Research On Intelligent Identification Model Of Microseismic And Blasting In Deep Mines

With the large-scale mining of shallow minerals and the rapid depletion of resources,the world has gradually entered the era of deep mining.However,as the mining operation goes deeper,great depth,high in-situ stress,and high geothermal temperature complicate the engineering environment.In addition,when the rock mass is subjected to external disturbance,the cracks within further evolve into fractures.The continuous expansion of the fractures will eventually release energy in the form of stress waves,which leads to geotechnical hazards such as rock mass instability,goaf collapse,roof collapse,and rockburst.As a result,rock mass stability in deep mines is crucial for the safety of employees and equipment,as well as for reasonably controlling the area where ground pressure is focused.The vibration signals generated by the rock mass rupture during the rock excavation is monitored in real time.Analyzing and identifying the source models of different signals lay the foundation for disaster prevention and control.The work in this study starts with analyzing the original waveform of the microseismic monitoring data.By extracting the waveform features and constructing an automatic identification model of microseismic/blasting events with strong learning ability,a study is carried out on topics such as feature extraction,feature selection,and supervised learning.Aiming to build an identification model of microseismic/blasting events with strong learning ability,the texture features,color features,and shape features in the waveform image were firstly selected and extracted.Then,based on the supervised learning method of Kernel Extreme Learning Machine(KELM),three identification models were proposed,namely,the model based on the Orthogonal Harris Hawk Optimization Algorithm(OHHO),the model based on the Chaotic Arc Adaptive Grasshopper Optimization Algorithm(AGOA),and the model based on the Orthogonal Sine Cosine Optimization Algorithm(OLSCA).The main achievements are as follows:(1)The current status of the research on automatic identifying rock’s microseismic and blasting in deep areas is comprehensively reviewed.The development trends and problems faced in these fields are analyzed and discussed.At the same time,the research status of swarm intelligence optimization and machine learning is briefly introduced,and the research status and existing problems in the Harris Hawk Optimization Algorithm(HHO),the Grasshopper Optimization Algorithm(GOA),the Sine Cosine Optimization Algorithm(SCA),and the KELM algorithm are mainly analyzed and discussed.In addition,some very commonly used feature extraction and selection methods are also briefly introduced.The discussion and analysis of them are the basis for the following research work.(2)Extract the characteristics of waveform images from various sources collected by the microseismic monitoring system.In the feature extraction of the source waveform image,the texture,color,and shape features were extracted.When extracting texture features,the global texture features were extracted by the method of Gray-level co-occurrence matrix(GLCM),but the local binary patterns(LBP)method was also used to extract the local texture features from the waveform images.The color moment method was used to extract various color features of the waveform images when extracting the color features.In the extraction of shape features,the shape features of various waveform images were extracted based on the geometric characteristics.(3)Explore an effective method for identifying the microseismic signals in addition to the traditional method of identifying the source parameters.Based on the supervised learning method of KELM,the works intensely studied the optimization method for hyperparameters,the selection method for waveform image features,the Orthogonal Harris Hawk Optimization Algorithm,the Grasshopper Optimization Algorithm,and Sine Cosine Optimization Algorithm in KELM.(4)Establish an identification model capable of identifying microseismic/blasting events.The works took the swarm intelligence optimization algorithm as the core and combined it with KELM to solve the problem of KELM’s low identification accuracy due to its hyperparameters.Combining the KELM with the wrapping feature selection method reduces the influences from feature redundancy on the identifying accuracy.By optimizing the hyperparameters and feature sets of KELM through OHHO,AGOA,and OLSCA,the works established a series of microseismic/blasting events identification model.In addition,to further improve the performance of identifying model,an integrated model that combines the above three identification models was proposed.