一种改进的专家系统自学习算法的研究

基于证据推理的置信规则库推理方法(RIMER)已被提出,由此产生了一种新型专家系统-RIMER专家系统.该专家系统的学习训练模型是一个带有线性约束的复杂非线性优化模型,传统优化方法求解该类模型有一定困难和局限性.本文结合梯度法和二分法提出一种新的优化算法实现了RIMER专家系统的自学习.采用该算法对一个实例进行了训练,训练结果令人满意.训练实例表明新算法具有简单、速度快、收敛精度高等特点.     

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.     

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。     

基于深度置信网络的广告点击率预估的优化

随着互联网广告的飞速发展,如何预测目标用户对互联网广告的点击率（click-through rate,简称CTR）,成为精确广告推荐投放的关键技术,并成为计算广告领域的研究热点和深度神经网络的应用热点.为了提高广告点击率预估的精确度,提出了基于深度置信网络的广告点击率预估模型,并通过基于Kaggle数据挖掘平台数据集的1 000万条随机数据的实验,研究不同的隐藏层层数和隐含节点数目对预测结果的影响.为了解决深度置信网络在数据规模较大的工业界解决方案中的训练效率问题,通过实验证明：广告点击率预估中,深度置信网络的损失函数存在大量的驻点,并且这些驻点对网络训练效率有极大的影响.为了提高模型效率,从发掘网络损失函数特性入手,进一步提出了基于随机梯度下降算法和改进型粒子群算法的融合算法,以优化网络训练.融合算法在迭代步长小于阈值时可以跳出驻点平面,继续正常迭代.实验结果表明,与传统的基于梯度提升决策树和逻辑回归的广告点击率预估模型以及模糊深度神经网络模型相比,基于深度置信网络的预估模型具有更好的预估精度,在均方误差、曲线下面积和对数损失函数指标上分别提升2.39%,9.70%,2.46%和1.24%,7.61%,1.30%;使用融合方法训练深度置信网络,训练效率提高30%~70%.     

抗推理攻击的隐私增强联邦学习算法

联邦学习在保证各分布式客户端训练数据不出本地的情况下，由中心服务器收集梯度协同训练全局网络模型，具有良好的性能与隐私保护优势。但研究表明，联邦学习存在梯度传递引起的数据隐私泄漏问题。针对现有安全联邦学习算法存在的模型学习效果差、计算开销大和防御攻击种类单一等问题，提出了一种抗推理攻击的隐私增强联邦学习算法。首先，构建了逆推得到的训练数据与训练数据距离最大化的优化问题，基于拟牛顿法求解该优化问题，获得具有抗推理攻击能力的新特征。其次，利用新特征生成梯度实现梯度重构，基于重构后的梯度更新网络模型参数，可提升网络模型的隐私保护能力。最后，仿真结果表明所提算法能够同时抵御两类推理攻击，并且相较于其他安全方案，所提算法在保护效果与收敛速度上更具优势。     

基于改进置信规则库推理的分类方法

通过引入置信规则库的线性组合方式,设定规则数等于分类数及改进个体匹配度的计算方法,提出了基于置信规则库推理的分类方法。比较传统的置信规则库推理方法,新方法中规则数的设置不依赖于问题的前件属性数量或候选值数量,仅与问题的分类数有关,保证了方法对于复杂问题的适用性。实验中,通过差分进化算法对置信规则库的规则权重、前件属性权重、属性候选值和评价等级的置信度进行参数学习,得到最优的参数组合。对3个常用的公共分类数据集进行测试,均获得理想的分类准确率,表明新分类方法合理有效。     

出租车乘车概率预测的置信规则库推理方法

出租车乘车概率预测中存在数据量级大,底层属性类型多,预测信息不确定的问题。鉴于此,整合大规模轨迹数据范畴中现有的挖掘算法对出租车GPS数据和路网数据进行离线处理;将多类型的不确定性数据转换为具有置信结构的规则形式,并以此构建置信规则库;通过置信规则库推理方法（belief rule-base infer-ence methodology using evidential reasoning,RIMER）在线预测路网道路上各个地点的乘车概率。以北京市2012年11月某天的出租车GPS数据为例说明该在线预测方法的应用。实验结果表明,该预测方法具有较高的实时性和准确性。     

基于AR模型的置信规则库结构识别算法

针对以置信规则推理作为系统控制器的应用,传统的置信K均值聚类算法往往不能充分利用数据中时间上的动态关联信息。因此,在模糊聚类算法的基础上引入自回归(AR)模型,将集约生产计划中的需求数据作为一组时间序列进行动态的聚类分析。该算法不仅可以充分利用集约生产计划中的需求数据的内部自相关性,而且可以进一步利用隶属度函数对AR模型的预测过程进行模糊化调整,从而得到更为理想的置信规则库结构,提高推理与决策的精度。     

Belief rule-base inference methodology using the evidential reasoning Approach-RIMER

In this paper, a generic rule-base inference methodology using the evidential reasoning (RIMER) approach is proposed. Existing knowledge-base structures are first examined, and knowledge representation schemes under uncertainty are then briefly analyzed. Based on this analysis, a new knowledge representation scheme in a rule base is proposed using a belief structure. In this scheme, a rule base is designed with belief degrees embedded in all possible consequents of a rule. Such a rule base is capable of capturing vagueness, incompleteness, and nonlinear causal relationships, while traditional if-then rules can be represented as a special case. Other knowledge representation parameters such as the weights of both attributes and rules are also investigated in the scheme. In an established rule base, an input to an antecedent attribute is transformed into a belief distribution. Subsequently, inference in such a rule base is implemented using the evidential reasoning (ER) approach. The scheme is further extended to inference in hierarchical rule bases. A numerical study is provided to illustrate the potential applications of the proposed methodology.     

