As an efficient incremental learning system based on random vector function-link network(RVFLN),broad learning system(BLS)has the characteristics of fast adaptive model structure selection and high precision.However,due to the lack of label data in target classification,the traditional BLS is difficult to improve the classi-fication effect of target domain by using relevant domain knowledge.Therefore,a domain adaptive BLS(DABLS)model based on manifold regularization framework and maximum mean discrepancy(MMD)is developed to achieve cross-domain image classification of target domain under unlabeled condition.Firstly,the feature nodes and en-hancement nodes of BLS are constructed to effectively extract features from the data of source domain and target domain.The manifold regularization framework is used to construct Laplacian matrix in order to explore the mani-fold characteristics of the target domain data and mine the potential information of the target domain data.Then the transfer learning method is used to construct the MMD penalty term between the source domain data and the target domain data to match the projection mean between the source domain and the target domain.The feature nodes,enhancement nodes,MMD penalty term and Laplacian matrix are combined to construct the objective func-tion.Ridge regression analysis is used to solve the objective function to obtain the output coefficients,so as to im-prove the cross-domain classification performance.Finally,a large number of validation and comparative experi-ments are carried out on different image data sets,and the experiment results show that the DABLS can better achieve cross-domain classification on different image data sets,and has strong generalization ability and better sta-bility.
The problem of finite-time global prescribed performance control (GPPC) for vehicular platoon control system (VPCS) with actuator nonlinearities is investigated in this paper. A smooth equivalent transformation is first presented to process both actuator dead-zone and saturation under the same framework. Then, a GPPC method based on improved finite-time performance function (IFnTPF) is proposed, and makes the tracking error tend to the prescribed region within the predefined time, while removes the strict restrictions on the initial conditions of the system. To realize the platoon in given time, an adaptive finite-time sliding mode control scheme is further developed, and both individual vehicle stability and string stability are guaranteed. Numerical simulations demonstrate the validity of the given scheme.
Considering the accuracy, generalization ability, stability, and training efficiency of a furnace temperature model in the process of municipal solid waste incineration, a heterogeneous feature ensemble modeling method for furnace temperature is proposed in this paper. First, heterogeneous features are generated according to the operation mechanism of the waste incineration process, and the training subset of the furnace temperature- and grate temperature-based model is determined from the historical data of this process. Second, the base model pools of furnace temperature and grate temperature are constructed by a regularized stochastic configuration network, and a set of optimal base models are retained by selective base model technology. Then, a negative correlation learning strategy is employed to establish a simultaneous training ensemble model of furnace temperature, and a regularized stochastic configuration network is used to establish a secondary training ensemble model of furnace temperature. The final output of the furnace temperature is obtained by the average value of the output of the above two ensemble models. Finally, a comparative experiment is carried out using the historical data of a waste incineration plant. The results show that the furnace temperature model established in this paper has advantages in accuracy, generalization ability, stability, and training efficiency. It can be applied to the field of furnace temperature prediction and control in the waste incineration process.
In this work, we tackle the problem of online adaptation for stereo depth estimation, that consists in continuously adapting a deep network to a target video recordedin an environment different from that of the source training set. To address this problem, we propose a novel Online Meta-Learning model with Adaption (OMLA). Our proposal is based on two main contributions. First, to reducethe domain-shift between source and target feature distributions we introduce an online feature alignment procedurederived from Batch Normalization. Second, we devise a meta-learning approach that exploits feature alignment forfaster convergence in an online learning setting. Additionally, we propose a meta-pre-training algorithm in order toobtain initial network weights on the source dataset whichfacilitate adaptation on future data streams. Experimentally, we show that both OMLA and meta-pre-training helpthe model to adapt faster to a new environment. Our proposal is evaluated on the wellestablished KITTI dataset,where we show that our online method is competitive withstate of the art algorithms trained in a batch setting.
基于深度学习的目标检测方法是目前计算机视觉领域的热点,在目标识别、跟踪等领域发挥了重要的作用.随着研究的深入开展,基于深度学习的目标检测方法主要分为有锚框的目标检测方法和无锚框的目标检测方法,其中无锚框的目标检测方法无需预定义大量锚框,具有更低的模型复杂度和更稳定的检测性能,是目前目标检测领域中较前沿的方法.在调研国内外相关文献的基础上,梳理基于无锚框的目标检测方法及各场景下的常用数据集,根据样本分配方式不同,分别从基于关键点组合、中心点回归、Transformer、锚框和无锚框融合等 4 个方面进行整体结构分析和总结,并结合COCO(Common objects in context)数据集上的性能指标进一步对比.在此基础上,介绍了无锚框目标检测方法在重叠目标、小目标和旋转目标等复杂场景情况下的应用,聚焦目标遮挡、尺寸过小和角度多等关键问题,综述现有方法的优缺点及难点.最后对无锚框目标检测方法中仍存在的问题进行总结并对未来发展的应用趋势进行展望.
