小波分析在瓦斯涌出量预测中的应用

发布时间：2018-03-15 10:57

本文选题：多分辨分析　切入点：小波包　出处：《西安科技大学》2013年硕士论文　论文类型：学位论文

【摘要】：煤矿瓦斯灾害是煤矿五大自然灾害之一，严重威胁煤矿的安全生产。瓦斯涌出量是研究瓦斯灾害的一个重要指标，，为防止瓦斯灾害事故的发生，对瓦斯涌出量进行预测尤为重要。从系统的观点来看，瓦斯涌出是一个复杂的非线性动力系统，其涌出量作为一种时间序列，依其数据的大小和顺序蕴含着大量有关系统动态演化过程的痕迹和特征，本文针对这一特点结合小波理论中的多分辨分析和小波包分析对矿井瓦斯涌出量进行了预测。主要内容和结论如下。通过小波多分辨分析把瓦斯涌出量这一非平稳时间序列分解为若干层近似意义上的平稳时间序列，再用AR模型对其单只重构序列建立模型（即建立了基于多分辨分析的预测模型），分析了不同小波基函数和同一小波基函数分解层数不同对预测效果的影响。经过仿真验证，基于多分辨分析的预测结果与直接用AR模型预测的结果相比较好。多分辨分析只能在固定的频率空间上分解时间轴，对于时间分辨率比较高的时间序列，可能会因为选取了比较低的频率尺度，导致某些在频率较高空间中反映瓦斯涌出系统状态特征的信息丢失。瓦斯涌出量常具有混沌特性且规律不易显现。因此，本文选用可以自适应选择频带的小波包变换对瓦斯涌出量的混沌时间序列进行分解和重构，在其混沌特性判别的基础上，改进了传统的小波包—混沌预测模型，对模型中小波包重构的每组序列的预测结果引入了权重，建立了加权小波包—混沌预测模型。仿真结果表明，此模型不但提高了预测精度还改善了预测误差的不稳定性，增加了可预测范围；鉴于最优小波包在分解层数一定的情况下重构结点数会尽可能少的优点，本文提出了用最优小波包进行分解的加权最优小波包—混沌预测模型，它不但保留了加权小波包—混沌预测模型的优点，还减小了计算量，经仿真验证，具有较好的预测效果。最后，本文用小波神经网络代替最优小波包—混沌预测模型中的加权一阶局域法，利用小波神经网络强大的非线性映射能力，建立了最优小波包—混沌—小波神经网络模型，对瓦斯涌出量进行预测。经过仿真实验和比较分析，最优小波包—混沌—小波神经网络模型预测精度较高，具有一定的推广和实用价值。
[Abstract]:Gas disaster in coal mine is one of the five natural disasters in coal mine, which seriously threatens the safe production of coal mine. The quantity of gas emission is an important index to study gas disaster, in order to prevent the occurrence of gas disaster accident, From the point of view of system, gas emission is a complex nonlinear dynamic system, and its emission is a time series. According to the size and order of the data, there are a lot of traces and characteristics about the dynamic evolution of the system. In this paper, combined with multi-resolution analysis and wavelet packet analysis in wavelet theory, the mine gas emission is predicted. The main contents and conclusions are as follows. The non-stationary time series of gas emission is decomposed into stationary time series in the sense of approximation by wavelet multi-resolution analysis. Then the AR model is established for its single reconstruction sequence (that is, a prediction model based on multi-resolution analysis is established, and the influence of different wavelet basis functions and different decomposition layers of the same wavelet basis function on the prediction results is analyzed. The prediction result based on Multiresolution analysis is better than that with AR model directly. Multiresolution analysis can only decompose the time axis in a fixed frequency space. For a time series with a higher time resolution, it may be possible to select a lower frequency scale. This results in the loss of some information which reflects the state characteristics of the gas emission system in the higher frequency space. The gas emission is often characterized by chaos and the law is not easy to appear. In this paper, the wavelet packet transform, which can adaptively select the frequency band, is used to decompose and reconstruct the chaotic time series of gas emission. On the basis of discriminating its chaotic characteristics, the traditional wavelet packet-chaos prediction model is improved. The weight is introduced into the prediction results of each set of sequences reconstructed by wavelet packets, and a weighted wavelet packet-chaotic prediction model is established. The simulation results show that the model not only improves the prediction accuracy but also improves the instability of prediction errors. In view of the advantage that the number of nodes reconstructed by optimal wavelet packets is as small as possible when the number of decomposition layers is constant, a weighted optimal wavelet packet-chaotic prediction model is proposed. It not only retains the advantages of the weighted wavelet packet-chaotic prediction model, but also reduces the computational complexity. The simulation results show that it has a good prediction effect. Finally, in this paper, wavelet neural network is used to replace the weighted first-order local method in the optimal wavelet packet-chaos prediction model, and the optimal wavelet packet-chaos wavelet neural network model is established by using the powerful nonlinear mapping ability of the wavelet neural network. The simulation experiment and comparative analysis show that the prediction accuracy of the optimal wavelet packet chaotic wavelet neural network model is high and has certain popularization and practical value.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TD712.5

【参考文献】