Three-dimensional modeling of virtual hoisting machinery is the critical works to structure the system of virtual construction, and the foundation to realize intelligent and interactive virtual hoisting. Aimed at enha...Three-dimensional modeling of virtual hoisting machinery is the critical works to structure the system of virtual construction, and the foundation to realize intelligent and interactive virtual hoisting. Aimed at enhancing the requests of image quality and stability of the virtual construction scene, taking a tower crane for example. We studied the technology of three-dimensional modeling and optimization of a virtual tower crane, and a method named two-stage model optimization was put forward. This depended on the modeling stage using Solidworks and 3DS Max and the performance optimization stage in EON. The practice of software development indicates that the proposed methods of three-dimensional modeling and optimization could satisfy the performance request of virtual construction system and be popularized to other virtual system.展开更多
In this paper, some experimental studies on the impact of effluent from an exhaust tower of an underground tunnel with special construction are reported. By measuring the flow field downstream of the tower in NJU mete...In this paper, some experimental studies on the impact of effluent from an exhaust tower of an underground tunnel with special construction are reported. By measuring the flow field downstream of the tower in NJU meteorological wind tunnel, some flow characteristics in the make area were established. Based on these, an advanced random\|walk dispersion model was set up and applied successfully to the simulation of dispersion in the wake area. The modelling results were in accordance with wind tunnel measurements. The computed maximum of ground surface concentration in the building case was a factor of 3-4 higher than that in the flat case and appeared much closer to the source. The simulation indicated that random walk modelling is an effective and practical tool for the wake stream impact assessment.展开更多
采用压缩和激励网络(squeeze and excitation network,SENET)构建双塔推荐模型,针对卷烟消费推荐问题进行了研究.基于长期采集的包含用户、商品和历史交互信息的数据集,使用SENET双塔推荐模型对该数据集进行建模,以预测用户的卷烟消费行...采用压缩和激励网络(squeeze and excitation network,SENET)构建双塔推荐模型,针对卷烟消费推荐问题进行了研究.基于长期采集的包含用户、商品和历史交互信息的数据集,使用SENET双塔推荐模型对该数据集进行建模,以预测用户的卷烟消费行为.实验结果表明,基于SENET双塔构架的卷烟消费推荐模型在消费者与产品间的交互信息获取上具有优势;与传统的推荐算法相比,SENET双塔推荐模型具有更好的推荐效果.展开更多
针对推荐系统中双塔型神经网络难以学习用户侧和商品侧交互信息以及图连接信息的问题,提出一种二阶段孪生卷积神经网络推荐算法(TSN)。首先,以用户行为构建异质图;然后,在双塔型神经网络之间设计图卷积孪生网络,从而在学习异质图连接信...针对推荐系统中双塔型神经网络难以学习用户侧和商品侧交互信息以及图连接信息的问题,提出一种二阶段孪生卷积神经网络推荐算法(TSN)。首先,以用户行为构建异质图;然后,在双塔型神经网络之间设计图卷积孪生网络,从而在学习异质图连接信息的同时进行信息交互;最后,通过设计特殊结构的二阶段孪生信息共享机制,使得用户侧和商品侧的神经网络在训练过程中能够动态地、双向地传输信息,且有效避免神经网络串联。在基于MovieLens和豆瓣电影数据集的对比实验中,NDCG@10、NDCG@50、NDCG@100相较于最优基准算法DAT(Dual Augmented Two-tower model for online large-scale recommendation)提升了11.39%~23.98%。结果表明,所提算法能够缓解双塔型神经网络缺乏信息交互的问题,较对比算法推荐性能提升显著。展开更多
基金supported by Special Project of Scientific Research of Education Department of Shaanxi Provincial Government under Grant No.11JK0967
文摘Three-dimensional modeling of virtual hoisting machinery is the critical works to structure the system of virtual construction, and the foundation to realize intelligent and interactive virtual hoisting. Aimed at enhancing the requests of image quality and stability of the virtual construction scene, taking a tower crane for example. We studied the technology of three-dimensional modeling and optimization of a virtual tower crane, and a method named two-stage model optimization was put forward. This depended on the modeling stage using Solidworks and 3DS Max and the performance optimization stage in EON. The practice of software development indicates that the proposed methods of three-dimensional modeling and optimization could satisfy the performance request of virtual construction system and be popularized to other virtual system.
文摘In this paper, some experimental studies on the impact of effluent from an exhaust tower of an underground tunnel with special construction are reported. By measuring the flow field downstream of the tower in NJU meteorological wind tunnel, some flow characteristics in the make area were established. Based on these, an advanced random\|walk dispersion model was set up and applied successfully to the simulation of dispersion in the wake area. The modelling results were in accordance with wind tunnel measurements. The computed maximum of ground surface concentration in the building case was a factor of 3-4 higher than that in the flat case and appeared much closer to the source. The simulation indicated that random walk modelling is an effective and practical tool for the wake stream impact assessment.
文摘采用压缩和激励网络(squeeze and excitation network,SENET)构建双塔推荐模型,针对卷烟消费推荐问题进行了研究.基于长期采集的包含用户、商品和历史交互信息的数据集,使用SENET双塔推荐模型对该数据集进行建模,以预测用户的卷烟消费行为.实验结果表明,基于SENET双塔构架的卷烟消费推荐模型在消费者与产品间的交互信息获取上具有优势;与传统的推荐算法相比,SENET双塔推荐模型具有更好的推荐效果.
文摘针对推荐系统中双塔型神经网络难以学习用户侧和商品侧交互信息以及图连接信息的问题,提出一种二阶段孪生卷积神经网络推荐算法(TSN)。首先,以用户行为构建异质图;然后,在双塔型神经网络之间设计图卷积孪生网络,从而在学习异质图连接信息的同时进行信息交互;最后,通过设计特殊结构的二阶段孪生信息共享机制,使得用户侧和商品侧的神经网络在训练过程中能够动态地、双向地传输信息,且有效避免神经网络串联。在基于MovieLens和豆瓣电影数据集的对比实验中,NDCG@10、NDCG@50、NDCG@100相较于最优基准算法DAT(Dual Augmented Two-tower model for online large-scale recommendation)提升了11.39%~23.98%。结果表明,所提算法能够缓解双塔型神经网络缺乏信息交互的问题,较对比算法推荐性能提升显著。
