
Applied Sciences, Год журнала: 2025, Номер 15(8), С. 4192 - 4192
Опубликована: Апрель 10, 2025
Coal is the main energy source in China, but coal mining a high-risk industry, making prevention and control of hazards an important topic. Constrained by complexity unpredictability underground spaces, current research on disaster technologies mainly focuses characteristics overlying strata laws mine pressure, resulting significant deficiencies accuracy. Given this, data-driven pressure prediction method proposed, which uses deep learning models to learn patterns existing data generate required predictions. This approach avoids challenges accurately extracting rock mass physical mechanical parameters geological structure modeling, thereby improving accuracy control. The stage working face exertion period prone disasters during mining. To achieve accurate task divided into three steps: first step predict support resistance ahead face, second classify labels coordinate units, third characteristic exertion. Deep were designed trained separately for each For step, Spatiotemporal sequence model was selected, achieved mean absolute error 4.65 kN prediction. image segmentation-based classification chosen, with reaching 97.77%. fusion consisting LSTM (Long Short-Term Memory) networks designed. 0.17 dynamic coefficient, maximum 810.93 period, 9.96 cycles duration, 92.35% type. Simulating actual situation application scenarios, input steps set as output from previous evaluated. 1035.21 82.90% units. In simulated scenario, there 9922 instances exertion, predicted 10,336 instances, 9046 them matching instances. evaluated 4946 included complete cycles. coefficient 0.21, 1218.31 kN, duration cycle 11.03 cycles, type 91.75%.
Язык: Английский