【作者】 李洪；

【导师】 蒋金泉；

【作者基本信息】 山东科技大学 ， 采矿工程， 2006， 博士

【摘要】 本文以冲击煤层的大量实测信息为基础，将混沌理论、小波理论、神经网络、模式识别等非线性学科的相关理论应用到冲击矿压危险性分析及预测识别领域中，提出了冲击矿压观测序列的混沌预测模型及模式识别方法，主要内容如下： (1) 为了提高冲击矿压预测和识别的准确性，利用小波变换对一维时间序列良好的去噪能力和奇异值分解对矩阵数据无损的特点，提出了小波和奇异值分解相结合的去噪方法。 (2) 基于大量实测信息，通过提取冲击矿压观测序列的关联维和最大Lyapunov指数来反演识别系统的混沌性态，深入研究了冲击运动孕育及变形破坏过程中的混沌特性及其变化规律。研究结果表明，冲击矿压观测数据序列中存在着混沌成分，从而明确了根据冲击矿压观测数据序列进行混沌预测和模式识别的可行性。 (3) 基于混沌分析的成果，构建了基于一阶局域近似、Lyapunov指数以及神经网络的冲击矿压观测序列的混沌预测模型，经实例分析验证，预测效果良好。 (4) 在混沌分析及混沌预测的基础上，运用模式识别理论，对冲击矿压观测序列计算了包括时域、频域、小波域及混沌域在内的数学特征值，采用欧氏距离测度的相似度量准则对这些特征值进行筛选和比较，选择和提取了最能反映原始量测数据本质和能有效识别冲击矿压危险的特征值组成模式识别的特征值空间，选择类内类间距离最小作为类别可分离性的判据，以Fisher函数和神经网络建立模式识别的Fisher准则识别器和径向基函数概率神经网络识别器，从而实现了观测序列的冲击矿压危险性预测和识别。更多还原

【Abstract】 In this paper, based on lots of measured information, the nonlinear science such as chaos theory, wavelet analysis theory, neural network, and pattern recognition is applied to the analysis and prediction of rock burst, and chaotic prediction models and pattern recognition method of rock burst are proposed in this dissertation. The main contents are as follows.(l)For having prediction and recognition more accurate, a united method which uses wavelet transform and strange value decomposition eliminate noise is proposed. The method takes advantage of capability which wavelet eliminate noise of one dimension time series and feature of non-damage to data which strange value decomposition eliminates noise.(2)based on lots of measured information, chaotic character of observation scries to rock burst can be offered(made) by back-analysis of correlation dimension and Lyapunov exponent extracted from these series to have character and rule of rock burst can be researched deeply in the processing of its evolution. The results show that rock burst observation series is a chaotic series and that chaotic prediction and pattern recognition is feasible.(3)based on the results of chaos analysis, the chaotic prediction mode of rock burst observation data series based to one-order approximation model, Lyapunov exponent, and neural network. The prediction effect is good by analysis of practical example.(4)based on the results of chaotic analysis and prediction, using theory of pattern recognition, math feature values of observation series to rock burst including time domain, frequency domain, wavelet domain, and chaos domain are calculated. By comparison and selection to these feature values using the similarity measure criterion based to Euclidean distance measurement, the feature values mapping essence of observation series and being able to recognize更多还原