基于区间灰作用量的GM(1,1)均值差分模型解的非唯一性研究

GM(1,1)是结构信息不完全的灰色预测模型，但当前其模拟及预测结果的实数形式不满足灰色理论解的非唯一性原理。文章从GM(1,1)网络模型出发，分析了灰作用量的背景与内涵，还原了影响因素不确定条件下灰作用量的区间灰数形式，构建了具有非唯一解的新型GM(1,1)均值差分模型。新模型具有更加完善的体系结构，同时能实现对传统 GM(1,1)均值差分模型的完全兼容。应用新模型对我国电力能源消费量进行建模，结果显示其建模结果的合理性优于传统 GM(1,1)模型。本研究成果对丰富灰色预测模型理论框架、完善灰色预测模型结构体系具有积极意义。     

基于新改进GM(1,1)模型的中国企业债券融资发展预测

我国企业债券市场明显滞后于整个资本市场的发展,加快发展企业债券市场的呼声日高。但是去除企业债券发展的束缚是一个渐进的过程。在此背景下,本文对我国企业债券融资发展进行定量预测。考虑到影响我国企业债券发展的因素较多且不确定,笔者采用灰色系统GM(1,1)模型进行预测。GM(1,1)模型是有偏差的灰指数模型,其精度取决于背景值的构造形式和初始条件的选取。已有的研究文献均是从一个侧面单独改进GM(1,1)模型,这里,笔者提出一种同时优化背景值和初始条件的新GM(1,1)模型。笔者发现新优化GM(1,1)模型比单独优化背景值或单独优化初始条件有更高的模拟精度。在此基础上,利用新改进GM(1,1)模型对我国2010年之前的企业债券余额进行了预测。     

惯性灰色模型及其应用性质研究

研究灰色预测模型建模的演化过程，可以更好地了解模型的本质特征和状态变化。惯性灰色模型主要研究灰色预测模型建模的演化过程，了解系统变化状态。本文根据数据的力学特性，利用矩阵分析方法研究惯性灰色模型的建模步骤，简化文献［1］中惯性模型的结构参数和分量参数形式，总结求解各种数据序列的力学变换式，获取各种惯性灰色模型的建模机理。最后通过实例研究系统状态的演变过程，将惯性灰色GM(1,1)模型应用到交通流状态的判定中，得到三相交通流与三种惯性灰色GM(1,1)模型的对应关系。利用三种惯性模型模拟效果来准确判断交通流的状态，揭示交通系统实时特性，为交通规划、控制和优化提供可靠的理论依据。     

提高灰色GM(1,1)模型精度的微粒群方法

改变背景值插值系数和边值是提高灰色GM(1,1)模型精度的途径之一。对于满足灰色GM(1,1)建模条件的序列,利用微粒群算法,给出了通过优化背景值插值系数和边值提高灰色GM(1,1)精度的新方法。给出了计算实例。计算表明,应用此方法可以提高灰色GM(1,1)模型的精度。     

基于初始条件优化的GM(1,1)幂模型及其应用

针对GM（1,1）幂模型幂指数和初始条件优化问题,提出了一种基于初始条件和幂指数协同优化的方法。根据新信息优先原理,通过引入权重信息控制函数优化初始条件,表现新旧信息在初始条件构建中作用大小的变化规律,最大限度提取小样本序列中的有效信息,反应新旧信息共同对系统趋势变化的影响;以平均相对误差最小化为目标,参数间约束关系作为条件,构建非线性优化模型,实现GM（1,1）幂模型的幂指数和初始条件协同优化。最后,通过我国网络购物用户规模预测实例研究表明,优化的模型实现模型平均相对误差在理论上的最小化,其建模效果要优于其他对比模型,并将其用于2016-2020年网购用户规模预测,表明本文模型的实用性和有效性。     

