Research On Disaster Feature Extraction And Information Fusion Of Coal Mine Gas

The coal mine gas disaster is one of the severe natural disasters that restrict the safe production of coal. Monitoring of the relevant parameters, forecasting of running trends and timely discovering gas disasters are of great significance to ensure the coalmine safe production and the miners'security. This thesis is focused upon studying the technology of the feature extraction and information fusion from gas disasters, analyzing of the characteristics of gas disasters, promoting of the methods for feature extraction, character- and decision-level fusions, and modeling the feature extraction, the character-level fusion and the decision-level fusion for the gas disasters.(1) Based on an analysis on the time and frequency domains of the information of gas disasters, the method of time-frequency combined analysis is used, with Gabor transformation and Wigner-Ville transformation respectively employed, to determine the time-frequency characteristic of gas. The author proposes the method and principle of multi- resolution analysis of time and frequency, and then gives the specific multi-resolution way of gas disaster information. This approach solves the problem of time-frequency analysis on gas signals, and provides theoretical evidence for analysis of the time-dependent or non-stationary signals, which resembles the cases of gas disasters.(2) Independent component analysis is a relatively new research field that has attracted the interests of many scholars and research groups in recent years. This method is used in the characteristic analysis of the gas disaster information, giving a multiplexed expression method and separation of the blind source from gas monitoring signals. A model of the independent component analysis of the gas disaster information is established with improvement and optimization, which provides a new theoretical and methodological approach to exactly analyzing the gas disaster features.(3) The feature extraction of gas disaster is studied. The process of feature extraction is analyzed, giving the corresponding evaluation criterion and the factors that should be considered as the feature extraction is carrying out. The problem of the feature extraction from the gas disaster information based on singular value of time-frequency distribution is studied, in which the expression method of the characteristic vectors is presented, and the model of feature extraction of gas disaster information based on singular value decomposition is established, Also, the evaluation criteria of the target identification of the gas disaster information is promoted, and a further progress has been made in developing the theory of gas disaster feature extraction.(4) A maximum entropy model of feature extraction for the gas disaster information is established, with the training algorithm of the maximum entropy presented. The maximum entropy method is statistically one of machine learning methods with the advantages of simplicity in principle and mathematical reasoning. The above model showing strong expression ability enables the multiple characteristics to be conveniently used, among which no independence assumption is necessary. It solves the problem of the multi-characteristic extraction difficulty from gas disasters, and gives a new idea for developing the technology.(5) This feature extraction is studied based on the supporting vector machines. On the basis of the relevant theory of feature extraction with the supporting vector machines, a model of feature extraction for the gas disaster is established based on the...