相关滤波算法(Correlation filter,CF)已广泛应用于无人机目标跟踪.然而,受无人机(Unmanned aerial vehicle,UAV)平台本身计算性能的制约,现有的无人机相关滤波跟踪算法大都仅采用手工特征来描述目标的外观,难以获得目标的全面语义信息.并且这些跟踪算法仅能较好地进行光照条件良好场景下的跟踪,而在跟踪夜间场景下的目标时性能严重下降.此外,相关滤波跟踪器采用余弦窗口来抑制循环移位产生的边界效应,缩小了样本提取区域,产生了训练样本污染的问题,这不可避免地降低了跟踪器的性能.针对以上问题,提出全天实时多正则化相关滤波算法(All-day and real-time multi-regularized correlation filter,AMRCF)跟踪无人机目标.首先,引入一个自适应图像增强模块,在不影响图像各通道颜色比例的前提下,对获得的图像进行增强,以提高夜间目标跟踪性能.其次,引入一个轻量型的深度网络来提取目标的深度特征,并与手工特征一起来表示目标的语义信息.此外,在算法框架中嵌入高斯形状掩膜,在抑制边界效应的同时,有效避免训练样本污染.最后,在5个公开的无人机基准数据集上进行充分的实验.实验结果表明,所提出的算法与多个先进的相关滤波跟踪器相比,取得了有竞争力的结果,且算法的实时速度约为25 fps,能够胜任无人机的目标跟踪任务.
针对机器人摄影测量中离线规划受初始位姿标定影响的问题,提出融合初始位姿估计的机器人摄影测量系统视点规划方法.首先构建基于YOLO(You only look once)的深度学习网络估计被测对象3D包围盒,利用PNP(Perspect-ive-N-point)算法快速求解对象姿态;然后随机生成机器人无奇异无碰撞的视点,基于相机成像的2D-3D正逆性映射,根据深度原则计算每个视角下目标可见性矩阵;最后,引入熵权法,以最小化重建信息熵为目标建立优化模型,并基于旅行商问题(Travelling saleman problem,TSP)模型规划机器人路径.结果表明,利用深度学习估计的平移误差低于5 mm,角度误差低于2°.考虑熵权的视点规划方法提高了摄影测量质量,融合深度学习初始姿态的摄影测量系统提高了重建效率.利用本算法对典型零件进行摄影测量质量和效率的验证,均获得优异的位姿估计和重建效果.提出的算法适用于实际工程应用,尤其是快速稀疏摄影重建,促进了工业摄影测量速度与自动化程度提升.
对于部分可观测环境下的多智能体交流协作任务,现有研究大多只利用了当前时刻的网络隐藏层信息,限制了信息的来源.研究如何使用团队奖励训练一组独立的策略以及如何提升独立策略的协同表现,提出多智能体注意力意图交流算法(Multi-agent attentional intention and communication,MAAIC),增加了意图信息模块来扩大交流信息的来源,并且改善了交流模式.将智能体历史上表现最优的网络作为意图网络,且从中提取策略意图信息,按时间顺序保留成一个向量,最后结合注意力机制推断出更为有效的交流信息.在星际争霸环境中,通过实验对比分析,验证了该算法的有效性.
为提高多无人机(Unmanned aerial vehicles,UAV)协同轨迹规划(Cooperative trajectory planning,CTP)效率,在解耦序列凸优化(Sequential convex programming,SCP)方法基础上,提出一种高效求解凸优化子问题的定制内点法.首先引入松弛变量,构建子问题的等价描述形式,并推导该形式下的子问题最优性条件.然后在预测-校正原对偶内点法的框架下,构建一套高效求解最优性条件方程组的计算流程以降低子问题计算复杂度,并利用约束矩阵特征提出一种快速计算原对偶搜索方向的方法以提高规划效率.仿真结果表明,在解耦序列凸优化框架下,定制内点法可将协同轨迹规划耗时降低一个数量级,达到秒级.
针对一些智能优化算法缺乏完备数学物理理论基础的现状,利用优化问题和量子物理在概率意义上的相似性,建立优化问题的薛定谔方程,将优化问题转化为以目标函数为约束条件的基态波函数问题,同时利用波函数定义了算法的能量、隧道效应和熵,实现了以波函数为中心的优化问题量子模型.这一纲要利用了量子物理完备的理论框架,建立起了优化问题与量子理论广泛的内在联系.从量子物理的角度回答了优化问题解的概率描述,邻域采样函数的选择,算法演化的过程设计,多尺度过程的必要性等问题.智能优化算法的量子理论纲要可以作为研究与构造算法的理论工具,其有效性己得到初步验证.
