使用GNN与RNN实现用户行为分析

随着国家高性能计算环境(CNGrid)各个节点产生日志数量不断增加,采用传统的人工方式进行用户行为分析已不能满足日常的分析需求.近年来,深度学习在入侵检测、图像识别、自然语言处理和恶意软件检测等与计算机科学相关的关键任务中取得了良好的效果.演示了如何将深度学习模型应用于用户行为分析.为此,在CNGrid中对用户行为进行...     

线性合成的双粒度 RNN 集成系统

针对脱机文字识别,提出了一种基于线性合成的双粒度递归神经网络(Recurrent neural net work, RNN)集成系统.首先,使用单词RNN对未知图 像进行识别;然后,依据识别结果进行字符分割,使用字符RNN对分割后的字符进行识别,并利用查表法计算字符的后验概率;最后,综合两个RNN的识别结果决定最终单词输出.在CAPTCHA识别 和手写识别上的实验结果证明了该系统的有效性.     

基于RNN的故障预测算法及在GIS上的应用

随着工业化程度的提高,设备的故障预测的重要性日趋提高。提出了一种基于循环神经网络（RNN）的故障预测算法,通过数据训练,充分发掘了RNN对时间序列数据的拟合能力。RNN故障预测模型由数据处理模块和神经网络识别模块组成。在数据处理模块中,采用数学函数分配的方法建立了RNN 模型的训练样本和测试样本。在神经网络识别模块中,针对当前故障预测技术中异常点难以确定的问题,应用了一种逐步逼近的神经网络训练方法。最后利用气体绝缘开关(GIS)故障数据对该算法进行了验证,结果表明,该方法可以在故障发生前检测到故障发生趋势,进而实现故障预测,并且能在逐步训练中确定异常点的位置。     

Application of an artificial immune system-based fuzzy neural network to a RFID-based positioning system

Due to the rapid development of globalization, which makes supply chain management more complicated, more companies are applying radio frequency identification (RFID), in warehouse management. The obvious advantages of RFID are its ability to scan at high-speed, its penetration and memory. In addition to recycling, use of a RFID system can also reduce business costs, by indentifying the position of goods and picking carts. This study proposes an artificial immune system (AIS)-based fuzzy neural network (FNN), to learn the relationship between the RFID signals and the picking cart’s position. Since the proposed network has the merits of both AIS and FNN, it is able to avoid falling into the local optimum and possesses a learning capability. The results of the evaluation of the model show that the proposed AIS-based FNN really can predict the picking cart position more precisely than conventional FNN and, unlike an artificial neural network, it is much easier to interpret the training results, since they are in the form of fuzzy IF–THEN rules.     

基于机器学习方法的高速信道建模研究

随着高速信道的传输速率变快,传输长度变长,结构复杂度变高,对信道进行建模也变得复杂与艰难.将目前比较火热的机器学习方法与高速信道结合起来,提出了一个新颖的方法.利用采集的大量模拟数据,采用深度神经网络DN N与循环神经网络RN N对信道建模,模型一旦训练成功,就可以通过该仿真模型预测输出信号的眼图,快速精准地对信号完整...     

基于云‒端融合的个性化推荐服务系统

主流个性化推荐服务系统通常利用部署在云端的模型进行推荐,因此需要将用户交互行为等隐私数据上传到云端,这会造成隐私泄露的隐患。为了保护用户隐私,可以在客户端处理用户敏感数据,然而,客户端存在通信瓶颈和计算资源瓶颈。针对上述挑战,设计了一个基于云?端融合的个性化推荐服务系统。该系统将传统的云端推荐模型拆分成用户表征模型和排序模型,在云端预训练用户表征模型后,将其部署到客户端,排序模型则部署到云端;同时,采用小规模的循环神经网络（RNN）抽取用户交互日志中的时序信息来训练用户表征,并通过Lasso算法对用户表征进行压缩,从而在降低云端和客户端之间的通信量以及客户端的计算开销的同时防止推荐准确率的下跌。基于RecSys Challenge 2015数据集进行了实验,结果表明,所设计系统的推荐准确率和GRU4REC模型相当,而压缩后的用户表征体积仅为压缩前的34.8%,计算开销较低。     