Iterative learning belief rule-base inference methodology using evidential reasoning for delayed coking unit

The belief rule-base inference methodology using evidential reasoning (RIMER) approach has been proved to be an effective extension of traditional rule-based expert systems and a powerful tool for representing more complicated causal relationships using different types of information with uncertainties. With a predetermined structure of the initial belief rule-base (BRB), the RIMER approach requires the assignment of some system parameters including rule weights, attribute weights, and belief degrees using experts’ knowledge. Although some updating algorithms were proposed to solve this problem, it is still difficult to find an optimal compact BRB. In this paper, a novel updating algorithm is proposed based on iterative learning strategy for delayed coking unit (DCU), which contains both continuous and discrete characteristics. Daily DCU operations under different conditions are modeled by a BRB, which is then updated using iterative learning methodology, based on a novel statistical utility for every belief rule. Compared with the other learning algorithms, our methodology can lead to a more optimal compact final BRB. With the help of this expert system, a feedforward compensation strategy is introduced to eliminate the disturbance caused by the drum-switching operations. The advantages of this approach are demonstrated on the UniSim? Operations Suite platform through the developed DCU operation expert system modeled and optimized from a real oil refinery.     

A neural fuzzy system with fuzzy supervised learning

A neural fuzzy system learning with fuzzy training data (fuzzy if-then rules) is proposed in this paper. This system is able to process and learn numerical information as well as linguistic information. At first, we propose a five-layered neural network for the connectionist realization of a fuzzy inference system. The connectionist structure can house fuzzy logic rules and membership functions for fuzzy inference. We use alpha-level sets of fuzzy numbers to represent linguistic information. The inputs, outputs, and weights of the proposed network can be fuzzy numbers of any shape. Furthermore, they can be hybrid of fuzzy numbers and numerical numbers through the use of fuzzy singletons. Based on interval arithmetics, a fuzzy supervised learning algorithm is developed for the proposed system. It extends the normal supervised learning techniques to the learning problems where only linguistic teaching signals are available. The fuzzy supervised learning scheme can train the proposed system with desired fuzzy input-output pairs which are fuzzy numbers instead of the normal numerical values. With fuzzy supervised learning, the proposed system can be used for rule base concentration to reduce the number of rules in a fuzzy rule base. Simulation results are presented to illustrate the performance and applicability of the proposed system.     

Fuzzy rule weight modification with particle swarm optimisation

The most challenging problem in developing fuzzy rule-based classification systems is the construction of a fuzzy rule base for the target problem. In many practical applications, fuzzy sets that are of particular linguistic meanings, are often predefined by domain experts and required to be maintained in order to ensure interpretability of any subsequent inference results. However, learning fuzzy rules using fixed fuzzy quantity space without any qualification will restrict the accuracy of the resulting rules. Fortunately, adjusting the weights of fuzzy rules can help improve classification accuracy without degrading the interpretability. There have been different proposals for fuzzy rule weight tuning through the use of various heuristics with limited success. This paper proposes an alternative approach using Particle Swarm Optimisation in the search of a set of optimal rule weights, entailing high classification accuracy. Systematic experimental studies are carried out using common benchmark data sets, in comparison to popular rule based learning classifiers. The results demonstrate that the proposed approach can boost classification performance, especially when the size of the initially built rule base is relatively small, and is competitive to popular rule-based learning classifiers.     

Applying a belief rule-base inference methodology to a guideline-based clinical decision support system

Abstract: A critical issue in the clinical decision support system (CDSS) research area is how to represent and reason with both uncertain medical domain knowledge and clinical symptoms to arrive at accurate conclusions. Although a number of methods and tools have been developed in the past two decades for modelling clinical guidelines, few of those modelling methods have capabilities of handling the uncertainties that exist in almost every stage of a clinical decision-making process. This paper describes how to apply a recently developed generic rule-base inference methodology using the evidential reasoning approach (RIMER) to model clinical guidelines and the clinical inference process in a CDSS. In RIMER, a rule base is designed with belief degrees embedded in all possible consequents of a rule. Such a rule base is capable of capturing vagueness, incompleteness and non-linear causal relationships, while traditional IF–THEN rules can be represented as a special case. Inference in such a rule base is implemented using the evidential reasoning approach which has the capability of handling different types and degrees of uncertainty in both medical domain knowledge and clinical symptoms. A case study demonstrates that employing RIMER in developing a guideline-based CDSS is a valid novel approach.     

