Detection And Early Warning Of Gas Anomaly Based On Correlation Analysis

With the constant promotion of intelligent mine construction,coal mining enterprises have accumulated massive monitoring data from coal mining,tunneling,mechanical and electrical engineering,transportation,ventilation and other production links of coal mines.However,numerous outliers often exist in the gas monitoring data,which the current mine monitoring systems cannot effectively distinguish the data,the hazard warning can only be realized according to the preset gas concentration threshold,which may lead to certain risks.The detection and early-warning mechanism of gas outlier were deeply studied from the following aspects:(1)Focusing on the low modeling efficiency of gas big data,the method of outlier detection based on the Data Sampling Optimization Logistic Regression(DSO-LR)was established.Firstly,applying the Lagrange Interpolation Polynomial,collected mine gas data and its missing values were interpolated,and the outliers were identified.Secondly,according to the principle of probability sampling,the Euclidean distance of the logistic regression equation established under various sampling probabilities was compared with that under all data,thereby obtaining the optimal sampling probability.Finally,based on the training set,the decision boundary of the regression equation was established.Based on this boundary,outliers were detected.The experimental results indicate that the optimal sampling probability of DSO-LR reaches up to 20%and the computational efficiency is 5 times higher than that of the LR method.(2)In order to solve the problem that K-Means algorithm was sensitive to the initial cluster center,the K-Means outlier detection method based on the initial cluster center optimization was provided.Firstly,the correlation of gas concentration data among various monitoring positions was analyzed.Secondly,according to the periodicity of coal mining process,based on the cluster center of the pre-order data,the initial cluster center of the post-order data was optimized,and the minimum variances of all clusters were selected as the conditions for measuring the data similarity within the cluster.Finally,by adopting the improved K-Means algorithm,based on the experimental data,the outlier detection was implemented.The experimental results indicate that the iteration number of the improved K-Means algorithm is less than that of the original algorithm,and its outlier detection accuracy is higher than that of the original algorithm.(3)In order to achieve the graded early-warning mechanism of gas hazard,based on the analysis of the correlation among outliers,the early-warning model of gas hazard based on the Apriori Model of Weight Optimization(WO-Apriori)was designed.Firstly,the dualization of gas concentration and related outliers detected by applying the above algorithm was implemented,and the learning set of association rule was built.Secondly,the support of the learning set was calculated and the high-frequency term set was identified.Thirdly,under the condition of minimum confidence of weight optimization,association rule within the high-frequency term set was generated.Finally,based on the association rule,the mechanism of the graded early warning was designed,thereby verifying the effectiveness of outliers and achieving the graded early warning of gas hazard.The experimental results indicate that among the 131 outliers detected based on the graded early-warning mechanism,38 outliers are identified as loud noise and warning is not needed;34 outliers are identified as low-grade risks;17 outliers are identified as medium-grade risks;and 42 outliers are identified as high-grade risks and warning of corresponding grade should be given.The early-warning results are consistent with the analysis results of the experts,which verifies the effectiveness of the early-warning mechanism of the WO-Apriori model.