Data-driven prediction framework of surrounding rock pressure in a fully mechanized coal face with temporal-spatial...

Area of Science:

• Mining Engineering
• Geotechnical Engineering
• Data Science

Background:

• Accurate prediction of surrounding rock pressure is crucial for roof management in fully mechanized coal faces (FMCF).
• Existing methods struggle to integrate diverse factors influencing rock pressure with high spatio-temporal correlation and heterogeneity.
• Hydraulic support load serves as a key indicator for monitoring surrounding rock pressure changes.

Purpose of the Study:

• To propose a novel data-driven prediction framework for surrounding rock pressure in FMCFs.
• To develop a system that effectively combines multidimensional data, temporal-spatial features, and adaptive strategies for accurate prediction.
• To enhance the reliability and applicability of surrounding rock pressure prediction models.

Main Methods:

• Development of a multidimensional working condition matrix (MWCM) module that adapts to mining processes and data quality.
• Implementation of a hydraulic support group temporal-spatial feature fusion (HTSFF) network to capture spatial correlations and temporal periodicity.
• Integration of an adaptive deployment strategy (ADS) to handle missing and abnormal data during model deployment.

Main Results:

• The proposed framework demonstrated high accuracy in predicting surrounding rock pressure, with an average error of 1.2406 MPa.
• Periodic pressure prediction achieved accuracy, precision, and recall rates exceeding 95%.
• The framework effectively addresses data challenges like missing values and abnormal data interference.

Conclusions:

• The developed data-driven framework provides an effective solution for surrounding rock pressure prediction in FMCFs.
• This approach significantly aids in FMCF roof management and control.
• Further fine-tuning may be necessary for broader applicability across diverse geological conditions and sensor qualities.