Metafrontier efficiency analysis for hospital pharmacy services using dynamic network DEA framework

Medication management is a complex but critical process within the healthcare delivery system. The pharmacy department in healthcare facilities is responsible for acquisition, distribution and dispensing of medication to patients that involves a series of network process. This process can be categorized into two main subprocesses: i) pharmacy store and ii) patient services (i.e., ambulatory and inpatient pharmacy). The objective of this study was to investigate the relative efficiency level of the two subprocesses of pharmacy services in Malaysia's public hospitals. We used a nonconvex metafrontier technology to estimate technical efficiency of hospital pharmacies between specialist and nonspecialist hospital categories under the dynamic network data envelopment analysis (DNDEA) framework. For overall efficiency, both specialist and nonspecialist categories had similar mean efficiency scores. However, when measuring the efficiency level according to pharmacy store and patient services separately, the nonspecialist category had a higher mean efficiency score in pharmacy store, whereas for patient service efficiency, both categories had similar results. The patient service efficiency level was lower consistently over time, suggesting that the overall inefficiency of hospital pharmacy is greatly influenced by the inefficiency of the patient service. This suggests that systematic improvements in patient service efficiency should be a higher priority than pharmacy store efficiency to further increase the overall efficiency level. We believe that our findings will provide valuable information to policymakers towards achieving an efficient public healthcare service.     

一种基于量子加密的软件定义网络南向安全防护策略

提出一种以量子加密技术为核心的软件定义网络（Software-defined Network，SDN）的南向安全防护策略，建立了基于量子密钥分发的SDN网络模型并分析了该模型下的身份认证、量子密钥分发和自适应免疫窃听流程。推导出了本策略下窃听攻击强度、用于检测窃听的量子态数目的比例α、系统发现窃听者的概率、密钥分发效率之间的定量关系。通过动态仿真分析确立了不同攻击强度下系统参数α的最优选值区间。SDN控制器可根据该仿真结果，动态地决策系统中α的取值进行自适应免疫密钥分发，实施更为精准的窃听免疫方案。     

基于深度学习的跳频信号识别

针对跳频信号分选存在人工提取参数特征具有复杂性的问题，提出了一种基于深度学习的识别方法。首先对跳频信号进行短时傅里叶变换，得到二维的时频矩阵；接着提取信号的轮廓特征，构造三维矩阵作等高线图，并对等高线图进行预处理；最后把预处理后的等高线图输入到卷积神经网络中进行训练、测试，进而实现分类识别。仿真结果表明，在不需要复杂的人工提取参数特征的基础上，在分选率为100〖WT《Times New Roman》〗%〖WTBZ〗时，所提方法经裁剪处理下的信噪比为-15 dB，比支持向量机和传统K-Means聚类算法都低10 dB。实测数据的算法验证表明，所提方法能够将大疆精灵4Pro、hm无人机、司马航模X8HW以及大疆悟2这四类无人机正确分类。     

基于图论的认知无线网络频谱动态分配

在认知无线电网络中，图论与量子遗传算法相结合的频谱分配策略能够提高频谱利用率，但存在早熟和收敛精度不够等缺点。为了解决该问题而实现算法的优化，对图着色理论的频谱分配模型进行数学建模，并针对该模型提出了改进的量子遗传算法。首先，通过使用小生境技术初始化种群，使种群分布更加广泛、算法的收敛度更高；其次，根据进化代数对量子旋转角进行实时动态调整，对染色体进行阈值变异，防止个体陷入早熟，跳出局部解；然后，对干扰约束条件进行重新设计，有效地避免盲目性，提高了网络的公平性和网络效益。仿真结果表明，所提算法有效地提高了频谱利用率，极大地增强了网络系统的性能。     

