基于GFU和分层LSTM的组群行为识别研究方法

提出一种以"关键人物"为核心，使用门控融合单元（GFU，Gated Fusion Unit）进行特征融合的组群行为识别框架，旨在解决两个问题：①组群行为信息冗余，重点关注关键人物行为特征，忽略无关人员对组群行为的影响；②组群内部交互行为复杂，使用GFU有效融合以关键人物为核心的交互特征，再通过LSTM时序建模成为表征能力更强的组群特征.最终，通过softmax分类器进行组群行为类别分类.该算法在排球数据集上取得了86.7%的平均识别率.     

改进ILoG算子的故障检测方法

针对强背景噪声干扰下轮对轴承故障特征微弱、难以准确检测的问题,提出了一种自适应改进高斯拉普拉斯(improved Laplacian of Gaussian,简称ILoG)算子的微弱故障检测方法。ILoG算子滤波器具有优良的信号突变特征检测能力,将其用于轮对轴承故障信号的冲击特征检测,同时利用水循环算法(water cycle algorithm,简称WCA)的寻优特性,并行搜寻筛选最佳的ILoG算子影响参数,通过对参数优化后ILoG算子滤波后信号做进一步包络解调分析,提取出轮对轴承微弱的故障特征信息。对实际轮对轴承外圈和内圈故障信号分析的结果表明,该方法可以有效检测出轴承微弱故障特征频率,故障检测效果优于小波阈值和多尺度形态学差值滤波方法。     

自适应融合层级特征的混合退化图像复原算法

多种退化类型混合的图像比单一类型的退化图像降质更严重,很难建立精确模型对其复原,研究端到端的神经网络算法是复原的关键.现有的基于操作选择注意力网络的算法(operation-wiseattentionnetwork,OWAN)虽然有一定的性能提升,但是其网络过于复杂,运行较慢,复原图像缺乏高频细节,整体效果也有提升的空间.针对这些问题,提出一种基于层级特征融合的自适应复原算法.该算法直接融合不同感受野分支的特征,增强复原图像的结构;用注意力机制对不同层级的特征进行动态融合,增加模型的自适应性,降低了模型冗余;另外,结合L1损失和感知损失,增强了复原图像的视觉感知效果.在DIV2K,BSD500等数据集上的实验结果表明,该算法无论是在峰值信噪比和结构相似性上的定量分析,还是在主观视觉质量方面,均优于OWAN算法,充分证明了该算法的有效性.     

辊速差对连铸连轧7075铝板显微组织、织构及力学性能的影响

研究辊速差对连铸连轧7075铝板显微组织、织构及力学性能的影响。采用3种不同上辊/下辊转速比(ω/ω0,ω为上辊转速,ω0为下辊转速)1:1、1:1.2及1:1.4进行多次试验。结果显示,在最大辊速差条件下(ω/ω0=1:1.4),7075铝板在轧制方向的屈服强度和极限抗拉强度分别提高41.5%和21.9%。此外,当辊速比ω/ω0为1:1.4时,成品轧制板的平均晶粒尺寸减小36%,横剖面平均硬度增加约9.2%。织构研究结果显示,辊速差越大,成品各向同性及硬度越大。然而,采用不同辊速度的连铸连轧会导致变形板伸长率降低约6%。     

数据中心内服务器能耗数据采集及特征分析

数据中心的高能耗低能效问题正受到广泛关注与研究，但目前没有公开的数据中心内服务器能耗数据集供研究人员使用，且现有过滤式特征选择并不能满足运维人员的需求。为此，提出了一套模拟数据中心内服务器运行状态的仿真环境架构，基于该架构采集了服务器运行各类任务时的多项性能指标和能耗数据。然后将基于因果关系的特征选择应用于能耗数据集的特征分析中，构建出可解释的特征子集和能耗预测结果。实验结果表明，因果特征子集大小约为过滤式特征子集大小的1/3到1/6，并且使用因果特征子集训练的模型在75%的情况下都取得了最优预测精度。     

Mixture ratio modeling of dynamic systems

Any knowledge extraction relies (possibly implicitly) on a hypothesis about the modelled-data dependence. The extracted knowledge ultimately serves to a decision-making (DM). DM always faces uncertainty and this makes probabilistic modelling adequate. The inspected black-box modeling deals with “universal” approximators of the relevant probabilistic model. Finite mixtures with components in the exponential family are often exploited. Their attractiveness stems from their flexibility, the cluster interpretability of components and the existence of algorithms for processing high-dimensional data streams. They are even used in dynamic cases with mutually dependent data records while regression and auto-regression mixture components serve to the dependence modeling. These dynamic models, however, mostly assume data-independent component weights, that is, memoryless transitions between dynamic mixture components. Such mixtures are not universal approximators of dynamic probabilistic models. Formally, this follows from the fact that the set of finite probabilistic mixtures is not closed with respect to the conditioning, which is the key estimation and predictive operation. The paper overcomes this drawback by using ratios of finite mixtures as universally approximating dynamic parametric models. The paper motivates them, elaborates their approximate Bayesian recursive estimation and reveals their application potential.     

Review of lithium-ion battery state of charge estimation

The technology deployed for lithium-ion battery state of charge (SOC) estimation is an important part of the design of electric vehicle battery management systems. Accurate SOC estimation can forestall excessive charging and discharging of lithium-ion batteries, thereby improving discharge efficiency and extending cycle life. In this study, the key lithium-ion battery SOC estimation technologies are summarized. First, the research status of lithium-ion battery modeling is introduced. Second, the main technologies and difficulties in model parameter identification for lithium-ion batteries are discussed. Third, the development status and advantages and disadvantages of SOC estimation methods are summarized. Finally, the current research problems and prospects for development trends are summarized.     

Data-driven parameterization of polymer electrolyte membrane fuel cell models via simultaneous local linear structured state space identification

In order to mitigate the degradation and prolong the lifetime of polymer electrolyte membrane fuel cells, advanced, model-based control strategies are becoming indispensable. Thereby, the availability of accurate yet computationally efficient fuel cell models is of crucial importance. Associated with this is the need to efficiently parameterize a given model to a concise and cost-effective experimental data set. A challenging task due to the large number of unknown parameters and the resulting complex optimization problem. In this work, a parameterization scheme based on the simultaneous estimation of multiple structured state space models, obtained by analytic linearization of a candidate fuel cell stack model, is proposed. These local linear models have the advantage of high computational efficiency, regaining the desired flexibility required for the typically iterative task of model parameterization. Due to the analytic derivation of the local linear models, the relation to the original parameters of the non-linear model is retained. Furthermore, the local linear models enable a straight-forward parameter significance and identifiability analysis with respect to experimental data. The proposed method is demonstrated using experimental data from a 30 kW commercial polymer electrolyte membrane fuel cell stack.