基于半监督动态深度融合神经网络的软测量

半监督深度神经网络建模方法已被广泛应用于软测量,但基于分层训练的网络在特征提取过程局限于挖掘每层输入的有效信息,忽略了原始输入有效信息的丢失,逐层累积,从而导致原始输入的特征表示准确率低下;另外,缺乏挖掘过程时空相关性,也会导致模型性能退化。提出一种半监督动态深度融合神经网络(semi-supervised dynamics deep fusion neural network,SS-DDFNN)方法。该方法在特征提取网络的每层都重构原始输入数据并预测质量变量,通过在预训练损失中使用重构原始输入误差,减小原始输入有效信息的丢失;同时融入注意力机制和t分布随机邻域嵌入提取时空相关信息,应用提取的特征建立门控神经网络质量预测模型。实验结果显示,相较于SAE、GSTAE和SIAE模型,所提方法在脱丁烷塔案例中的预测精度分别提升了2.8%、1.1%和0.9%;在工业聚乙烯生产案例中,分别提升了2.7%、1.0%和0.7%。实验结果验证了所提方法的有效性。

Nature and Kinetic Architecture： The Development of a New Type of Transformable Structure for Temporary Applications

The contemplation of contemporary architectural designs shows an increasing demand for the development of more adaptable, flexible and transformable structures. This type of structures can adapt with different environmental conditions and meet different functions. This can help in reducing environmental waste and pollution associated with many buildings and above all can save on cost and time. Natural systems have inspired human being, since they began to build and design. Architects and designers have utilized nature as one of the main resources of information for the creation of innovative architectural spaces. One of the unique features of natural structures is the way that their components open and close in order to respond to a particular requirement or the environmental changes. This aspect has inspired many designers for the development of transformable architectural structures that can change their shape and geometry to be able to adapt with specific conditions. To make a way toward the design of transformable structure for temporary applications, the authors have developed a new type of adaptable structures according to natural forms. The proposed design applies the transformation principles that exist in potato＇s flower and the movement mechanisms used in a spider＇s leg. The design is able to fit to different topographies and have a potential to be folded to a very compact state in a very short period of time. The detailed design and the different configurations of the system applications will be presented in this paper. The result of the study shows that using modular triangular plates can create a changeable module that is not only able to respond to different functions and environmental changes but it is also able to shape different configuration to be able to respond to different user＇s ambitions. The compactability of this structure into 1/3 of its base dimensions; makes its transportation fast and with minimum costs. These capabilities make this structure suitable for temporary buildings such as exhibitions, temporary settlements or hospital in damaged areas.