小波-极限学习机在瓦斯涌出量时变序列预测中的应用  

Application of wavelet-extreme learning machine in time-varying series prediction of gas emission quantity

作　　者：肖鹏 谢行俊[1,2] 双海清 刘朝阳 王海宁 徐经苍 马军红 XIAO Peng;XIE Xing-jun;SHUANG Hai-qing;LIU Chao-yang;WANG Hai-ning;XU Jing-cang;MA Jun-hong(College of Safety Science and Engineering,Xi’an University of Science & Technology,Xi’an 710054,China;Key Laboratory of Western Mine Exploitation and Hazard Prevention,Ministry of Education,Xi’an University of Science & Technology,Xi’an 710054,China;Shaanxi Chenghe Mining Co. ,Ltd. ,Weinan 714000,China)

机构地区：[1]西安科技大学安全科学与工程学院,陕西西安710054 [2]西安科技大学西部矿井开采及灾害防治教育部重点实验室,陕西西安710054 [3]陕西陕煤澄合矿业有限公司,陕西渭南714000

出　　处：《西安科技大学学报》2020年第5期839-845,共7页Journal of Xi’an University of Science and Technology

基　　金：国家自然科学基金项目(51774235,51904238);陕西省自然科学基础研究计划项目(2020JM-530,2019JQ-337);陕西省教育厅专项科学研究计划项目(19JK0534)。

摘　　要：针对煤矿瓦斯涌出量预测中经常出现变量难以获取等问题,为了提高瓦斯涌出量的预测精度和可靠性,提出将小波包分解方法与极限学习机相结合,构建瓦斯涌出量的小波-极限学习机时变预测模型。首先,通过小波包分解重构将瓦斯涌出量时变序列分解成高、低频率不同的分量,然后采用极限学习机对小波包分解重构后的时间序列进行预测,再叠加预测值,得到最终的预测结果。以山西天池煤矿某工作面瓦斯涌出量监测时序样本为例,为体现模型的优越性,设置2个对照模型,即小波-BP模型和未经小波处理的极限学习机模型。结果表明:该模型预测相对误差为0.42%~10.45%,平均相对误差仅为2.50%,小波-BP模型的预测相对误差为0.33%~7.33%,平均相对误差为3.42%,未经小波处理的极限学习机模型的预测相对误差为1.59%~13.09%,平均相对误差为4.25%,小波-极限学习机模型的预测精度和泛化能力均高于对照模型;小波包分解重构方法的引入能有效降低数据复杂度,大幅度提高预测精度,为瓦斯涌出量时变序列的预测提供了新的思路。Aiming at the difficulty of obtaining variables in the prediction of coal mine gas emission,in order to improve the prediction accuracy and reliability of gas emission,a wavelet packet decomposition method and an extreme learning machine were combined to construct a wavelet-extreme learning machine time-varying prediction model for gas emission.Firstly,the time-varying sequence of gas emission was decomposed into high and low frequency components by wavelet packet decomposition and reconstruction,and the extreme time machine was used to predict the time series with a wavelet packet decomposed and reconstructed,and then the predicted values were superimposed to obtain the final forecast results.The time series samples of gas emission monitoring of a working face in Tianchi coal mine of Shanxi province were used as an example.In order to reflect the superiority of the model,two contrast models were set up,namely a wavelet-BP model and an extreme learning machine model without wavelet processing.The results indicate that the relative prediction errors of this model was 0.42%to 10.45%,and the average relative error was only 2.50%.The relative prediction errors of the wavelet-BP model was 0.33%to 7.33%,and the average relative error was 3.42%.The prediction relative errors of the extreme learning machine model without wavelet processing was 1.59%to 13.09%,and the average relative error was 4.25%.The prediction accuracy and generalization ability of the wavelet-extreme learning machine model are higher than those of the contrast model.The introduction of wavelet packet decomposition and reconstruction method can effectively reduce the complexity of the data,greatly improve the prediction accuracy,and provide new ideas for the prediction of time-varying sequences of gas emission.

关 键 词：瓦斯涌出量预测 时变序列 小波包分解 极限学习机 预测精度 

分 类 号：TD712[矿业工程—矿井通风与安全]