【作者】 乔美英；

【导师】 马小平；

【作者基本信息】 中国矿业大学 ， 控制理论与控制工程， 2012， 博士

【摘要】 我国是世界上煤与瓦斯突出灾害最严重的国家之一，具有突出矿井多、分布范围广、始突深度浅，突出强度大等一系列问题。由于尚未认清煤与瓦斯突出机理，故完全杜绝煤与瓦斯突出发生还不太可能。尽管目前国内外在局部检测与解危措施方面的研究已近成熟，但在防治煤与瓦斯突出的实施方面存在很大的难度。大量的事实表明，重特大的突出事故大多发生在煤巷掘进工作面，所以研究掘进工作面非接触式动态连续的煤与瓦斯突出预警方法具有重要的现实意义。首先，本文从瓦斯在煤层中的流动理论出发，得出掘进工作面的瓦斯涌出量与煤层瓦斯含量、瓦斯压力、煤层透气性、地质构造等因素有关，由此得出瓦斯突出与工作面瓦斯涌出量的复杂特性密切相关。其次，利用Taken的相空间重构理论来分析工作面瓦斯涌出时间序列的动力学模型。相空间重构参数的选取，文中采用了微熵率最小原则，最终确定所研究实例的几个嵌入维数均大于8，说明工作面瓦斯涌出动力学系统是一个高维的复杂系统。在此基础上，利用功率谱与Poincare截面定性分析工作面瓦斯涌出量时间序列数据具有混沌特性。针对实际时间序列长度有限且噪声未知的情况，在定量混沌特性分析中，提出了基于NLSR的LLE的求取，和基于FCM的关联维数的求取。为了进一步验证所提方法的有效性，利用确定的混沌时间序列Logistic和Hénon系统对算法进行验证，之后将所提算法用于实际时间序列混沌特性判别。LLE的求取结果进一步确定了工作面的瓦斯时间序列数据具有混沌特性。关联维数的确定也进一步验证了前面嵌入维数求取的正确。第三，引入Hurst指数来分析时间序列的长程相关性，并针对传统的R/S分析法存在的缺点，引入了V/S分析法来求其长程趋势特性，同时文中引入有效关联长度来进一步分析长程特性。最终得出几组实际的瓦斯时间序列数据的Hurst指数范围介于0.6301到0.9137，其有效关联长度范围为101.415到102.134。结果表明：工作面瓦斯时间序列具有长程正向趋势特性，而且未发生突出时的瓦斯时间序列数据正向长程相关性要强于已发生突出时的数据。最后，提出利用Bayesian推理框架下的LS-SVM回归模型对工作面瓦斯涌出量时间序列进行了多步预测。进一步地为了提高模型的鲁棒性，针对此BayesianLS-SVM对异常点的敏感而引入了BWLS-SVM。文中引入四个误差指标来评价所有给出模型的预测效果。由于在预测函数中自变量的更新值使用的是实测数据，所以所提方法均可较好地实现多步预测，而且预测误差也较小。为了进一步说明所提多步预测方法的效果，文中给出在Bayesian推理框架下LS-SVM模型的传统迭代预测效果。仿真结果表明传统迭代预测效果并没有文中所提方法好。作者从混沌时间序列的角度研究了工作面瓦斯涌出量的复杂特性，并在此基础上建立了时间序列预测模型。所研究内容为工作面非接触式煤与瓦斯突出动态预测奠定了一定的理论基础，同时也为深入分析工作面瓦斯涌出量的复杂特性开辟了新思路。更多还原

【Abstract】 China is one of the most serious countries in the world where the coal and gasoutburst often happens, with a series of problems such as many outburst wells, widelydistribution, shallow penetration outburst depth, high outburst frequency, hugeoutburst strength etc. However, the coal and gas outburst mechanism still hasn’t beenfully understood yet. No country can absolutely prevent the occurrence of outburst.Moreover, at presents, research and methodology on local test gas and getting rid ofdanger is gradually improved at home and abroad, there is a great difficulty in theprevention and control of coal and gas outburst. A large number of data indicate thataccidents which created particularly serious coal and gas outburst mostly take place inmine workface. So seeking a better non-contact dynamic continuous predictionmeasure is quite necessary.Firstly, the relationship between workface gas emission with coal seam gascontent, gas pressure, seam permeability, geological structure factors and other factorshave been proposed according to the flow theory of gas in coal seams. Further, therelated complex features of gas outburst with gas emission time-series be obtained,and it is as studying content in full text.Secondly, the dynamics simulation of gas emission time-series is analyzed byapplying Taken’s phase space reconstruction theory. The phase space reconstructionparameters are selected by using the differential entropy minimum principle. Severalembedding dimension of studying instance been ultimately identified are great than8.It indicates that the gas emission dynamics system is a high-dimensional complexsystem. On this basis, the chaotic characteristics of gas emission time-series data isqualitatively analyzed by using power spectrum and Poincare section. LargestLyapunov exponent solution based on nonlinear least squares regression andcorrelation dimension solution based on FCM are proposed in thesis because oflimited actual length of time-series and unknown noise. To further verifying thevalidity of the method, the algorithm has been carried out to test by using identifiedLogistic chaotic time-series and Hénon system. Then, the method is used todistinguish characteristics of the actual time-series, and determination of thecorrelation dimension also further validates the correctness of the embeddingdimension solution mentioned in front.Thirdly, Hurst exponent is appointed to analyze long-range correlation oftime-series in this thesis, and appointed V/S method to solve the long-range trend characteristics for shortcomings of the traditional R/S method. Further, the effectivecorrelation length is quoted to analyze the long-range features. The Hurst exponentrange of several groups of actual data is between0.6301-0.9137, and its effectivecorrelation length range is from101.415to102.134. These data suggest that workface gastime-series with positive long-range trend characteristics, and this characteristics ofno-outburst gas time-series data is significantly stronger than outburst gas time-seriesdata.Finally, the LS-SVM regression model within the Bayesian inference frameworkis adapted to multi-step predict workface gas emission time-series. Further, in order toimprove the robustness of the models, BWLS-SVM model is introduced for theBayesian LS-SVM is sensitive to outliers. In this paper four error indicators areintroduced to evaluate prediction effect of the given model. The updated value of theprediction function variables is used the measured data, so the mentioned methods canachieve better multi-step prediction; the prediction error is also smaller. To furthercompare the proposed multi-step prediction effect, the traditional iterative predictionby using LS-SVM model within Bayesian inference framework is given. The finalresults show that the traditional iterative prediction results are unsatisfactory.The author studies the complex nature of workface gas emission from the pointof view of the chaotic time-series, and on this basis to establish time-series forecastingmodels. The contents of this article has laid a theoretical basis for non-contact coaland gas outburst dynamic forecast, but also open up new ideas for in-depth analysis ofthe gas emission complex.更多还原