基于梯度分布调节策略的Xgboost算法优化

为了解决集成学习模型Xgboost在二分类问题中少数类检出率低的问题，提出了基于梯度分布调节策略的改进的Xgboost算法——LCGHA-Xgboost。首先，通过定义损失贡献（LC）来模拟Xgboost算法中样本个体的损失量；而后，通过定义损失贡献密度（LCD）来衡量Xgboost算法中样本被正确分类的难易程度；最后，提出了梯度分布调节算法LCGHA，依据LCD动态调整样本个体的一阶梯度分布，间接地增大难分样本（主要存在于少数类中）的损失量，减小易分样本（主要存在于多数类中）的损失量，使Xgboost算法偏向对难分样本的学习。实验结果表明,与Xgboost、GBDT、随机森林（Random_Forest）这三大集成学习算法相比，LCGHA-Xgboost算法在多个UCI数据集上的召回率（Recall）值有5.4%~16.7%的提高，AUC值有0.94%~7.41%的提高；在垃圾网页数据集WebSpam-UK2007和DC2010数据集上所提算法的Recall值更是有44.4%~383.3%的提高，AUC值有5.8%~35.6%的提高。LCGHA-Xgboost算法可以有效提高对少数类的分类检出能力，减小少数类的分类错误率。     

[可靠性建模及试验技术]重型机床液压元件油液污染退化量分布剖面可靠性建模

针对重型机床液压系统故障频繁且多与油液中的固态颗粒污染物相关的问题，进行了油液污染趋势变化试验。通过时域分析获得了油样颗粒数的有量纲和量纲一参数，通过Q-Q图和K-S检验分析有量纲参数，污染颗粒数是退化量服从正态分布的退化数据。进行了油液污染与环境相关性分析试验，采用相关系数法分析得到，颗粒数变化量与一定范围内的温度、流量、压力的相关性小；将液压元件分为管路、阀、过滤器三类，用直径5 μm左右的颗粒和直径大于15 μm的颗粒分别研究管路及阀件的堵塞和磨损情况，以过滤器过滤精度大小的颗粒研究过滤器的堵塞情况，设定ISO4406标准20/17级对应的颗粒数为阈值，利用退化量分布建立了液压元件单一故障模式的可靠性模型；利用竞争失效模型将上述模型融合为多故障模式下的可靠性模型。     

Integrated multi-factory production and distribution scheduling applying vehicle routing approach

This paper introduces a new integrated multi-factory production and distribution scheduling problem in supply chain management. This supply chain consists of a number of factories joined together in a network configuration. The factories produce intermediate or finished products and supply them to other factories or to end customers that are distributed in various geographical zones. The problem consists of finding a production schedule together with a vehicle routing solution simultaneously to minimise the sum of tardiness cost and transportation cost. A mixed-integer programming model is developed to tackle the small-sized problems using CPLEX, optimally. Due to the NP-hardness, to deal with medium- and large-sized instances, this paper develops a novel Improved Imperialist Competitive Algorithm (IICA) employing a local search based on simulated annealing algorithm. Performance of the proposed IICA is compared with the optimal solution and also with four variants of population-based metaheuristics: Imperialist Competitive Algorithm, Genetic Algorithm, Particle Swarm Optimisation (PSO), and Improved PSO. Based on the computational results, it is statistically shown that quality of the IICA’s solutions is the same as optimal ones solving small problems. It also outperforms other algorithms in finding near-optimal solutions dealing with medium and large instances in a reasonably short running time.     

Modeling and analysis for multilayered complex communication network of satellite navigation system

This study is extended to construct the network model, the node model, and the link model of complex communication network for satellite navigation system (CCN‐SNS) based on the hierarchical architecture. Firstly, a method called snapshots was proposed to describe the dynamic topology for CCN‐SNS; secondly, another method was put forward to model the different nodes of the CCN‐SNS; thirdly, the different links between every two different nodes were modeled. Therefore, based on the OPNET tools, a simulation for the CCN‐SNS, which contains the models that proposed earlier used to analyze the navigation accuracy and network transmission performance, was performed.     

