstrates the feasibility of machine learning for 3D inversion in transient electromagnetic applications within mines. Finally, machine learning inversion was conducted by field data from Huaibei coal mining and the inversion results were generally consistent with information of drilling geology. Hence, the field data inversion demonstrates its effectiveness. Additionally, the total inversion time was 485 milliseconds and it is less than one second. All in all, this high inversion speed lays the foundation for real-time processing and intelligent detection of transient electromagnetic signals in the coal mine.