Adaptive RTRL based neurocontroller for damping subsynchronous oscillations using TCSC

Modern interconnected electrical power systems are complex and require perfect planning, design and operation. Hence the recent trends towards restructuring and deregulation of electric power supply has put great emphasis on the system operation and control. Flexible AC transmission system (FACTS) devices such as thyristor controlled series capacitor (TCSC) are capable of controlling power flow, improving transient stability and mitigating subsynchronous resonance (SSR). In this paper an adaptive neurocontroller is designed for controlling the firing angle of TCSC to damp subsynchronous oscillations. This control scheme is suitable for non-linear system control, where the exact linearised mathematical model of the system is not required. The proposed controller design is based on real time recurrent learning (RTRL) algorithm in which the neural network (NN) is trained in real time. This control scheme requires two sets of neural networks. The first set is a recurrent neural network (RNN) which is a fully connected dynamic neural network with all the system outputs fed back to the input through a delay. This neural network acts as a neuroidentifier to provide a dynamic model of the system to evaluate and update the weights connected to the neurons. The second set of neural network is the neurocontroller which is used to generate the required control signals to the thyristors in TCSC. This is a single layer neural network. Performance of the system with proposed neurocontroller is compared with two linearised controllers, a conventional controller and with a discrete linear quadratic Gaussian (DLQG) compensator which is an optimal controller. The linear controllers are designed based on a linearised model of the IEEE first benchmark system for SSR studies in which a modular high bandwidth (six-samples per cycle) linear time-invariant discrete model of TCSC is interfaced with the rest of the system. In the proposed controller, since the response time is highly dependent on the number of states of the system, it is often desirable to approximate the system by its reduced model. By using standard Hankels norm approximation technique, the system order is reduced from 27 to 11th order by retaining the dominant dynamic characteristics of the system. To validate the proposed controller, computer simulation using MATLAB is performed and the simulation studies show that this controller can provide simultaneous damping of swing mode as well as torsional mode oscillations, which is difficult with a conventional controller. Moreover the fast response of the system can be used for real-time applications. The performance of the controller is tested for different operating conditions.     

Back propagation genetic and recurrent neural network applications in modelling and analysis of squeeze casting process

Today, in competitive manufacturing environment reducing casting defects with improved mechanical properties is of industrial relevance. This led the present work to deal with developing the input-output relationship in squeeze casting process utilizing the neural network based forward and reverse mapping. Forward mapping is aimed to predict the casting quality (such as density, hardness and secondary dendrite arm spacing) for the known combination of casting variables (that is, squeeze pressure, pressure duration, die and pouring temperature). Conversely, attempt is also made to determine the appropriate set of casting variables for the required casting quality (that is, reverse mapping). Forward and reverse mapping tasks are carried out utilizing back propagation, recurrent and genetic algorithm tuned neural networks. Parameter study has been conducted to adjust and optimize the neural network parameters utilizing the batch mode of training. Since, batch mode of training requires huge data, the training data is generated artificially using response equations. Furthermore, neural network prediction performances are compared among themselves (reverse mapping) and with those of statistical regression models (forward mapping) with the help of test cases. The results shown all developed neural network models in both forward and reverse mappings are capable of making effective predictions. The results obtained will help the foundry personnel to automate and précised control of squeeze casting process.     