大跨度天基数据链的TDD时隙调度分析与优化

针对天基数据链系统卫星间的传播时延较大且节点间距离变化较大等特点，提出了一种提高时频资源利用率、提升系统容量、减小时延的方法。在时分多址与空分多址相结合的接入方式下，进行了时隙调度的分析与优化，利用卫星星座的空间复用与大跨度传播时延特性，采用时隙编排、收发状态调整、帧格式优化等手段，将信道利用率和系统容量提升了1倍，将卫星节点在某一方向获得服务的平均等待时间缩短为原来的1/4，提高了天基数据链系统的服务质量。     

“北斗”救生终端海上自适应功率退避算法

受海况、浸水、结冰、对星等因素影响，飞行员海上遇险后报位可靠性较低，工作时长较短。为此，提出一种自适应功率退避方法，在对报位影响因素分析基础上，利用深度神经网络，通过对经纬度、高度、俯仰角、方向、加速度、速度、浸水、信号、电量等多维度数据进行学习，拟合出覆冰、浪涌、液面等信号特征与环境特征联想模型，自适应控制信号发射时机，实现功耗的有效控制。试验表明，该方法将通信成功率及工作时长分别由36.3%、6.0 h提升至73.3%、8.6 h。     

平台经济视域下的商家投机行为治理——基于平台主动治理视角

平台型网络市场的虚拟性与隐蔽性导致商家隐瞒、欺骗消费者的行为愈演愈烈，衍生了新的治理难题。基于交易成本理论和社会交换理论，立足平台主动治理视角，文章探索平台主动治理中的事前商家规范机制、事中交易保障机制和事后争议处理机制对商家强、弱投机行为的影响，以及平台声誉对其影响的调节作用。基于消费者视角开展问卷调查，对数据进行结构方程模型分析和分层回归分析，结果表明：弱投机行为对强投机行为具有显著正向影响；商家规范机制和争议处理机制对强、弱投机行为均具有抑制作用；而交易保障机制对弱投机行为具有促进作用；平台声誉对交易保障机制、争议处理机制和弱投机行为之间的关系具有负向调节作用。     

Knowledge distance and R&D collaboration in Cournot oligopoly

A Cournot collaboration game is complemented with the hypothesis that the cost of R&D collaboration depends on firm’s directed knowledge distance with respect to potential R&D partners. The further hypothesis that directed distance may be non‐uniform and pair‐wise asymmetric reconciles theoretical predictions with qualitative evidence about concentration and multiplicity of components in industrial R&D networks. The welfare properties of sparse stable networks are also affected.     

基于机器学习的斜坡堤越浪量预测方法研究

针对斜坡堤越浪量预测方法,分别建立集成神经网络(ensemble neural network,ENN)、随机森林(random for-eset,RF)和支持向量回归机(suppport vector regression,SVR)3种机器学习模型对斜坡堤越浪量进行预测,并利用决定系数R2和均方根误差RMSE来评估模型性能.最后,对3种模型的性能进行分析.结果显示,集成神经网络模型的决定系数R2和均方根误差RM S E分别约为0.96和0.0018,随机森林模型的决定系数R2和均方根误差RMSE分别约为0.97和0.0014,支持向量回归机模型的决定系数R2和均方根误差RMSE分别约为0.94和0.002.对比发现,3种模型的决定系数都达到0.9以上,都具有较高的预测精度,随机森林相比其他两个模型精度更高.     

Dynamics of firm's network community associations and firm's innovation performance

ABSTRACT

R&D network structures have crucial impacts on firm's innovation performance. However, most previous studies have been based on the whole or ego network perspective, few studies have investigated the influence of community structure that a firm is engaged in on its innovation performance, and it is still unclear how a firm's relation to network community affect its innovation performance. This research aims to address this gap by focusing on the dynamics of firm's network community associations, empirically investigate the relationship between dynamics of firm's network community associations and its innovation performance. Based on the unbalanced panel data of smartphone R&D network during year 2004–2017, the results demonstrate that change in community member associations and movement across communities both have inverted-U-shaped effects on firm's innovation performance. Moreover, innovation openness depth has moderating effects on the relationships between dynamics of firm's network community associations and its innovation performance.