灰色组合预测模型及其应用

针对传统回归模型需要的数据量大且建模复杂等缺陷,提出了一种基于灰色关联度和GM(1,1)的灰色组合预测模型;从灰色关联度的视角寻找数据之间的依赖关系,运用GM(1,1)模型预测数据关系的未来发展趋势,进而建立因变量的预测模型;模型体现了回归分析基于事物因果关系的建模思想,同时又具有灰色理论小样本建模的特点;应用该模型对我国2007和2008年度的GDP进行预测,预测结果表明了该模型的有效性及实用性。     

一类调节强度可变的弱化缓冲算子及其应用研究

在灰色系统缓冲算子公理体系下,根据灰色系统的“新信息优先原理”,构造了一类调节强度可变的弱化缓冲算子,并对其特性进行了研究。针对不同的建模背景,通过调整缓冲算子的参数,可以生成具有不同弱化效果的序列。最后,利用本文提出的缓冲算子,对两个实际案例进行了建模计算,计算结果表明,这类新的弱化缓冲算子能够有效提高GM（1,1）模型的预测精度,解决定量预测结果与定性分析结论不符的问题。     

非等间距GM(1,1)模型背景值的优化

基于背景值是影响灰色建模精度的重要因素之一,本文对非等间距GM(1,1)模型中的背景值构造进行了研究,根据灰色模型的指数特性和积分特点,利用非齐次指数函数来拟合一次累加生成序列,提出了一种重构非等间距GM(1,1)模型背景值的方法。实例表明利用优化的背景值计算公式建立的非等间距GM(1,1)模型显著地改善了模拟和预测精度。该背景值不仅适合于非等间距建模,也适合于等间距建模,具有精度高、适用性强的特点。     

优化离散灰色幂模型及其应用

考虑已有的灰色预测模型主要能对指数型发展系统或幂函数型发展系统进行模拟预测,本文构建了一种不仅能够模拟指数型和幂函数型的发展系统,并且能够体现出二者之间的相互作用关系的离散灰色幂模型;并针对初始条件对离散灰色幂模型模拟精度的影响,首先给出了离散灰色幂模型的建模步骤,然后以平均相对误差最小化为目标、参数之间的关系为约束条件,构建了离散灰色幂模型初始条件的优化模型,实现对离散灰色幂模型初始条件的优化。结果表明,优化的离散灰色幂模型使得平均相对误差在理论上达到了最小化,其模拟精度和预测精度都高于传统模型。最后,通过中国网络购物人数数据预测和仿真数据分析,说明了本文优化方法的有效性和适用性。     

基于GM(1,1)模型的Kano质量要素分类动态预测方法

针对目前缺乏对Kano模型质量要素的分类结果进行动态预测研究的问题,结合灰色系统理论在处理不规则数据方面的优势,提出基于GM(1,1)模型的Kano质量要素分类动态预测方法。针对传统Kano模型分类结果过于主观的缺点,设计模糊Kano模型实现质量要素的客观化分类;结合GM(1,1)模型能处理少数据、贫信息,具有高预测精度的优点,利用GM(1,1)模型研究Kano质量要素分类动态预测问题;最后通过在快递业的实证验证模型的可行性和有效性。     

粒子群优化灰色模型在负荷预测中的应用

针对电力系统负荷特性,分析灰色模型GM(1,1)的应用局限性,引入向量α改进灰色模型背景值序列的计算公式,从而构建了适应性更强的GM(1,1,α)模型。应用粒子群优化算法非线性全局寻优能力来求解最优α值,提出了基于粒子群优化算法的灰色模型PSOGM,并给出了电力负荷预测的应用实例。实例证明PSOGM模型具有较高的预测精度和较广的应用范围。     

基于组合插值的GM(1,1)模型预测方法的改进与应用

从理论上分析了GM(1,1)模型中的背景值,提出组合插值的思想,利用分段线性插值函数与Newton插值公式结合的方法构造一类新的灰色预测模型CIGM(1,1),改进背景值的构造方法,克服现有的灰色改进模型的不足,为提高预测精度提供了新的途径。最后以2008年江苏省工业用电量数据为例,用本文提出的方法进行预测仿真,理论分析和应用实例表明了本文所提方法的有效性。     

