Study On Gas Concentration Prediction And Pre-warning Based On Real-time Monitoring Data Of Safety Monitoring System Safety Technology And Engineering

Gas disaster has long been a major threat to safe production of coal mines in China. Through effective analysis of gas monitoring data in mine-site to gain accurate and reliable gas concentration prediction, pre-warning of gas abnormal situation can be achieved to enhance the gas safety management in the mine. Based on real-time gas monitoring data, the methods for gas concentration prediction and pre-warning analysis are studied in the dissertation, in which the main research contents are presented as follows:Study on data pre-processing: The characteristics of real-time monitoring data in a mine is analyzed and data pre-processing is performed, including abnormal data substitution, supplementary treatment of missing data and noise suppression of data, thus to eliminate the potential interference from random and uncertain factors as much as possible. The HHT analysis on the time series of gas concentration data is studied through its decomposition into IMF components with different instantaneous frequencies based on EMD model, and suitable prediction methods are selected to reduce the complexity in prediction, and to improve its accuracy.Study on the gas concentration prediction at a sensor site: The effectiveness of gas concentration prediction is taken as the criteria of accuracy assessment in prediction performance. The grey cluster relational analysis and GPR are adopted jointly to determine the best sample dimension through reconstruction of the data dynamically. The data is divided into several categories of strong relevance as dummy variables to express the dynamic characteristics of gas concentration influenced by unavoidably random and uncertain factors, so as to relieve the influence exerted by those factors. Based on HHT analysis, AR and GPR are integrated to achieve an adaptive gas concentration prediction at the sensor site. Study on multi-variable gas concentration prediction in a face region: The monitoring data collected from multi-sensors in a face region is taken as Bayesian network variables, and the multi-variable correlation analysis is performed with Bayesian network learning, and the multivariate time series relating to the gas concentration at the face is formed by Bayesian network inference to extract the most closely related variables, the dimension of multivariate data is determined by chaotic phase space reconstruction, and the multi-variable gas concentration prediction model is established with the gas concentration prediction at the face achieved by GPR application.Study on the judgment of gas abnormality pre-warning: On the basis of gas concentration prediction, two kinds of pre-warning index, i.e., basic index and correlation index are determined. The threshold of basic index is identified by statistical analysis of the gas concentration data, and the threshold of correlation index is determined by correlation analysis of gas concentration data, the analysis of gas abnormality is conducted by comparing the prediction value of gas concentration with the pre-mentioned thresholds to determine the pre-warning level, thus the dynamic and meticulous gas pre-warning analysis is achieved.Study on the application of gas prediction and pre-warning: The data collected from various locations and regions in Rujigou Coal Mine is thoroughly analyzed. The results show that the accuracy of prediction is satisfactory, the range of prediction fields is reliable, the pre-warning thresholds are reasonable, which indicates that the gas prediction and pre-warning analysis is verified.The achievements of gas concentration prediction and pre-warning models are suitable for the real-time analysis of monitoring data, and can provide a new technique to the management and control of gas safety in coal mines.