Research On Coal-rock Identification Technology Based On GPR And PSO-BP Neural Networ

Realization of full machine intelligent unmanned operation is the trend of coal underground mining,coal rock identification as a key technology to realize intellige nt construction of coal mine working face,it is of great theoretical significance and practical significance to conduct in-depth research on it.There are various methods of coal rock identification,among which radar detection method has the advantages of small size,convenient operation,strong detection ability and low cost,and has b een widely used and developed.In this paper,we establish the physical model of co al-rock interface,use ground-penetrating radar for coal-rock layer tracking,analyze the law of radar single-channel wave,and use PSO optimized BP neural network m achine learning algorithm for intelligent prediction,in order to realize the dynamic i dentification of unknown coal-rock interface.The main conclusions of this paper ar e as follows:1.The radar prediction experiments show that: the stability test images of rada r all present a curved and smooth reflective waveform set with high repeatability an d accuracy;the performance test images of radar respond strongly to strongly reflec tive materials with outstanding signal characteristics.The information comparison o f the radar images verifies that the instrument used in the experiment has high relia bility.2.The relative permittivity test experiments show that the relative permittivity and water content show positive correlation,the relative permittivity changes are s mall,and the coal samples in this area have weak water absorption ability;48 sets o f univariate crossover experiments under different parameters(water content,coal-r ock interface connection,coal-rock inclusions gangue)show that coal rock contains multiple inclusions and changing the relative permittivity difference of coal-rock m ixed media has the greatest influence on coal-rock interface identification.3.Establish a prediction model based on PSO-BP neural network.The 48 sets of physical experimental radar raw data are extracted from the feature values,and t he ordinary BP neural network prediction model and PSO-BP neural network predi ction model are trained and learned respectively.The results show that the predictio n accuracy of PSO-BP neural network for coal rock layers is improved from 95% to98%,and the optimization effect is obvious,and the convergence speed is improve d,which has strong applicability in coal rock identification.