Optimization Models for Training Belief-Rule-Based Systems

A belief rule-base inference methodology using the evidential reasoning approach (RIMER) has been developed recently, where a new belief rule representation scheme is proposed to extend traditional IF-THEN rules. The belief rule expression matrix in RIMER provides a compact framework for representing expert knowledge. However, it is difficult to accurately determine the parameters of a belief rule base (BRB) entirely subjectively, particularly, for a large-scale BRB with hundreds or even thousands of rules. In addition, a change in rule weight or attribute weight may lead to changes in the performance of a BRB. As such, there is a need to develop a supporting mechanism that can be used to train, in a locally optimal way, a BRB that is initially built using expert knowledge. In this paper, several new optimization models for locally training a BRB are developed. The new models are either single- or multiple-objective nonlinear optimization problems. The main feature of these new models is that only partial input and output information is required, which can be either incomplete or vague, either numerical or judgmental, or mixed. The models can be used to fine tune a BRB whose internal structure is initially decided by experts' domain-specific knowledge or common sense judgments. As such, a wide range of knowledge representation schemes can be handled, thereby facilitating the construction of various types of BRB systems. Conclusions drawn from such a trained BRB with partially built-in expert knowledge can simulate real situations in a meaningful, consistent, and locally optimal way. A numerical study for a hierarchical rule base is examined to demonstrate how the new models can be implemented as well as their potential applications.     

应用模糊神经网络预测油田产量

为了研究受多变量、时变和不确定因素影响的油田产量预测问题,将模糊逻辑推理技术与人工神经网络相结合,构建具有模糊逻辑推理和学习功能的模糊神经网络(FNN)系统。该系统基于现有的油田开发历史数据,建立相应的规则集,使用神经网络的训练方法(如梯度下降学习算法),在训练过程中调整参数,并自适应增加规则,以使系统的输出最佳地逼近于目标样本。通过对某油田的实际开发历史数据的拟合与测试,结果表明该模糊神经网络能够较精确地预测未来的油产量,与常规的BP神经网络相比,其预测精度更高、训练速度更快。因此,基于模糊神经网络(FNN)的油田产量预测方法研究具有较好的实际应用价值。     

Online updating belief rule based system for pipeline leak detection under expert intervention

A belief rule base inference methodology using the evidential reasoning approach (RIMER) has been developed recently, where a new belief rule base (BRB) is proposed to extend traditional IF-THEN rules and can capture more complicated causal relationships using different types of information with uncertainties, but these models are trained off-line and it is very expensive to train and re-train them. As such, recursive algorithms have been developed to update the BRB systems online and their calculation speed is very high, which is very important, particularly for the systems that have a high level of real-time requirement. The optimization models and recursive algorithms have been used for pipeline leak detection. However, because the proposed algorithms are both locally optimal and there may exist some noise in the real engineering systems, the trained or updated BRB may violate some certain running patterns that the pipeline leak should follow. These patterns can be determined by human experts according to some basic physical principles and the historical information. Therefore, this paper describes under expert intervention, how the recursive algorithm update the BRB system so that the updated BRB cannot only be used for pipeline leak detection but also satisfy the given patterns. Pipeline operations under different conditions are modeled by a BRB using expert knowledge, which is then updated and fine tuned using the proposed recursive algorithm and pipeline operating data, and validated by testing data. All training and testing data are collected from a real pipeline. The study demonstrates that under expert intervention, the BRB expert system is flexible, can be automatically tuned to represent complicated expert systems, and may be applied widely in engineering. It is also demonstrated that compared with other methods such as fuzzy neural networks (FNNs), the RIMER has a special characteristic of allowing direct intervention of human experts in deciding the internal structure and the parameters of a BRB expert system.     

Research and Design of a Fuzzy Neural Expert System

We have developed a fuzzy neural expert system that has the precision and learning ability of a neural network.Knowledge is acquired from domain experts as fuzzy rules and membership functions.Then,they are converted into a neural network which implements fuzzy inference without rule matching.The neural network is applied to problem-solving and learns from the data obtained during operation to enhance the accuracy.The learning ability of the neural network makes it easy to modify the membership functions defined by domain experts.Also,by modifying the weights of neural networks adaptively,the problem of belief propagation in conventional expert systems can be solved easily.Converting the neural network back into fuzzy rules and membership functions helps explain the inner representation and operation of the neural network.     

A methodology to generate a belief rule base for customer perception risk analysis in new product development

New product development (NPD) is crucial for a company’s success in a competitive market. Meanwhile, NPD is a process associated with great complexity and high risk. To ensure its smooth operation, risks involved in an NPD process need to be analyzed in a proper way. In this paper, a novel method is proposed to generate a belief rule base (BRB), which is the basis of the Belief Rule-Base Inference Methodology using the Evidential Reasoning (RIMER). Due to its capability in dealing with complex reasoning problems under uncertainty, RIMER is then applied to assess customer perception risk (CPR) in an NPD process. To test and validate the method proposed in this paper, a case study of an “Interactive Doll” is conducted at the end of the paper.