Into the study of quasi-relaxation, in the past researches it has been concluded that the condition of meta-stability in the metallic specimen is given by the plasticity explained by the plastic energy in the process ...Into the study of quasi-relaxation, in the past researches it has been concluded that the condition of meta-stability in the metallic specimen is given by the plasticity explained by the plastic energy in the process of the quasi-relaxation. It is calculated through quasi-relaxation functional of this energy to obtain a spectra in the space D(σ – ε;t), that induces the existence of functions φ(t), and Ψ(t), related with the fundamental curves of quasi-relaxation given by σ(t), with their poles in , which is got in the maximum of stress given by σ0 = σ1. Also the tensor of plastic deformation that represents the plastic load during the application of specimen machine, cannot be obtained without poles in the space D(σ;t), corresponding the curves calculated into the space D(σ – ε;t), by curves that in the kinetic process of quasi-relaxation are represented by experimental curves in coordinates log σ – t. This situation cannot be eluded, since in this phenomena exist dislocations that go conform fatigue in the nano-crystalline structure of metals. From this point of view, is necessary to obtain a spectral study related to the energy using functions that permits the modeling and compute the states of quasi-relaxation included in the poles in the deformation problem to complete the solutions in the space D(σ – ε;t), and try a new method of solution of the differential equations of the quasi-relaxation analysis. In a nearly future development, the information obtained by this spectral study (by our integral transforms), will be able to give place to the programming through the spectral encoding of the materials in the meta-stability state, which is propitious to a nano-technological transformation of materials, concrete case, some metals.展开更多
针对地图综合中建筑多边形化简方法依赖人工规则、自动化程度低且难以利用已有化简成果的问题,本文提出了一种基于Transformer机制的建筑多边形化简模型。该模型首先把建筑多边形映射至一定范围的网格空间,将建筑多边形的坐标串表达为...针对地图综合中建筑多边形化简方法依赖人工规则、自动化程度低且难以利用已有化简成果的问题,本文提出了一种基于Transformer机制的建筑多边形化简模型。该模型首先把建筑多边形映射至一定范围的网格空间,将建筑多边形的坐标串表达为网格序列,从而获取建筑多边形化简前后的Token序列,构建出建筑多边形化简样本对数据;随后采用Transformer架构建立模型,基于样本数据利用模型的掩码自注意力机制学习点序列之间的依赖关系,最终逐点生成新的简化多边形,从而实现建筑多边形的化简。在训练过程中,模型使用结构化的样本数据,设计了忽略特定索引的交叉熵损失函数以提升化简质量。试验设计包括主试验与泛化验证两部分。主试验基于洛杉矶1∶2000建筑数据集,分别采用0.2、0.3和0.5 mm 3种网格尺寸对多边形进行编码,实现了目标比例尺为1∶5000与1∶10000的化简。试验结果表明,在0.3 mm的网格尺寸下模型性能最优,验证集上的化简结果与人工标注的一致率超过92.0%,且针对北京部分区域的建筑多边形数据的泛化试验验证了模型的迁移能力;与LSTM模型的对比分析显示,在参数规模相近的条件下,LSTM模型无法形成有效收敛,并生成可用结果。本文证实了Transformer在处理空间几何序列任务中的潜力,且能够有效复用已有化简样本,为智能建筑多边形化简提供了具有工程实用价值的途径。展开更多
文摘Into the study of quasi-relaxation, in the past researches it has been concluded that the condition of meta-stability in the metallic specimen is given by the plasticity explained by the plastic energy in the process of the quasi-relaxation. It is calculated through quasi-relaxation functional of this energy to obtain a spectra in the space D(σ – ε;t), that induces the existence of functions φ(t), and Ψ(t), related with the fundamental curves of quasi-relaxation given by σ(t), with their poles in , which is got in the maximum of stress given by σ0 = σ1. Also the tensor of plastic deformation that represents the plastic load during the application of specimen machine, cannot be obtained without poles in the space D(σ;t), corresponding the curves calculated into the space D(σ – ε;t), by curves that in the kinetic process of quasi-relaxation are represented by experimental curves in coordinates log σ – t. This situation cannot be eluded, since in this phenomena exist dislocations that go conform fatigue in the nano-crystalline structure of metals. From this point of view, is necessary to obtain a spectral study related to the energy using functions that permits the modeling and compute the states of quasi-relaxation included in the poles in the deformation problem to complete the solutions in the space D(σ – ε;t), and try a new method of solution of the differential equations of the quasi-relaxation analysis. In a nearly future development, the information obtained by this spectral study (by our integral transforms), will be able to give place to the programming through the spectral encoding of the materials in the meta-stability state, which is propitious to a nano-technological transformation of materials, concrete case, some metals.
文摘针对地图综合中建筑多边形化简方法依赖人工规则、自动化程度低且难以利用已有化简成果的问题,本文提出了一种基于Transformer机制的建筑多边形化简模型。该模型首先把建筑多边形映射至一定范围的网格空间,将建筑多边形的坐标串表达为网格序列,从而获取建筑多边形化简前后的Token序列,构建出建筑多边形化简样本对数据;随后采用Transformer架构建立模型,基于样本数据利用模型的掩码自注意力机制学习点序列之间的依赖关系,最终逐点生成新的简化多边形,从而实现建筑多边形的化简。在训练过程中,模型使用结构化的样本数据,设计了忽略特定索引的交叉熵损失函数以提升化简质量。试验设计包括主试验与泛化验证两部分。主试验基于洛杉矶1∶2000建筑数据集,分别采用0.2、0.3和0.5 mm 3种网格尺寸对多边形进行编码,实现了目标比例尺为1∶5000与1∶10000的化简。试验结果表明,在0.3 mm的网格尺寸下模型性能最优,验证集上的化简结果与人工标注的一致率超过92.0%,且针对北京部分区域的建筑多边形数据的泛化试验验证了模型的迁移能力;与LSTM模型的对比分析显示,在参数规模相近的条件下,LSTM模型无法形成有效收敛,并生成可用结果。本文证实了Transformer在处理空间几何序列任务中的潜力,且能够有效复用已有化简样本,为智能建筑多边形化简提供了具有工程实用价值的途径。
