基于互逆分数阶算子的离散灰色模型及阶数优化

针对现有灰色预测模型主要以一阶累加生成序列为建模序列这一现象, 在互逆的分数阶累加生成算子与分数阶累减生成算子的基础上, 建立分数阶算子离散灰色模型, 并给出最小平均相对误差下最优阶数的自适应粒子群优化算法. 多个实例表明, 通过阶数优化, 分数阶算子离散灰色模型相对于灰色模型GM(1,1) 和离散灰色模型DGM(1,1) 表现出更优的拟合精度.

     

命名数据网络中面向移动用户的内容预取方法*

针对命名数据网络(Named Data Networking, NDN)移动性增强支持不足的问题,基于城区移动场景提出了一种基于移动预测的内容预取方法(Mobility-aware Prediction Approach for Content Prefetching, MAP-CP)及扩展方案(extended MAP-CP, eMAP-CP)。MAP-CP和eMAP-CP通过预测移动用户的下一接入点(Access Point, AP),预取在当前AP下因切换AP不能获取到的Data并缓存到所预测的下一AP相关联的NDN节点中,为切换AP后的移动用户就近服务,达到降低内容获取延迟和提高响应率的目的。eMAP-CP在MAP-CP的基础上,可预取切换AP过程中请求的Data,能够进一步提高性能。基于ndnSIM的仿真实验表明,和NDN本身相比,MAP-CP和eMAP-CP具有更低的平均内容获取延迟和更高的响应率,而且eMAP-CP更优。在移动速度为20m/s,发包速率为200pps时,MAP-CP和eMAP-CP的平均内容获取延迟可分别降低约4.0%和38.7%,响应率可分别提高约1.2%和4.9%。     

基于小波滤波背景预测的红外弱小目标检测方法

红外弱小目标检测是红外图像研究领域的热点与难点。有效地从背景中检测出弱小目标对于后续的跟踪、识别工作具有十分重要的意义。针对现有检测方法的不足,提出了一种基于小波滤波背景预测的红外弱小目标检测方法。该方法利用小波滤波去噪的优良特性将目标作为噪声滤除,然后构建近似的前景分布图与背景分布图,最后基于连通体筛选与对比度门限完成弱小目标的提取。采用实测光电图像对该方法进行了验证,结果表明,提出的方法能够有效抑制噪声,完成背景预测以及红外弱小目标的检测。     

深度随机森林在离网预测中的应用

在电信运营商领域,离网预测模型是企业决策者用来发现潜在离网用户(即停用运营商服务)的主要手段。目前离网预测模型都是基于逻辑回归、决策树、神经网络及随机森林等浅层机器学习算法,但是在大数据的背景下,这些浅层算法在预测问题上很难取得更高的精度。因此,提出了一种新型的深层结构模型——深度随机森林,通过将传统浅层随机森林堆积成深层结构模型,获得更高的预测精度。在运营商真实数据上进行了大量实验,结果证明深层随机森林模型比传统浅层机器学习算法在离网预测问题上可以得到更好的效果。同时,增大训练数据量可以进一步提升深层随机森林的预测能力,从而证明了在大数据环境下深层模型的潜力。     

一种基于概率的路径预测与查询算法

研究了路网空间内的路径预测与查询技术,设计了基于统计信息和概率论的最优路径预测算法。实际应用中,路网错综复杂。提出可能路径集合的概念,并设计算法来提取当前路径预测涉及到的路网子网,减小路网规模和路径预测的复杂度。在空间网络环境下,现有移动对象位置预测技术主要针对短期预测,不能预测下一路口的交通情况。为了弥补这一缺陷,降低用户端的位置更新率,设计了路网移动模型来简洁描述提取自大量历史移动路径的移动统计特征,捕捉路口处转向模式。基于移动模型,提出了具有较高精度的交通预测模型来预测对象的运动路径。     

