【摘要】 煤矿顶板事故是矿井生产的重大安全隐患，实现顶板压力的准确和多步长预测对于煤层安全智能开采具有重大意义。以往的顶板压力预测大多预测支架下一时刻的压力数据，注重提高单步长预测准确率，而缺乏对于多步长预测的研究。基于顶板压力多步长预测研究，提出了一种基于改进Informer网络的预测顶板压力的新方法，并采用Savitzky-Golay滤波对原始压力进行数据分析，得出了该方法在预测未来多个时刻的顶板压力时具有较好的效果。选用了山东省枣庄市付村煤矿某工作面的压力数据进行试验，结果表明：在预测第24个时间点时，相比于LSTM和GRU网络，改进Informer网络的顶板压力多步长预测模型RMSE分别减少了70.18%和38.31%,MAE分别减少了73.94%和40.50%。更多还原

【Abstract】 Coal mine roof accident is a major safety hazard in mine production. It is of great significance to realize the accurate and multi-step prediction of roof pressure for safe and intelligent mining of coal seams. Previous roof pressure predictions mostly predicted the pressure data of the next moment of the support, focused on improving the accuracy of single-step prediction, and lacked the research on multi-step prediction. Based on the multi-step prediction of roof pressure, a new method for predicting roof pressure based on improved Informer network was proposed, and Savitzky-Golay filter was used to analyze the original pressure. This method has a good effect in predicting roof pressure at multiple moments in the future. The pressure data of a working face in Fucun Coal Mine, Zaozhuang City, Shandong Province were selected for testing. The results show that when predicting the 24th time point, compared with LSTM and GRU network, the multi-step prediction model of roof pressure based on improved Informer network, RMSE is reduced by 70.18% and 38.31% respectively, and MAE is reduced by 73.94% and 40.50% respectively.更多还原