基于频率分解与LSTM网络模型的隧道施工期瓦斯浓度预测  

Prediction of Gas Concentration During Tunnel Construction Based on Frequency Decomposition and LSTM Network Model

作　　者：张丹锋 ZHANG Danfeng(Qinghai Communications Construction Management Co.,Ltd.,Xi′ning 810001,China)

机构地区：[1]青海省交通建设管理有限公司,西宁810001

出　　处：《公路交通技术》2023年第5期169-176,共8页Technology of Highway and Transport

摘　　要：针对隧道现场瓦斯数据具有非线性、非平稳性、复杂性特点,提出一种基于互补集合经验模态分解(CEEMD)与长短期记忆网络(LSTM)组合模型的隧道瓦斯浓度预测方法。该方法采用CEEMD将瓦斯浓度数据分解为具有不同特征的子序列分量IMF,通过对分解后的各子序列分别建立LSTM网络模型进行单步预测,并进一步叠加各子序列预测结果得到隧道瓦斯浓度最终预测值。结果表明:CEEMD-LSTM的均方根误差(RMSE)和平均绝对误差(MAE)值分别为0.023 4和0.017 3,相较多层感知器(MLP)、支持向量回归(SVR)、门控循环单元(GRU)、长短期记忆网络(LSTM)等预测模型具有更高的准确性;该方法可为精确预测开挖隧道内瓦斯浓度提供理论依据,可供预防隧道瓦斯事故发生和保障隧道安全施工参考。A tunnel gas concentration prediction method based on the combination model of complementary set empirical mode decomposition(CEEMD) and short-term memory network(LSTM) is proposed to address the nonlinear,non-stationary,and complex characteristics of gas data in tunnel sites..The method first uses CEEMD to decompose the gas concentration data into sub-series components IMF with different characteristics.Then,the LSTM network model is established for single-step prediction of each subsequence after decomposition.Finally,the final prediction value of tunnel gas concentration is obtained by superimposing the prediction results of each subsequence.The results show that the root mean square error(RMSE) value of CEEMD-LSTM is 0.023 4 and the mean absolute error(MAE) value is 0.017 3.Compared with the basic prediction models such as multilayer perceptron(MLP),support vector regression(SVR),gated recurrent unit(GRU),long and short-term memory network(LSTM),this method demonstrates higher accuracy.It can serve as a theoretical basis for accurately predicting gas concentration inside excavated tunnels,providing reference for preventing tunnel gas accidents and ensuring tunnel construction safety.

关 键 词：瓦斯隧道 互补集合经验模态分解 长短期记忆网路 瓦斯预测 

分 类 号：U455[建筑科学—桥梁与隧道工程]