瓦斯时间序列混沌特性分析及预测研究

发布时间：2019-01-01 13:17

【摘要】：瓦斯灾害是威胁我国煤矿安全生产的重要因素之一。从煤矿实际监测数据中挖掘瓦斯涌出规律，实现瓦斯的准确、可靠预测，能够有效避免灾害的发生。瓦斯涌出系统受多种因素影响，且随机干扰成分很多，因此通过分析其影响因素建立预测模型的方法不易实现。针对以上问题，本文尝试从瓦斯历史数据中挖掘规律，建立预测模型，通过对一维时间序列的相空间重构还原系统空间，从而避免了预测建模的经验性和主观性。本文的主要工作有：首先，以某煤矿1024工作面的实际瓦斯采样数据为研究对象，通过最大Lyapunov指数法和功率谱法验证了瓦斯涌出量时间序列的混沌特性。利用经验模态分解方法对瓦斯涌出量时间序列进行了预处理，其中的高频分量作为噪声滤除，其余分量作为有用信息，分别建立预测模型，将各分量的预测结果叠加作为最终预测结果。然后，在分析前馈神经网络和反馈神经网络建模特点的基础上，根据瓦斯时间序列的动态特性，提出了基于回声状态网络的建模方案，并针对随着预测步长的增加，，固定相空间无法满足预测要求的特点，提出了基于最大互信息的变相空间重构方法：以互信息函数作为个体适应度函数，计算不同m和取值组合下的个体适应度值，选择最有利于提供信息的空间结构，并通过实验验证了该方法的有效性。最后，建立了基于混沌分析，经验模态分解及回声状态网络的瓦斯涌出量单步预测模型、直接多步预测模型及迭代多步预测模型，并将本文所建回声状态网络单步预测模型和同等条件下的BP、ELMAN预测模型作对比分析，验证了回声状态网络模型的优势。将直接多步预测模型和迭代多步预测模型作对比分析，验证了直接多步预测模型的优势。
[Abstract]:Gas disaster is one of the important factors threatening the safety of coal mine production in China. Mining the law of gas emission from the actual monitoring data of coal mine, realizing the accurate and reliable prediction of gas, can effectively avoid the occurrence of disaster. The gas emission system is affected by many factors and has a lot of random interference components. Therefore, it is difficult to establish a prediction model by analyzing the influencing factors. Aiming at the above problems, this paper tries to excavate the laws from the gas historical data and establish the prediction model. By reconstructing the system space of the one-dimensional time series in the phase space, the experience and subjectivity of the prediction modeling are avoided. The main work of this paper is as follows: firstly, taking the actual gas sampling data of a coal mine 1024 face as the research object, the chaotic characteristics of the time series of gas emission quantity are verified by the maximum Lyapunov exponent method and the power spectrum method. The time series of gas emission are pretreated by the empirical mode decomposition method. The high frequency component is used as noise filter and the other components are used as useful information to establish the prediction model. The prediction results of each component are superimposed as the final prediction results. Then, on the basis of analyzing the modeling characteristics of feedforward neural network and feedback neural network, according to the dynamic characteristics of gas time series, a modeling scheme based on echo state network is proposed. The fixed phase space can not meet the requirement of prediction. A new method for reconstruction of variable phase space based on maximum mutual information is proposed. The mutual information function is used as the individual fitness function to calculate the individual fitness values under different combinations of m and values. The spatial structure which is most favorable for providing information is selected, and the effectiveness of the method is verified by experiments. Finally, based on chaos analysis, empirical mode decomposition and echo state network, single-step prediction model of gas emission, direct multi-step prediction model and iterative multi-step prediction model are established. The one-step prediction model of echo state network is compared with the BP,ELMAN prediction model under the same conditions, and the advantages of echo state network model are verified. The direct multi-step prediction model is compared with the iterative multi-step prediction model, and the advantages of the direct multi-step prediction model are verified.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TD712.5

【相似文献】