Study On Prediction For Gas Concentration In Fully-Mechanized Coal Mine Face Based On R Language

As one of the five big disasters in coal mine,gas disaster has a larger proportion in coal mine accidents.Effective analysis of gas monitoring data on-site to gain accurate and reliable gas concentration prediction,can provide decision support in preventing gas disaster.For real-time gas monitoring data time series,we study time series decomposition-combination prediction method based on R language.The main contents are presented as follows:On the basis of statistical analysis and pre-processing of original gas monitoring data in working face,gas concentration time series is decomposed into trend part,periodic part and random part,combining with the coal mine's actual production situation of"production-maintenance-production" three alternating phases,to analyze the changing law of the three parts.Use the single exponential smoothing,Holt exponential smoothing and Holt-Winters exponential smoothing method to fit the random part,the result shows that,the single and Holt exponential smoothing fittings have lagged results,the Holt-Winters exponential smoothing can reflect the real changing law of the random part within reasonable fitting error range.Further comparison analysis shows that Holt-Winters exponential smoothing is suitable for fitting the random part.Under the original gas monitoring data sampling frequency of every 5 minutes,analyzing the prediction error of the ARIMA direct multi-step prediction,piecewise direct multi-step prediction,decomposition-combination prediction,the comparision results shows that the decomposition-combination prediction has the minimum error and the mean absolute percentage error is 11.71%.Comparing the prediction error of ARIMA direct multi-step prediction and decomposition-combination prediction under different frequencies,the results shows that,the prediction error presents a downward trend with the increase of the time interval.Under the decomposition-combination prediction error is smaller.Use 205 working face's gas monitoring data in HuangLing No.2 coal mine as the research object to verify the decomposition-combination prediction method,by predicting the gas concentration in top coner and tailentry of the fully-mechanized coal mine face,the mean absolute percentage error of gas prediction in top coner and tailentry are 7.76%and 5.45%respectively,and the prediction accuracy is higher.By using R language to study the prediction method of gas monitoring data time series in fully-mechanized coal mine face,the results shows good visual effect and higher prediction accuracy,providing a new method and technique ways for analyzing gas monitoring data.