Study On Classification And Source Location Method Based On Knowledge For Multi-channel Micro-seismic Waveform

Rockburst is a typical coal mine disaster.It is a huge threat to the security of coal mine production.Therefore,the prevention and early warning of rockburst disasters is very important.The micro-seismic monitoring technique is a real-time and continuous monitoring method,which has been widely used in the geotechnical engineering,such as: mining,petroleum,and slopes monitoring.It is an effective method to timely warn rockburst by monitoring the micro-seismic signals of coal mine.To identify waveforms is the basis of data processing,and locate the source of causing shake is one of the most critical things to be determined.However,the micro-seismic waveform of coal mines disturbed by various noises is complicated.The existing identification systems and positioning methods have limited performances.Thus,how to improve the recognition performance of the micro-seismic waveforms and the localization accuracy of earthquake sources is a problem that needs to be solved urgently.To this end,the domain knowledge and the imbalance characteristics of the microseismic data are both taken into account in the paper.Moreover,because the number of monitoring channels plays a significant impact on the positioning accuracy,the inversion positioning method of earthquake source is studied.The main contributions of this paper is as follows:(1)For a multi-channel micro-seismic dataset,a knowledge-based SMOTE classification method is proposed to determine the effective micro-seismic waveforms.After preprocessing the micro-seismic data sampling from the SOS micro-seismic monitoring system,an imbalanced training dataset is constructed in terms of the spatial correlations among the micro-seismic signals in the time window.According to the imbalanced rate of the training dataset,the minority data is oversampled by SMOTE based on the sampling rate.Finally,an ensemble learning method is employed to implement the classification.(2)A micro-seismic classification method based on dual evolutionary ensemble learning is proposed.In the inner ensemble model,hybrid sub-classifiers are built by three kinds of learning methods and their parameters are optimized by grid search,with the purpose of enhancing the diversity among them.For each sub-dataset,the classifier having the best classification accuracy is selected as the base classifier.In the outer ensemble model,a multi-modal genetic algorithm with a niche strategy is employed to seek all optimal combinations of base classifiers in terms of classification performances.The combination aggregating a smallest amount of base classifiers by the weighted sum forms the final ensemble structure.The experimental results prove that the performance of the proposed algorithm is stable and is superior to other ensemble learning methods.(3)To position the source of micro-seismic signals,a hybrid multi-objective localization method based on multi-objective particle swarm optimization and simulated annealing is proposed.As we known,the number of detectors involved in positioning system has a direct influence on the accuracy.However,too many positioning channels will increase the computation cost without the effective improvement of the positioning accuracy.Therefore,to choose the appropriate number of detectors for positioning is an important issue.To address the issue,a multi-objective position model that takes both the number of detectors and the positioning accuracy into account is built with the condition of homogeneous medium.Following that,a hybrid positioning algorithm based on multi-objective particle swarm optimization and simulated annealing is presented.The proposed method provides a better initial population for global evolution and then the simulated annealing algorithm for local search effectively prevents falling into the local optima.The experimental results show that the proposed algorithm has best positioning accuracy.The thesis has 28 figures,20 tables,and 142 references.