基于DFNN的动态矩阵网络控制系统的应用研究

针对网络控制系统中的随机时延,提出一种基于动态模糊神经网络的动态矩阵网络控制系统。利用动态模糊神经网络的特点,提高系统动态响应性能。在以太网的网络环境下,通过实验仿真结果表明,该方法响应快,提高了系统的跟踪精度,具有更理想的控制效果。     

基于FNN的覆冰机器人越障机械臂轨迹跟踪控制

覆冰机器人除冰时要跨越各种障碍物。采用卡尔曼滤波学习算法,将自适应模糊神经网络控制器用于覆冰机器人越障时的机械臂轨迹跟踪控制,解决了BP算法实时性差的问题。经过仿真实验论证,该方法对覆冰机器人越障时的机械臂轨迹跟踪控制具有很好的效果,表明控制策略和理论分析的可行性。     

基于迁移学习的知识图谱问答语义匹配模型

针对单一事实类问答系统中问句和关系的语义匹配在小规模标注样本中难以获得较高准确率的问题,提出一种基于循环神经网络（RNN）的迁移学习模型。首先,使用基于RNN的序列到序列无监督学习算法,通过序列重构的方式在大量无标注样本中学习问句的语义空间分布,即词向量和RNN;然后,通过给神经网络参数赋值的方式,使用此语义空间分布作为有监督语义匹配算法的参数;最后,通过使用问句特征和关系特征计算内积的方式,在有标注样本中训练并生成语义匹配模型。实验结果表明,在有标注数据量较少而无标注数据量较大的环境下,与有监督学习方法Embed-AVG和RNNrandom相比,所提模型的语义匹配准确率分别平均提高5.6和8.8个百分点。所提模型通过预学习大量无标注样本的语义空间分布可以明显提高在小规模标注样本环境下的语义匹配准确率。     

基于多特征提取的图像语义描述算法

针对图像语义描述方法中存在的图像特征信息提取不完全以及循环神经网络(RNN)产生的梯度消失问题,提出了一种基于多特征提取的图像语义描述算法.所构建模型由三个部分组成:卷积神经网络(CNN)用于图像特征提取,属性提取模型(ATT)用于图像属性提取,而双向长短时记忆(Bi-LSTM)网络用于单词预测.该模型通过提取图像属性...     

基于胶囊网络的方面级跨领域情感分析

在跨领域情感分析任务中,目标领域带标签样本严重不足,并且不同领域间的特征分布差异较大,特征所表达的情感极性也有很大差别,这些问题都导致了分类准确率较低。针对以上问题,提出一种基于胶囊网络的方面级跨领域情感分析方法。首先,通过BERT预训练模型获取文本的特征表示;其次,针对细粒度的方面级情感特征,采用循环神经网络（RNN）将上下文特征与方面特征进行融合;然后,使用胶囊网络配合动态路由来区分重叠特征,并构建基于胶囊网络的情感分类模型;最后,利用目标领域的少量数据对模型进行微调来实现跨领域迁移学习。所提方法在中文数据集上的最优的F1值达到95.7%,英文数据集上的最优的F1值达到了91.8%,有效解决了训练样本不足造成的准确率低的问题。     

模糊神经网络PID在电动舵机控制中的应用

研究电动舵机控制系统优化问题。针对传统控制器响应速度慢,由于系统本身是多变量非线性的复杂系统,存在时滞问题,系统参数不易整定,为了优化电动舵机控制系统的快速性性能,设计了一种改进的模糊神经网络PID控制器,提出了分离学习算法,利用自组织学习整定隶属度函数参数和误差反向传播学习整定加权系数。将算法用于电动舵机控制,实现对舵机以及导弹姿态的快速控制。仿真结果表明,改进控制器的响应时间达到4.8ms,优于传统的PID控制器和模糊PID控制器,为电动舵机控制系统快速性、高精度的设计提供了依据。     

基于PID神经元网络和内模控制的拥塞控制算法*

针对网络系统的大时滞和非线性特性,设计了一种新的拥塞控制算法,将PID神经元网络与内模控制相结合应用于主动队列管理中,并使用Lyapunov理论证明了此算法的稳定性。NS仿真结果表明,这种算法的稳态和瞬态性能都优于PID算法,并且在参数变化和负载扰动时具有很强的鲁棒性。     