Refinement of primary carbides in AISI DC 53 cold work tool steel by thin section copper permanent mould casting and water atomization

In this study, as alternative methods to conventional production methods involving thin section copper permanent mould casting (fast solidification) and water atomization (rapid solidification) of AISI DC 53 cold work tool steel have been investigated. Thin slabs obtained by copper mould casting was homogenized at 1150 °C for one hour and then hot rolling was be applied. After hot rolling, conventional heat treatment have been applied to thin slabs. The same heat treatment procedure have been performed for a commercially available AISI DC 53 which is manufactured by thick section ingot casting and rolling. In order to investigate also the effect of rapid solidification on AISI DC 53 tool steel, steel powder was produced by water atomization technique. Experiments have revealed that thin steel slabs in permanent copper mould and rapidly solidified tool steel powder by water atomization have a more refined primary M₇C₃ carbides than commercially available steel.     

改进的卷积神经网络单幅图像超分辨率重建

针对经典的基于卷积神经网络的单幅图像超分辨率重建方法网络较浅、提取的特征少、重建图像模糊等问题，提出了一种改进的卷积神经网络的单幅图像超分辨率重建方法，设计了由密集残差网络和反卷积网络组成的新型深度卷积神经网络结构。原始低分辨率图像输入网络，利用密集残差学习网络获取更丰富的有效特征并加快特征梯度流动，其次通过反卷积层将图像特征上采样到目标图像大小，再利用密集残差学习高维特征，最后融合不同卷积核提取的特征得到最终的重建图像。在Set5和Set14数据集上进行了实验，并和Bicubic、K-SVD、SelfEx、SRCNN等经典重建方法进行了对比，重建出的图像在整体清晰度和边缘锐度方面更好，另外峰值信噪比（PSNR）平均分别提高了2.69?dB、1.68?dB、0.74?dB和0.61?dB。实验结果表明，该方法能够获取更丰富的细节信息，得到更好的视觉效果，达到了图像超分辨率的增强任务。     

A parametric investigation on seat/occupant contact forces and their relationship with initially perceived discomfort using a configurable seat

The present work investigates the contact forces between sitters and seat as well as their correlations with perceived discomfort. Twelve different economy class aeroplane seat configurations were simulated using a multi-adjustable experimental seat by varying seat pan and backrest angles, as well as seat pan compressed surface. Eighteen males and 18 females, selected by their body mass index and stature, tested these configurations for two sitting postures. Perceived discomfort was significantly affected by seat parameters and posture and correlated both with normal force distribution on the seat-pan surface and with normal forces at the lumbar and head supports. Lower discomfort ratings were obtained for more evenly distributed normal forces on the seat pan. Shear force at the seat pan surface was at its lowest when sitters were allowed to self-select their seat-pan angle, supporting that a shear force should be reduced but not zeroed to improve seating comfort.

Practitioner Summary: The effects of seat-pan and backrest angle, anthropometric dimensions and sitting posture on contact forces and perceived discomfort were investigated using a multi-adjustable experimental seat. In addition to preferred seat profile parameters, the present work provides quantitative guidelines on contact force requirement for improving seating comfort.     

Super-Resolution Reconstruction of Images Based on Microarray Camera

In the field of images and imaging, super-resolution (SR) reconstruction of images is a technique that converts one or more low-resolution (LR) images into a highresolution (HR) image. The classical two types of SR methods are mainly based on applying a single image or multiple images captured by a single camera. Microarray camera has the characteristics of small size, multi views, and the possibility of applying to portable devices. It has become a research hotspot in image processing. In this paper, we propose a SR reconstruction of images based on a microarray camera for sharpening and registration processing of array images. The array images are interpolated to obtain a HR image initially followed by a convolution neural network (CNN) procedure for enhancement. The convolution layers of our convolution neural network are 3×3 or 1×1 layers, of which the 1×1 layers are used to improve the network performance particularly. A bottleneck structure is applied to reduce the parameter numbers of the nonlinear mapping and to improve the nonlinear capability of the whole network. Finally, we use a 3×3 deconvolution layer to significantly reduce the number of parameters compared to the deconvolution layer of FSRCNN-s. The experiments show that the proposed method can not only ameliorate effectively the texture quality of the target image based on the array images information, but also further enhance the quality of the initial high resolution image by the improved CNN.