Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memory

Coal main adit is susceptibility to external factors, so it is crucial to monitor and predict its deformation. Based on the application of fiber optic monitoring for adit deformation, A multi-step prediction model is proposed which is built with Ensemble Empirical Mode Decomposition (EEMD) integrate...

Full description

Saved in:
Bibliographic Details
Main Authors: Wenli JI, Xin DAN, Chenyang MA, Jing CHAI, Yuyi WU, Fengqi QIU, Wentao LIU, Wulin LEI, Yongliang LIU
Format: Article
Language:zho
Published: Editorial Department of Coal Science and Technology 2025-04-01
Series:Meitan kexue jishu
Subjects:
Online Access:http://www.mtkxjs.com.cn/article/doi/10.12438/cst.2023-1480
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849326637995261952
author Wenli JI
Xin DAN
Chenyang MA
Jing CHAI
Yuyi WU
Fengqi QIU
Wentao LIU
Wulin LEI
Yongliang LIU
author_facet Wenli JI
Xin DAN
Chenyang MA
Jing CHAI
Yuyi WU
Fengqi QIU
Wentao LIU
Wulin LEI
Yongliang LIU
author_sort Wenli JI
collection DOAJ
description Coal main adit is susceptibility to external factors, so it is crucial to monitor and predict its deformation. Based on the application of fiber optic monitoring for adit deformation, A multi-step prediction model is proposed which is built with Ensemble Empirical Mode Decomposition (EEMD) integrated with TCN-LSTM deep learning network for adit deformation. Firstly, the monitoring data containing noise is decomposed into several Intrinsic Mode Functions (IMF) components using the EEMD method. Then, the fuzzy entropy of each IMF component series is calculated, and effective IMF components are selected. Finally, the TCN network is used to extract long-term features from different effective component series, while the LSTM network captures nonlinear features. The prediction results of each component are combined. The multi-output strategy is adopted in the training process of the prediction model, and the output is the fiber optic monitoring value for multiple times in the future. Experimental results on different fiber optic grating sensors show that the EEMD combined with fuzzy entropy method can filter out more noise while retaining the roadway deformation information. Compared with existing methods, the proposed prediction method has a coefficient of determination (R2) of 0.99, and the root mean square error (RMSE) and mean absolute error (MAE) are reduced by 3.0%−10.0% and 5.0%−20.0% in single-step prediction, respectively, resulting in more accurate predictions. Under the multi-output strategy, the average R2 of this proposed method for three steps ahead is 0.95, and the RMSE and MAE values of the strain gauge are reduced by at least 75.0% and 31.5%. The RMSE and MAE values of the displacement meter were reduced by at least 50.0% and 66.7%, respectively, while the RMSE and MAE values of the pressure gauge were reduced by at least 85.7% and 62.3%. the proposed prediction method with multi-output strategy has the lowest error accumulation. The EEMD-TCN-LSTM multi-step prediction method for adit deformation provides a technical basis for predicting the deformation of roadway surrounding rock.
format Article
id doaj-art-cf997f81c83b4bd389bccb7638ed8331
institution Kabale University
issn 0253-2336
language zho
publishDate 2025-04-01
publisher Editorial Department of Coal Science and Technology
record_format Article
series Meitan kexue jishu
spelling doaj-art-cf997f81c83b4bd389bccb7638ed83312025-08-20T03:48:06ZzhoEditorial Department of Coal Science and TechnologyMeitan kexue jishu0253-23362025-04-0153417619010.12438/cst.2023-14802023-1480Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memoryWenli JI0Xin DAN1Chenyang MA2Jing CHAI3Yuyi WU4Fengqi QIU5Wentao LIU6Wulin LEI7Yongliang LIU8College School of Communication and Information Engineering, Xi’an University of Science and Technology, Xi’an 710054, ChinaCollege School of Communication and Information Engineering, Xi’an University of Science and Technology, Xi’an 710054, ChinaCollege of Energy Engineering, Xi’an University of Science and Technology, Xi’an 710054, ChinaCollege of Energy Engineering, Xi’an University of Science and Technology, Xi’an 710054, ChinaChina Coal Energy Research Institute Co., Ltd., Xi’an 710000China Coal Energy Research Institute Co., Ltd., Xi’an 710000China Coal Energy Research Institute Co., Ltd., Xi’an 710000School of Energy Engineering, Longdong University, Qingyang 745000, ChinaKey Laboratory of Western Mine Exploitation and Hazard Prevention, Ministry of Education, Xi’an University of Science and Technology, Xi’an 710054, ChinaCoal main adit is susceptibility to external factors, so it is crucial to monitor and predict its deformation. Based on the application of fiber optic monitoring for adit deformation, A multi-step prediction model is proposed which is built with Ensemble Empirical Mode Decomposition (EEMD) integrated with TCN-LSTM deep learning network for adit deformation. Firstly, the monitoring data containing noise is decomposed into several Intrinsic Mode Functions (IMF) components using the EEMD method. Then, the fuzzy entropy of each IMF component series is calculated, and effective IMF components are selected. Finally, the TCN network is used to extract long-term features from different effective component series, while the LSTM network captures nonlinear features. The prediction results of each component are combined. The multi-output strategy is adopted in the training process of the prediction model, and the output is the fiber optic monitoring value for multiple times in the future. Experimental results on different fiber optic grating sensors show that the EEMD combined with fuzzy entropy method can filter out more noise while retaining the roadway deformation information. Compared with existing methods, the proposed prediction method has a coefficient of determination (R2) of 0.99, and the root mean square error (RMSE) and mean absolute error (MAE) are reduced by 3.0%−10.0% and 5.0%−20.0% in single-step prediction, respectively, resulting in more accurate predictions. Under the multi-output strategy, the average R2 of this proposed method for three steps ahead is 0.95, and the RMSE and MAE values of the strain gauge are reduced by at least 75.0% and 31.5%. The RMSE and MAE values of the displacement meter were reduced by at least 50.0% and 66.7%, respectively, while the RMSE and MAE values of the pressure gauge were reduced by at least 85.7% and 62.3%. the proposed prediction method with multi-output strategy has the lowest error accumulation. The EEMD-TCN-LSTM multi-step prediction method for adit deformation provides a technical basis for predicting the deformation of roadway surrounding rock.http://www.mtkxjs.com.cn/article/doi/10.12438/cst.2023-1480adit deformationmulti-step predictiontcn-lstm prediction modelensemble empirical mode decompositioncoal mine intelligence
spellingShingle Wenli JI
Xin DAN
Chenyang MA
Jing CHAI
Yuyi WU
Fengqi QIU
Wentao LIU
Wulin LEI
Yongliang LIU
Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memory
Meitan kexue jishu
adit deformation
multi-step prediction
tcn-lstm prediction model
ensemble empirical mode decomposition
coal mine intelligence
title Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memory
title_full Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memory
title_fullStr Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memory
title_full_unstemmed Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memory
title_short Multi-step prediction of coal mine adit deformation based on time convolutional long short-term memory
title_sort multi step prediction of coal mine adit deformation based on time convolutional long short term memory
topic adit deformation
multi-step prediction
tcn-lstm prediction model
ensemble empirical mode decomposition
coal mine intelligence
url http://www.mtkxjs.com.cn/article/doi/10.12438/cst.2023-1480
work_keys_str_mv AT wenliji multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT xindan multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT chenyangma multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT jingchai multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT yuyiwu multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT fengqiqiu multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT wentaoliu multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT wulinlei multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory
AT yongliangliu multisteppredictionofcoalmineaditdeformationbasedontimeconvolutionallongshorttermmemory