基于栈式循环神经网络的血液动力学状态估计方法

利用fMRI数据准确地估计血液动力学状态, 能得到一种更接近神经元层面的大脑活动的客观表示, 这将促进人们对大脑运行机理的深刻理解, 推动脑认知的进一步发展.迄今为止, 人们已经提出了许多血液动力学状态估计方法.然而, 这些方法大都只考虑了相邻时刻血液动力学状态之间的关系, 忽视了更深层次的时序特征.而对模型参数先验信息的需求也使一些方法在实际应用中受到了限制.为此, 本文提出了一种基于循环神经网络的血液动力学状态估计新方法.首先, 利用血液动力学模型中非线性函数的反函数建立BOLD信号与血液动力学状态之间的映射关系, 并构建模型的反演过程.然后, 采用一种堆叠三个RNN模块的栈式神经网络结构来拟合这种映射关系, 使其能够以BOLD信号作为输入, 得到血液动力学状态的估计值.最后, 在仿真数据上验证新方法的性能.实验结果表明:与一些代表算法相比, 新方法能够更合理地提取fMRI数据中的时间特性, 有效地拟合BOLD信号与血液动力学状态之间的动态非线性关系.     

Inventory lot-sizing with supplier selection using hybrid intelligent algorithm

In supply chain management (SCM), multi-product and multi-period models are usually used to select the suppliers. In the real world of SCM, however, there are normally several echelons which need to be integrated into inventory management. This paper presents a hybrid intelligent algorithm, based on the push SCM, which uses a fuzzy neural network and a genetic algorithm to forecast the rate of demand, determine the material planning and select the optimal supplier. We test the proposed algorithm in a case study conducted in Iran.     

一种利用函数链神经网络的传感器建模新方法

讨论基于函数链神经网络 (FLNN)的传感器建模新方法 ,其结构简单、使用灵活、建模容易 ,易于实时硬件实现。两个算例说明网络的训练和非线性逼近方法 ,显示出网络的自适应能力、学习能力 ,基于FLNN的传感器模型可同时实现温度补偿和非线性校正。实际上 ,利用这种模型可以跟踪补偿环境改变引起的传感器特性的各种变化 ,在测控系统中具有良好的应用前景。     

基于自适应AGBFN的不确定非线性系统的跟踪控制

针对一类具有未知不确定性的非线性系统,提出了一种基于观测器的自适应不对称高斯基函数网络(AGBFN)跟踪控制方案.当系统只有输出可以测量时,通过设计观测器对其进行在线状态估计,进而构造反馈控制律和自适应控制律.所提出的完全自适应AGBFN,可以在线更新网络所有参数,克服了传统RBF网络对称性约束,提高了网络的适应性和学习能力,可以有效地对消系统未知不确定项的影响.证明了闭环系统所有误差信号最终一致有界,且系统输出较好地跟踪参考模型输出.仿真结果表明了所提出方法的有效性.     

Neuro-fuzzy control of antilock braking system using sliding mode incremental learning algorithm

A neuro-fuzzy adaptive control approach for nonlinear dynamical systems, coupled with unknown dynamics, modeling errors, and various sorts of disturbances, is proposed and used to design a wheel slip regulating controller. The implemented control structure consists of a conventional controller and a neuro-fuzzy network-based feedback controller. The former is provided both to guarantee global asymptotic stability in compact space and as an inverse reference model of the response of the controlled system. Its output is used as an error signal by an incremental learning algorithm to update the parameters of the neuro-fuzzy controller. In this way the latter is able to gradually replace the conventional controller from the control of the system. The proposed new learning algorithm makes direct use of the variable structure systems theory and establishes a sliding motion in terms of the neuro-fuzzy controller parameters, leading the learning error toward zero. In the simulations and in the experimental studies, it has been tested on the control of antilock breaking system model and the analytical claims have been justified under the existence of uncertainty and large nonzero initial errors.