【摘要】 在近距离煤层开采过程中，遗留煤柱下区段煤柱的变形影响巷道的稳定性。以大柳塔矿活鸡兔井22208工作面回采巷道为工程背景，采用分布式光纤传感技术对区段煤柱内部的水平变形进行表征。在此基础上，提出一种改进的自适应脉冲编码方法，将采集的应变数据转换为脉冲信号，作为脉冲神经网络(SNN)预测模型的输入。采用泄漏积分发放(LIF)神经元模型构建脉冲神经网络，并引入奖励调节的突触可塑性(R-STDP)学习机制进行训练和测试。结果表明:分布式光纤传感技术能够有效反映区段煤柱内部变形的真实状态;相较于传统神经网络模型，改进自适应脉冲编码的SNN模型在煤柱变形预测方面表现更优，其预测结果的均方根误差降低了0.81，平均绝对百分比误差减少了27%。研究为近距离煤层开采过程中的巷道稳定性评估与安全控制提供了重要的技术依据。更多还原

【Abstract】 In close-distance coal seam mining,the deformation of section coal pillars beneath remnant coal pillars compromises the stability of the roadway.The internal horizontal deformation of section coal pillars was characterized using distributed optical fiber sensing(DOFS) technology within the roadway of working face 22208,coal seam,Huojitu Mine,Daliuta Coal Mine.Building on this,an improved adaptive pulse encoding method was proposed to convert the acquired strain data into pulse signals,these signals served as critical input parameters for a spiking neural network prediction model.The SNN model was constructed using the leaky integrate-and-fire(LIF) neuron,and trained and tested by incorporating the reward-modulated spike timing-dependent plasticity(R-STDP) learning mechanism.The results demonstrate that: DOFS technology effectively reflects the true deformation state within section coal pillars.Compared with traditional neural network models,the SNN model utilizing the improved adaptive pulse encoding exhibits superior performance in predicting coal pillar deformation,achieving a reduction in root mean square error of 0.81 mm and a decrease in mean absolute percentage error by 27%.This research provides a robust technical basis for stability assessment and safety control in roadways during close-distance coal seam mining operations.更多还原