中国安全科学学报 ›› 2020, Vol. 30 ›› Issue (5): 39-47.doi: 10.16265/j.cnki.issn1003-3033.2020.05.007

• 安全工程技术 • 上一篇    下一篇

基于KPCA-CMGANN算法的瓦斯涌出量预测研究

肖鹏1,2 副教授, 谢行俊1,2, 双海清1,2 讲师, 刘朝阳3, 王海宁3, 徐经苍3   

  1. 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 教育部西部矿井开采及灾害防治重点实验室,陕西 西安 710054;
    3.陕西陕煤澄合矿业有限公司,陕西 澄城 715200
  • 收稿日期:2020-02-05 修回日期:2020-04-15 出版日期:2020-05-28 发布日期:2021-01-28
  • 作者简介:肖鹏(1982—),男,陕西西安人,博士,副教授,主要从事采动覆岩裂隙演化与瓦斯储运及瓦斯高效抽采理论与技术方面的研究。E-mail:xiaopeng@xust.edu.cn。
  • 基金资助:
    国家自然科学基金资助(51774235,51904238);陕西省自然科学基础研究计划(2020JM-530,2019JQ-337);陕西省教育厅专项科学研究计划(19JK0534)。

Prediction of gas emission quantity based on KPCA-CMGANN algorithm

XIAO Peng1,2, XIE Xingjun1,2, SHUANG Haiqing1,2, LIU Chaoyang3, WANG Haining3, XU Jingcang3   

  1. 1. College of Safety Science and Engineering, Xi'an University of Science & Technology, Xi'an Shaanxi 710054, China;
    2. Key Laboratory of Western Mine Exploitation and Hazard Prevention, Ministry of Education, Xi'an University of Science & Technology, Xi'an Shaanxi 710054, China;
    3. Shaanxi Chenghe Mining Co., Ltd., Chengcheng Shaanxi 715200, China
  • Received:2020-02-05 Revised:2020-04-15 Online:2020-05-28 Published:2021-01-28

摘要: 为了精准预测瓦斯涌出量,针对绝对瓦斯涌出量非线性、时变性、复杂性等特点,提出采用核主成分分析法(KPCA)对影响因素进行降维处理;针对BP神经网络(BPNN)中存在的收敛速度慢和易陷入局部最优解的问题,采用压缩映射遗传算法(CMGA)优化BPNN;构建CMGA与BPNN的耦合算法(CMGANN),计算分析某低瓦斯矿井监测历史数据形成的样本集,建立KPCA-CMGANN预测模型;用KPCA-CMGANN预测模型和其他3种网络模型分别对煤矿现场数据进行预测。结果表明:KPCA-CMGANN预测模型在379个时间步长里达到收敛,4个回采工作面的瓦斯涌出量预测相对误差分别为0.58%、0.63%、0.57%和0.45%,平均相对误差仅为0.56%,预测精度和收敛速度均优于对比模型,可实现瓦斯涌出量的快速精准预测。

关键词: 瓦斯涌出量预测, 核主成分分析法(KPCA), 压缩映射遗传算法(CMGA), BP神经网络(BPNN), 样本集

Abstract: In order to accurately predict gas emission quantity, considering the nonlinearity, time-varying characteristic and complexity of absolute gas emission, KPCA was proposed to conduct dimensionality reduction for influencing factors. Secondly, targeting at problems of BPNNs' slow convergence and tendency to fall into local optimal solution, CMGA was adopted to optimize BPNN. Then, a coupling algorithm CMGANN based on CMGA and BPNN was constructed to calculate and analyze sample sets formed by historical data of a low gas mine, and KPCA-CMGANN prediction model was established which together with three other network models were used to predict coal mine field data. The results show that KPCA-CMGANN model achieves convergence in 379 time steps, and relative errors of gas emission prediction for four working faces are 0.58%, 0.63%, 0.57% and 0.45% with an average relative error at only 0.56%. Its prediction accuracy and convergence speed are superior to comparative model, making it ready to predict gas emission amount accurately and quickly.

Key words: predication of gas emission quantity, kernel principal component analysis (KPCA), compression mapping genetic algorithm (CMGA), back propagation neural network(BPNN), sample sets

中图分类号: