基于置信规则库的工业控制系统安全状态评估

当前工业控制系统存在严重安全问题,针对现有工业控制系统安全状态评估模型存在的不足,提出一种基于置信规则库(BRB)专家系统的工业控制系统安全状态评估方法。该方法首先利用置信规则库专家系统将工业控制系统中定性知识与定量监测数据相结合。然后采用证据推理(ER)算法进行知识推理,并对所建立的BRB模型初始参数进行优化。最后以某燃气工业控制系统数据为例,实验结果表明,其评估精度高于SVM和BP神经网络模型。     

基于置信规则库的飞控系统故障诊断

针对传统飞控系统故障诊断中存在的因引入专家知识引起的主观偏差问题和使用数据驱动方法因数据量不足导致的过拟合问题,提出了基于置信规则库推理的飞控系统故障诊断。根据已有故障知识构建飞控系统故障诊断置信规则库,利用测试过程中获得的故障数据,以数值样本优化学习模型对置信规则库参数进行训练。实例表明,经少量样本训练后的置信规则库可以很好地解决初始置信规则库参数存在主观偏差的问题,经实验证明该方法能够实现高效可靠的飞控系统故障诊断。     

基于规则推理网络的分类模型

为了缓解神经网络的“黑盒子”机制引起的算法可解释性低的问题,基于使用证据推理算法的置信规则库推理方法(以下简称RIMER)提出了一个规则推理网络模型.该模型通过RIMER中的置信规则和推理机制提高网络的可解释性.首先证明了基于证据推理的推理函数是可偏导的,保证了算法的可行性;然后,给出了规则推理网络的网络框架和学习算法,利用RIMER中的推理过程作为规则推理网络的前馈过程,以保证网络的可解释性;使用梯度下降法调整规则库中的参数以建立更合理的置信规则库,为了降低学习复杂度,提出了“伪梯度”的概念;最后,通过分类对比实验,分析了所提算法在精确度和可解释性上的优势.实验结果表明,当训练数据集规模较小时,规则推理网络的表现良好,当训练数据规模扩大时,规则推理网络也能达到令人满意的结果.     

基于置信规则库的工业控制网络入侵检测

针对工业控制网络易遭受恶意攻击,本文提出了一种基于置信规则库的工业控制网络入侵检测方法。当置信规则库的前提属性数目过多时,置信规则库的规则条数呈指数级别增长,容易导致"组合爆炸"问题,本文提出利用线性组合方式构建置信规则库中的规则。本文还利用证据推理算法对置信规则库中的置信规则进行组合,并且优化置信规则库初始参数,提高了入侵检测的精确度。     

基于证据推理和置信规则库的装备寿命评估

针对航天产品试验样本少,寿命评估难的特点,结合产品在研制阶段多种工作环境的失效数据,提出了一种基于证据推理(evidential reasoning,ER)和置信规则库(belief-rule-base,BRB)进行装备寿命评估的新方法.首先,分析了模型的合理性并使用多维BRB模型将多种环境下的寿命数据折合为标准工作环境下的寿命数据,然后通过ER算法将折合后数据和实际工作环境数据进行融合.其次,详细说明了BRB--ER模型的推理过程和寿命评估的步骤.最后,采用某航天产品的失效数据对该方法进行了验证,并用已有的产品寿命的固定值进行BRB的参数更新.研究结果表明,在专家知识准确合理时,该模型能够准确地评估产品寿命,并可根据已有的产品的固定寿命进行训练,建立更加准确的寿命预测模型.     

数据缺失的扩展置信规则库推理方法

数据驱动的扩展置信规则库专家系统能够处理含有定量数据或定性知识的不确定性问题.该方法已被广泛地研究和应用,但仍缺乏在不完整数据问题上的研究.鉴于此,针对不完整数据集上的问题,提出一种新的扩展置信规则库专家系统推理方法.首先提出基于析取范式的扩展规则结构,并通过实验讨论了在新的规则结构下,置信规则前提属性参考值个数对推理...     

基于GDA的置信规则库参数训练的集成学习方法

目前对置信规则库(belief rule base,BRB)的研究主要针对单个BRB系统,然而单个BRB系统的推理性能不仅受参数取值的影响,而且当训练集分布不均衡或数据量较少时,容易导致参数训练不全面,从而使得推理结果所提供的决策信息存在局部性。通过引入Bagging算法和Ada Boost算法,分别与BRB相结合提出了基于梯度下降法(gradient descent algorithm,GDA)的置信规则库系统的集成学习方法,并分别应用于输油管道检漏、多峰函数的置信规则库训练,将多个BRB子系统集成,提高系统的推理性能。在实验中,以收敛精度和曲线拟合效果作为衡量指标来分析集成系统的性能,并将集成系统与其他单个BRB系统进行比较,实验结果表明BRB集成学习方法合理有效。     

基于置信规则库推理的二择众仓分类方法

针对线性组合方式所构建的置信规则库存在常常无法准确发挥前件属性权重的效能,且随着评价等级个数的增加,新激活权重公式往往会对结果造成不利影响的不足,本文在现有置信规则库推理分类算法的基础上,提出二择众仓决策法,以此改进置信规则库决策系统。首先仅设置两个规则的后件评价等级,对一个决策问题仅做出二择判定,即回答是与否;其次,设置多个置信规则库同时处理若干个子问题;最后通过众仓决策方式融合多个子问题的结果,进而解决最终的分类问题。实验结果表明,改进后的置信规则库推理分类方法可行有效。     

基于改进规则激活率的扩展置信规则库推理方法

数据驱动的扩展置信规则库系统,是在传统置信规则库的基础上利用关系数据来生成规则,使用该方法构建规则库简单有效。然而,该方法激活的规则存在不一致与不完整,并且该方法无法处理零激活的输入。鉴于此,本文提出基于改进规则激活率的扩展置信规则库方法,通过高斯核改进个体匹配度计算方法,权衡激活规则的一致性与完整性,并利用k近邻思想解决规则零激活问题。最后,本文选取非线性函数拟合实验和输油管道检漏实验来检验所提方法的效率和准确度。实验结果表明该方法既保证了扩展置信规则库系统的推理效率,也提高了推理结果的精度。     

基于置信规则库专家系统的发动机故障诊断

针对发动机故障原因和征兆之间存在的复杂非线性关系,利用RIMER(基于证据推理算法的置信规则库推理方法)对发动机进行故障诊断,克服了传统专家系统或神经网络技术只能单一利用专家知识或训练数据的缺点,将定性知识与定量数据有效结合,对发动机故障原因进行了研究,给维修人员提供了重要参考依据,仿真实验结果表明该方法可行有效.     

Milling Fault Detection Method Based on Fault Tree Analysis and Hierarchical Belief Rule Base

Expert knowledge is the key to modeling milling fault detection systems based on the belief rule base. The construction of an initial expert knowledge base seriously affects the accuracy and interpretability of the milling fault detection model. However, due to the complexity of the milling system structure and the uncertainty of the milling failure index, it is often impossible to construct model expert knowledge effectively. Therefore, a milling system fault detection method based on fault tree analysis and hierarchical BRB (FTBRB) is proposed. Firstly, the proposed method uses a fault tree and hierarchical BRB modeling. Through fault tree analysis (FTA), the logical correspondence between FTA and BRB is sorted out. This can effectively embed the FTA mechanism into the BRB expert knowledge base. The hierarchical BRB model is used to solve the problem of excessive indexes and avoid combinatorial explosion. Secondly, evidence reasoning (ER) is used to ensure the transparency of the model reasoning process. Thirdly, the projection covariance matrix adaptation evolutionary strategies (P-CMA-ES) is used to optimize the model. Finally, this paper verifies the validity model and the method's feasibility techniques for milling data sets.     

基于AE-LSTM的多目标硬盘故障预测方法

硬盘故障预测是在故障发生前发出预警,避免数据丢失或服务中断,提高数据中心的可靠性和安全性。然而,大多数故障预测模型将硬盘故障问题转化为二分类任务,忽略了硬盘故障是渐变过程的,并且缺乏故障诊断功能。因此,提出了一种基于AE-LSTM的硬盘故障预测框架,实现多目标任务：硬盘健康状态分级、硬盘剩余使用寿命预测、硬盘故障诊断。首先,采用回归决策树模型智能化对硬盘健康状态进行标记;其次,通过AE-LSTM模型提取鲁棒的隐藏变量,并构建剩余使用寿命预测模型和硬盘健康状态分级模块;最后,根据AE模块的输入输出差异进行硬盘故障诊断。在Backblaze公开数据集上,对比了RF、LSTM和AE-LSTM三种算法,实验结果证实了AE-LSTM算法在多目标硬盘故障预测中的有效性和优势。     

Characteristic Model-based Adaptive Fault Tolerant Control for Four-motor Synchronization Systems Considering Actuator Failure

This article proposes a characteristic model-based adaptive fault tolerant control scheme to deal with actuator failure in four-motor synchronization systems, which usually causes sudden inertia ratio change and backlash increase. Firstly, the characteristic modeling method is applied into servo system to obtain a simplified system model without losing high-order features. Also, this model could reflect real-time system status through three characteristic parameters. Secondly, a particle swarm optimization algorithm-based estimator is designed to identify characteristic parameters online. By this way, the characteristic model could react to inertia ratio changes quickly and eliminate its negative effect in signal tracking. Thirdly, an improved adaptive electric anti-backlash method is used to restrain backlash. Compared to regular anti-backlash technique, this adaptive one uses a neural network-based fault detector to monitor motors and adjust bias current according to different actuator status, even when one motor breaks down. With these three steps combined, a fast terminal sliding mode controller is finally designed as fault tolerant controller and the stability of this closed-loop system is guaranteed by Lyapunov stability theorem. At last, the simulation and experiment results prove the effectiveness of the proposed control scheme in system control and fault tolerance.

     

基于机器学习的列车设备故障预测模型研究

决策树作为机器学习和数据挖掘领域中广泛应用的预测模型,其输出结果易于理解和解释。针对高速铁路车载智能设备数量庞大的流数据且设备故障复杂和诊断效率低等问题,采用CVFDT决策树算法,通过对规范化的列控设备流数据进行机器学习,构建车载设备智能故障预测模型（低概率发生、高概率发生和已发生故障）,实现对设备潜在故障“事前排除”,提高故障分类精度、定位和诊断准确性,保障高速铁路运营安全和运输效率。     

深海载人潜水器推进器系统故障诊断的新型主元分析算法

针对"蛟龙号"深海载人潜水器多推进器系统的故障检测与快速定位难题,将基于信度分配的模糊小脑神经网络(credit assignment-based fuzzy cerebellar model articulation controller, FCA–CMAC)应用于主元分析模型,提出一种基于主元分析(principal component analysis, PCA)的深海载人潜水器推进器系统故障诊断模型.首先,应用推进器系统正常运行的历史电流样本数据,由主元分析模型得到各推进器的电流预测值.其次,计算出故障检测统计量均方预测误差(squared prediction error, SPE),根据SPE值是否跳变,判断推进器系统有无故障发生.通过分别重构各推进器电流信号的SPE值对故障推进器进行定位和隔离.最后,通过对实际海试数据进行仿真处理说明了该算法的可行性,并通过与多层前馈神经网络(back propagation, BP)和常规小脑神经网络(cerebellar model articulation control-ler, CMAC)神经网络进行比较,说明基于FCA–CMAC神经网络的主元分析模型的优越性.     

基于预测控制的高超声速飞行器容错控制器设计

针对存在升降舵面偏转角卡死故障的高超声速飞行器,提出一种基于预测控制的容错控制器设计方法.利用输入输出反馈线性化,对高超声速飞行器纵向模型进行变换;对于速度和高度的高阶导数以及故障项,设计扩张状态观测器在线观测;采用泰勒展开得到预测模型,建立连续预测控制器,分析证明闭环控制系统的稳定性和观测误差的有界性.仿真结果验证了所提方法的有效性.     

Dynamic modeling and fuzzy logic control of vibrations of a railway vehicle for different track irregularities

Safe and comfortable transportation of passengers and goods on railways can be achieved by solving the vibration problem. In this study, the dynamic modeling of the full railway vehicle is used to perform vibration analysis in order to observe displacements and accelerations. The full railway vehicle model consists of 54 degrees of freedom which are defined by differential equations. Additionally, wheel–rail contact problem (i.e. creepage factors and hertzian spring stiffness of rails) is analyzed by finite element method. Dynamic modeling and vibration analysis are carried out using Matlab–Simulink software. Using the developed model, the car body vibrations, caused by a lateral and two vertical sinusoidal track irregularities, are controlled by fuzzy logic controllers placed between the car body and bogies. The fuzzy logic algorithm herein is used for realizing the active control of car body vibrations. The simulations of vibration analysis are obtained in time and frequency domains and compared with passive controlled status. The robustness of the designed controller is verified by simulations, carried out for the cases of car body mass variations. The results show the effectiveness of the proposed algorithm.     

A method for predicting the network security situation based on hidden BRB model and revised CMA-ES algorithm

It is important to establish the forecasting model of the network security situation. But the network security situation cannot be observed directly and can only be measured by other observable data. In this paper the network security situation is considered as a hidden behavior. In order to predict the hidden behavior, some methods have been proposed. However, these methods cannot use the hybrid information that includes qualitative knowledge and quantitative data. As such, a forecasting model of network security situation is proposed on the basis of the hidden belief rule base (BRB) model when the inputs are multidimensional. The initial parameters of the hidden BRB model given by experts may be subjective and inaccurate. In order to train the parameters, a revised covariance matrix adaption evolution strategy (CMA-ES) algorithm is further developed by adding a modified operator. The revised CMA-ES algorithm can optimize the parameters of the hidden BRB model effectively. The case study shows that compared with other methods, the proposed hidden BRB model and the revised CMA-ES algorithm can predict the network security situation effectively to improve the forecasting precision by making full use of qualitative knowledge.     

基于文本挖掘的高速铁路动车组故障多级分类研究

针对高速铁路信号设备故障发生后记录的文本数据,提出基于文本挖掘方式的高速铁路信号设备故障多级分类模型研究。提出TF-IDF词汇权重与词汇字典结合的特征表示方法实现信号设备故障文本数据的特征提取。多级分类模型中,基于Stacking集成学习思想设计单层分类模型,将循环神经网络BiGRU和BiLSTM作为初级学习器,设计权重组合计算方法作为次级学习器,将多级分类任务分解为各层单分类任务,并采用K折交叉验证训练Stacking模型。采用高速铁路自开通至十年的信号转辙机故障数据,通过对故障原因文本数据的分析,实现故障部位和故障原因的二级分类,经过K=5次训练,BiGRU较BiLSTM各评价指标都较高,经实验BiGRU分配权重为0.7,BiLSTM权重为0.3,组合加权对两个网络的输出计算,准确率提高为0.8814,召回率提高为0.8642。实验表明多级分类模型能够有效提升信号设备故障多级分类任务的分类评价指标,并能够保证分类结果隶属关系的正确性。