Two case studies were conducted in the Shennan mining area of Shaanxi Province,China to evaluate the surrounding rock deformation and stress evolution in pre-driven longwall recovery rooms· These studies mainly m...Two case studies were conducted in the Shennan mining area of Shaanxi Province,China to evaluate the surrounding rock deformation and stress evolution in pre-driven longwall recovery rooms· These studies mainly monitored the surrounding rock deformation and coal pillar stress in the recovery rooms of the N1206 panel of 2-2 coal seam at Ningtiaota Coal Mine and the 15205 panel of 5-2 coal seam at Hongliulin Coal Mine.The monitoring results showed that the surrounding rock deformation of the main recovery room and the coal pillar stress in the N1206 and 15205 panels began to increase significantly when the face was 36 m and 42 m away from the terminal line,respectively.After the face entered the main recovery room,the maximum roof-to-floor convergence in the N1206 and 15205 panels was 348.03 mm and 771.24 mm,respectively,and the coal pillar stresses increased more than 5 MPa and 7 MPa,respectively.In addition,analysis of the periodic weighting data showed that the main roof break position of the N1206 and 15205 panels after the longwall face entered the main recovery room was-3.8 m and-8.2 m,respectively.This research shows that when the main roof breaks above the coal pillar,the surrounding rock deformation of the main recovery room and the coal pillar stress increase sharply.The last weighting is the key factor affecting the stability of the main recovery room and the coal pillar;main roof breaks at disadvantageous positions are the main cause of the support crushing accidents.展开更多
面向推荐的对话生成任务旨在通过人机对话交互获取用户偏好,以实现精准推荐。针对现有研究工作存在对话推荐类型单一和生成回复质量低的问题,本文提出一种基于统一预训练语言模型(Unified Language Model pre-training,UniLM)的目标驱...面向推荐的对话生成任务旨在通过人机对话交互获取用户偏好,以实现精准推荐。针对现有研究工作存在对话推荐类型单一和生成回复质量低的问题,本文提出一种基于统一预训练语言模型(Unified Language Model pre-training,UniLM)的目标驱动的推荐对话生成模型(Goal Driven Recommendation-oriented Dialog Generation model, GDRDG)。该模型包括文本表示模块、多头编码模块、解码模块以及一种特殊的注意力掩码机制。其中,文本表示模块通过UniLM对输入文本进行向量化表示,确保模型能捕获文本的深层次语义特征;多头编码模块利用多头自注意力机制捕捉全局上下文信息,提高生成回复的连贯性和相关性;解码模块生成当前轮对话目标及基于该目标的回复,确保回复符合上下文并将对话向预期目标引导;特殊的注意力掩码机制则通过控制解码过程中的信息流,确保模型仅关注当前轮次相关信息,以提高回复质量。实验结果表明,GDRDG模型在BLEU、Distinct、F1和Hit@1等指标上均优于现有方法,验证了模型的有效性和先进性。展开更多
基金Support for this work was provided by the National Natural Science Foundation of China(No.51679199)Key Laboratory for Science and Technology Co-ordination and Innovation Projects of Shaanxi Province(No.2014SZS15-Z01)and is thankfully acknowledged by the authors.
文摘Two case studies were conducted in the Shennan mining area of Shaanxi Province,China to evaluate the surrounding rock deformation and stress evolution in pre-driven longwall recovery rooms· These studies mainly monitored the surrounding rock deformation and coal pillar stress in the recovery rooms of the N1206 panel of 2-2 coal seam at Ningtiaota Coal Mine and the 15205 panel of 5-2 coal seam at Hongliulin Coal Mine.The monitoring results showed that the surrounding rock deformation of the main recovery room and the coal pillar stress in the N1206 and 15205 panels began to increase significantly when the face was 36 m and 42 m away from the terminal line,respectively.After the face entered the main recovery room,the maximum roof-to-floor convergence in the N1206 and 15205 panels was 348.03 mm and 771.24 mm,respectively,and the coal pillar stresses increased more than 5 MPa and 7 MPa,respectively.In addition,analysis of the periodic weighting data showed that the main roof break position of the N1206 and 15205 panels after the longwall face entered the main recovery room was-3.8 m and-8.2 m,respectively.This research shows that when the main roof breaks above the coal pillar,the surrounding rock deformation of the main recovery room and the coal pillar stress increase sharply.The last weighting is the key factor affecting the stability of the main recovery room and the coal pillar;main roof breaks at disadvantageous positions are the main cause of the support crushing accidents.
文摘面向推荐的对话生成任务旨在通过人机对话交互获取用户偏好,以实现精准推荐。针对现有研究工作存在对话推荐类型单一和生成回复质量低的问题,本文提出一种基于统一预训练语言模型(Unified Language Model pre-training,UniLM)的目标驱动的推荐对话生成模型(Goal Driven Recommendation-oriented Dialog Generation model, GDRDG)。该模型包括文本表示模块、多头编码模块、解码模块以及一种特殊的注意力掩码机制。其中,文本表示模块通过UniLM对输入文本进行向量化表示,确保模型能捕获文本的深层次语义特征;多头编码模块利用多头自注意力机制捕捉全局上下文信息,提高生成回复的连贯性和相关性;解码模块生成当前轮对话目标及基于该目标的回复,确保回复符合上下文并将对话向预期目标引导;特殊的注意力掩码机制则通过控制解码过程中的信息流,确保模型仅关注当前轮次相关信息,以提高回复质量。实验结果表明,GDRDG模型在BLEU、Distinct、F1和Hit@1等指标上均优于现有方法,验证了模型的有效性和先进性。
