Multi-parameter prediction of goaf environment based on time-series optimized long short-term memory network
ZHANG Pengyu
MA Li
SHI Xinhui
WANG Shaorong
LI Zhao
Abstract：Spontaneous combustion of coal and harmful gas emissions in goaf areas affect mine safety production,making environmental prediction in these areas crucial for hazard identifica-tion.Environmental characteristic data from goaf areas were collected and analyzed to deter-mine the variation patterns of absolute pressure,temperature,O2 concentration,and CO con-centration,with correlations between parameters calculated using Pearson correlation coeffi-cients.Empirical Mode Decomposition(EMD)was applied for adaptive time-frequency decom-position of environmental characteristic sequences,while Kernel Principal Component Analysis(KPCA)was used to map the decomposed data into high-dimensional feature space.A Long Short-Term Memory(LSTM)neural network model with time-series optimization was devel-oped to predict multiple environmental parameters in the No.5 mining district goaf of a coal mine.The results show that environmental parameters in goaf exhibit multi-scale and non-sta-tionary temporal characteristics,with EMD decomposition revealing intrinsic features of pa-rameter sequences at different time scales.O2 concentration fluctuations show significant cor-relation with absolute pressure changes,while CO concentration and temperature sequences demonstrate strong nonlinearity and abrupt features.The dimensionality reduction through KPCA preserves 98.12％of the data information content while significantly reducing feature redundancy.The LSTM model after time series optimization has achieved significant improve-ment in prediction performance.Compared with the non-optimized models,the prediction ac-curacy of the optimized model has been greatly improved,and it has high generalization ability and robustness.This approach effectively mines coupling relationships between environmental parameters,thereby improving the accuracy of goaf environmental parameter prediction.
Keywords：coal spontaneous combustionlarge-area goaftime seriesfeature extractiondeep learning
Publication Date：2025-05-30
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：14( 653-666 )