可充电传感网中移动式能量补给及数据收集策略研究

可充电无线传感器网络是一种新型的无线传感网,它利用移动充电车在收集数据的同时给能量低的节点充电,可广泛应用于需要长期监测环境的应用中。但是,移动充电车如何在给定的延迟内完成数据收集,降低网络能耗并尽可能多地给低能量节点补充能量是一个具有挑战性的问题。因此提出一个新的算法RSEP(Root Selection with Energy Prediction)。首先,限定充电车的路径长度以保证延迟。然后,将路径上的低能量节点作为根节点,构造多棵数据收集树。若根节点能量可以保证其短期内不会死亡,则从树中寻找一条等于树的直径的路径。在该路径上选取网络中邻居最多的节点作为新的根节点,以改变树的结构来降低树高。树上的节点将它们的数据及能量信息沿着树传送到根节点。最后,移动充电车沿着充电路径为各个根节点充电时,就可以收集各个树上节点的数据及能量信息。此外,充电车收集到的能量信息会随着时间推移而“过时”,而能量信息是根节点选择时的重要参考因素。因此,充电车利用马尔科夫模型预测节点在下一轮数据收集开始时的能量,从而优化根节点的选择。仿真实验结果表明,与目前已有的算法相比,RSEP算法可以以较少的网络总能耗完成充电,并且每轮充电时间均较短。     

Robust Cardiac Function Assessment in 4D PC‐MRI Data of the Aorta and Pulmonary Artery

Four‐dimensional phase‐contrast magnetic resonance imaging (4D PC‐MRI) allows the non‐invasive acquisition of time‐resolved, 3D blood flow information. Stroke volumes (SVs) and regurgitation fractions (RFs) are two of the main measures to assess the cardiac function and severity of valvular pathologies. The flow rates in forward and backward direction through a plane above the aortic or pulmonary valve are required for their quantification. Unfortunately, the calculations are highly sensitive towards the plane's angulation since orthogonally passing flow is considered. This often leads to physiologically implausible results. In this work, a robust quantification method is introduced to overcome this problem. Collaborating radiologists and cardiologists were carefully observed while estimating SVs and RFs in various healthy volunteer and patient 4D PC‐MRI data sets with conventional quantification methods, that is, using a single plane above the valve that is freely movable along the centerline. By default it is aligned perpendicular to the vessel's centerline, but free angulation (rotation) is possible. This facilitated the automation of their approach which, in turn, allows to derive statistical information about the plane angulation sensitivity. Moreover, the experts expect a continuous decrease of the blood flow volume along the vessel course. Conventional methods are often unable to produce this behaviour. Thus, we present a procedure to fit a monotonous function that ensures such physiologically plausible results. In addition, this technique was adapted for the usage in branching vessels such as the pulmonary artery. The performed informal evaluation shows the capability of our method to support diagnosis; a parameter evaluation confirms the robustness. Vortex flow was identified as one of the main causes for quantification uncertainties.     

Fast fractal prediction of surge over underwater vehicle

In order to realize accurate control of underwater vehicles, this paper proposed a fractal prediction model of surge. Firstly, we introduce the current state-of-the-art of fractal prediction; secondly, we complete a fractal analysis with the R_S method, calculating the fractal dimension of surge using the short-time dimension method; thirdly, we correct the vertical scaling factor of the Iterated Function System (IFS) using the fractal dimension. Then the predictive value can be calculated through anti-iteration according to the statistical significantly IFS and proposing a fractal prediction model. Finally, we complete a simulation analysis based on the actual data and a comparison to another prediction algorithm. The result shows that the fractal prediction algorithm of this article has better comprehensive performance.     

基于模体演化的时序链路预测方法

时序链路预测是动态网络分析的重要组成部分,具有极大的理论和应用价值. 传统的时序链路预测方法往往直接对边的演化规律进行分析,忽略了网络中其他微观结构的演化对链路形成的影响. 基于此分析,本文引入非负张量分解和时间序列分析对网络模体的演化规律进行研究,进而提出一种基于模体演化的链路预测方法. 在三个真实数据集上的实验结果表明,该方法能有效提高链路预测精度.     

Single Criterion Net Flow Using Concordance and Discordance Relations

This article proposes a flow-based method using the single criterion net flow for pattern classification. Traditional single criterion net flow, generated on the basis of net partial concordance indices, measures the preference intensity for one pattern with respect to all others on this criterion. A concordance index measures the intensity of preference for one pattern over another pattern. Contrary to traditional flow-based methods, such as PROMETHEE, in which only the concordance relation is taken into account, the proposed method proposes a new definition of single criterion net flow using both concordance and discordance relations. This can balance pros and cons for the preference of one pattern over another pattern for a criterion. The final classification decision for a new pattern depends on its net flow. Criteria weights are determined using a genetic-algorithm-based approach. Empirical results involving bankruptcy prediction demonstrate that the proposed method performs well compared to other well-known classification methods.