在城市固体废弃物焚烧(Municipal solid waste incineration,MSWI)过程中,烟气含氧量是影响焚烧效果的重要工艺参数.由于固废焚烧过程的复杂性,在实际应用过程中,难以实现烟气含氧量的有效控制.面向城市固废焚烧过程烟气含氧量控制的实际需求,提出一种基于数据驱动的烟气含氧量自适应预测控制方法.首先,采用自适应模糊C均值(Fuzzy C-means,FCM)算法辅助确定径向基函数(Radial basis function,RBF)神经网络隐含层神经元个数及初始中心,建立基于FCM算法的径向基函数神经网络预测模型,并在控制过程中通过自适应更新策略在线调节预测模型参数;然后,利用梯度下降算法求解控制律,并基于李雅普诺夫理论分析了所提控制方法的稳定性;最后,基于城市固废焚烧厂实际数据,验证了所提控制方法的有效性.
针对黑猩猩优化算法(Chimp optimization algorithm,ChOA)存在收敛速度慢、精度低和易陷入局部最优值问题,提出一种融合多策略的黄金正弦黑猩猩优化算法(Multi-strategy golden sine chimp optimization algorithm,IChOA).引入Halton序列初始化种群,提高初始化种群的多样性,加快算法收敛,提高收敛精度;考虑到收敛因子和权重因子对于平衡算法勘探和开发能力的重要作用,引入改进的非线性收敛因子和自适应权重因子,平衡算法的搜索能力;结合黄金正弦算法相关思想,更新个体位置,提高算法对局部极值的处理能力.通过对23个基准测试函数的寻优对比分析和Wilcoxon秩和统计检验以及部分CEC2014测试函数寻优结果对比可知,改进的算法具有更好的鲁棒性;最后,通过2个实际工程优化问题的实验对比分析,进一步验证了 IChOA在处理现实优化问题上的优越性.
高炉料面视频关键帧是视频中的中心气流稳定、清晰、无炉料及粉尘遮挡且特征明显的图像序列,对于及时获取炉内运行状态、指导炉顶布料操作具有重要的意义.然而,由于高炉内部恶劣的冶炼环境及布料的周期性和间歇性等特征,料面视频存在信息冗余、图像质量参差不齐、状态多变等问题,无法直接用于分析处理.为了从大量高炉冶炼过程料面视频中自动准确筛选清晰稳定的料面图像,提出基于状态识别的高炉料面视频关键帧提取方法.首先,基于高温工业内窥镜采集高炉冶炼过程中的料面视频,并清晰完整给出料面反应新现象和形貌变化情况;然后,提取能够表征料面运动状态的显著性区域的特征点密集程度和像素位移特征,并提出基于局部密度极大值高斯混合模型(Local density maxima-based Gaussian mixture model,LDGMM)聚类的方法识别料面状态;最后,基于料面状态识别结果提取每个布料周期不同状态下的关键帧.实验结果表明,该方法能够准确识别料面状态并剔除料面视频冗余信息,能提取出不同状态下的料面视频关键帧,为优化炉顶布料操作提供指导.
针对基于Docker容器的分布式云计算下出现负载不均衡问题,有必要将较高负载服务器中的Docker容器进程迁移到其他相对空闲的服务器上.而传统的容器迁移算法忽视了容器本身的特征,从而导致在迁移过程中传输效率低下.基于此,利用第三方管理平台和数据预存储阈值机制,提出一种Docker容器动态迁移预存储算法PF-Docker.首先将Dock-er 容器内部进程运行相关文件和流动数据预存至云端存储器,然后通过预存储阈值机制减少流动数据的无效传输,最后在停机传输阶段将流动数据和冗余数据传输给目的服务器.实验表明,该方法在Docker容器迁移中能有效地降低迁移时间,减少数据传输量,提高容器的容错率.
作为一种不需要事先获得训练数据的机器学习方法,强化学习(Reinforcement learning,RL)在智能体与环境的不断交互过程中寻找最优策略,是解决序贯决策问题的一种重要方法.通过与深度学习(Deep learning,DL)结合,深度强化学习(Deep reinforcement learning,DRL)同时具备了强大的感知和决策能力,被广泛应用于多个领域来解决复杂的决策问题.异策略强化学习通过将交互经验进行存储和回放,将探索和利用分离开来,更易寻找到全局最优解.如何对经验进行合理高效的利用是提升异策略强化学习方法效率的关键.首先对强化学习的基本理论进行介绍;随后对同策略和异策略强化学习算法进行简要介绍;接着介绍经验回放(Experience replay,ER)问题的两种主流解决方案,包括经验利用和经验增广;最后对相关的研究工作进行总结和展望.
时间敏感网络(Time-sensitive networking,TSN)作为一种新兴工业通信技术,能够为工业控制业务提供高可靠及确定性时延保障.针对时间敏感网络在工业场景中广泛采用的时间感知整形(Time-aware shaper,TAS)机制,提出一种基于网络演算的时延上界分析模型,对多节点组网下端到端时延上界进行定量分析,用以评估门控(Gate control list,GCL)设置是否满足业务服务质量(Quality of service,QoS)需求,有助于简化多节点组网场景下门控设置复杂度.模型仿真部分对影响端到端时延的主要因素进行了对比分析,并通过OMNeT++实时仿真验证了所提出时延上界分析模型的有效性.
