The finite element method is used to simulate the rectification process of shield machine, to study the relationship between rectification moment and angle and to explore the influence laws of different soil parameter...The finite element method is used to simulate the rectification process of shield machine, to study the relationship between rectification moment and angle and to explore the influence laws of different soil parameters and buried depth on rectification moment. It is hoped that the reference value of rectification moment can be offered to operator, and theoretical foundation can be laid for future automatic rectification technology. The results show that the rectification moment and angle generally exhibit good linear behavior in clay layers with different soil parameters or buried depths, and then the concept of rectification coefficient, that is, the ratio of rectification angle to rectification moment, is proposed; different soil parameters and buried depths have different influences on rectification coefficient, in which elastic modulus has great influence but others have little influences; the simulations of rectification process are preformed in clay layers with different elastic modulus, and fitting results show that elastic modulus and rectification coefficient present the quadratic function relation.展开更多
According to the actual engineering problem that the precise load model of shield machine is difficult to achieve,a design method of sliding mode robust controller oriented to the automatic rectification of shield mac...According to the actual engineering problem that the precise load model of shield machine is difficult to achieve,a design method of sliding mode robust controller oriented to the automatic rectification of shield machine was proposed. Firstly,the nominal load model of shield machine and the ranges of model parameters were obtained by the soil mechanics parameters of certain geological conditions and the messages of the self-learning of shield machine by tunneling for previous segments. Based on this rectification mechanism model with known ranges of parameters,a sliding mode robust controller was proposed. Finally,the simulation analysis was developed to verify the effectiveness of the proposed controller. The simulation results show that the sliding mode robust controller can be implemented in the attitude rectification process of the shield machine and it has stronger robustness to overcome the soil disturbance.展开更多
针对小样本学习过程中样本数量不足导致的性能下降问题,基于原型网络(Prototype network,ProtoNet)的小样本学习方法通过实现查询样本与支持样本原型特征间的距离度量,从而达到很好的分类性能.然而,这种方法直接将支持集样本均值视为类...针对小样本学习过程中样本数量不足导致的性能下降问题,基于原型网络(Prototype network,ProtoNet)的小样本学习方法通过实现查询样本与支持样本原型特征间的距离度量,从而达到很好的分类性能.然而,这种方法直接将支持集样本均值视为类原型,在一定程度上加剧了对样本数量稀少情况下的敏感性.针对此问题,提出了基于自适应原型特征类矫正的小样本学习方法(Few-shot learning based on class rectification via adaptive prototype features,CRAPF),通过自适应生成原型特征来缓解方法对数据细微变化的过度响应,并同步实现类边界的精细化调整.首先,使用卷积神经网络构建自适应原型特征生成模块,该模块采用非线性映射获取更为稳健的原型特征,有助于减弱异常值对原型构建的影响;然后,通过对原型生成过程的优化,提升不同类间原型表示的区分度,进而强化原型特征对类别表征的整体效能;最后,在3个广泛使用的基准数据集上的实验结果显示,该方法提升了小样本学习任务的表现.展开更多
基金Project(2007CB714006)supported by the National Basic Research Program of China
文摘The finite element method is used to simulate the rectification process of shield machine, to study the relationship between rectification moment and angle and to explore the influence laws of different soil parameters and buried depth on rectification moment. It is hoped that the reference value of rectification moment can be offered to operator, and theoretical foundation can be laid for future automatic rectification technology. The results show that the rectification moment and angle generally exhibit good linear behavior in clay layers with different soil parameters or buried depths, and then the concept of rectification coefficient, that is, the ratio of rectification angle to rectification moment, is proposed; different soil parameters and buried depths have different influences on rectification coefficient, in which elastic modulus has great influence but others have little influences; the simulations of rectification process are preformed in clay layers with different elastic modulus, and fitting results show that elastic modulus and rectification coefficient present the quadratic function relation.
基金Project(2007CB714006) supported by the National Basic Research Program of China
文摘According to the actual engineering problem that the precise load model of shield machine is difficult to achieve,a design method of sliding mode robust controller oriented to the automatic rectification of shield machine was proposed. Firstly,the nominal load model of shield machine and the ranges of model parameters were obtained by the soil mechanics parameters of certain geological conditions and the messages of the self-learning of shield machine by tunneling for previous segments. Based on this rectification mechanism model with known ranges of parameters,a sliding mode robust controller was proposed. Finally,the simulation analysis was developed to verify the effectiveness of the proposed controller. The simulation results show that the sliding mode robust controller can be implemented in the attitude rectification process of the shield machine and it has stronger robustness to overcome the soil disturbance.
文摘针对小样本学习过程中样本数量不足导致的性能下降问题,基于原型网络(Prototype network,ProtoNet)的小样本学习方法通过实现查询样本与支持样本原型特征间的距离度量,从而达到很好的分类性能.然而,这种方法直接将支持集样本均值视为类原型,在一定程度上加剧了对样本数量稀少情况下的敏感性.针对此问题,提出了基于自适应原型特征类矫正的小样本学习方法(Few-shot learning based on class rectification via adaptive prototype features,CRAPF),通过自适应生成原型特征来缓解方法对数据细微变化的过度响应,并同步实现类边界的精细化调整.首先,使用卷积神经网络构建自适应原型特征生成模块,该模块采用非线性映射获取更为稳健的原型特征,有助于减弱异常值对原型构建的影响;然后,通过对原型生成过程的优化,提升不同类间原型表示的区分度,进而强化原型特征对类别表征的整体效能;最后,在3个广泛使用的基准数据集上的实验结果显示,该方法提升了小样本学习任务的表现.