Effects of Knowledge on Attitude Formation and Change Toward Genetically Modified Foods

In three waves, this study investigates the impact of risk and benefit knowledge on attitude formation toward genetically modified (GM) foods as well as the moderating effect of knowledge level on attitude change caused by receiving information. The data in Wave 1 (N = 561) demonstrate that both benefit and risk knowledge either directly contribute to attitude formation or indirectly affect attitudes through the mediating roles of benefit and risk perceptions. Overall, benefit and risk knowledge affect consumer attitudes positively and negatively, respectively. In Wave 2, 486 participants from Wave 1 were provided with information about GM foods, and their attitudes were assessed. Three weeks later, 433 of these participants again reported their attitudes. The results indicate that compared with the benefit and mixed information, risk information has a greater and longer lasting impact on attitude change, which results in lower acceptance of GM foods. Furthermore, risk information more strongly influences participants with a higher knowledge level. The moderating effect of knowledge on attitude change may result from these participants’ better understanding of and greater trust in the information. These findings highlight the important role of knowledge in attitude formation and attitude change toward GM foods as well as the necessity of considering the determinants of attitude formation in attitude change studies.     

Trust, the Asymmetry Principle, and the Role of Prior Beliefs

Within the risk literature there is an ongoing debate on whether trust is vulnerable or enduring. Previous research on nuclear energy by Slovic in 1993 has shown that negative events have much greater impact on self-reported trust than do positive events. Slovic attributes this to the asymmetry principle: specifically, that trust is much easier to destroy than to create. In a questionnaire survey concerning genetically modified (GM) food in Britain (n= 396) we similarly find that negative events have a greater impact on trust than positive events. Because public opinion in Britain is skewed in the direction of opposition toward GM food, the pattern of results could either be caused by the fact that negative information is more informative than positive information (a negativity bias) or reflect the influence of people's prior attitudes toward the issue (a confirmatory bias). The results were largely in line with the confirmatory bias hypothesis: participants with clear positive or negative beliefs interpreted events in line with their existing attitude position. However, for participants with intermediate attitudes, negative items still had greater impact than the positive. This latter finding suggests that, congruent with the negativity bias hypothesis, negative information may still be more informative than positive information for undecided people. The study also identified the labeling of GM products, consulting the public, making biotechnology companies liable for any damage, and making a test available to detect GM produce as being particularly important preconditions for maintaining trust in the regulation of agricultural biotechnology.     

基于多尺度组合模型的铜价预测研究

铜价预测是国际大宗商品市场研究的一个重要领域。本文运用经验模态分解法（EMD）、人工神经网络（ANN）、支持向量机（SVM）和时间序列方法,基于分解-重构-集成的思想,构建了一个多尺度组合预测模型。在模型构建过程中,提出了运用游程判定法对分量序列进行重构的新思路。然后,运用此模型对LME铜价波动特点和走势进行分析：将铜价序列分解并重构成高频、低频和趋势三个部分,并从不规则因素、重大事件以及长期趋势三个角度解释了重构项的波动特征;实证分析表明,与灰色模型GM（1,1）、Elman神经网络方法等单模型,以及ARIMA-SVM组合模型相比,多尺度组合模型取得了最好的预测效果。     

The Role of Similarity Cues in the Development of Trustin Sources of Information About GM Food

In evaluating complex new technologies, people are usually dependent on information provided by others, for example, experts or journalists, and have to determine whether they can trust these information sources. This article focuses on similarity as the basis for trust. The first experiment ( N  = 261) confirmed that a journalist writing about genetically modified (GM) food was trusted more when his attitude was congruent with that of his readers. In addition, the experiment showed that this effect was mediated by the perceived similarity of the journalist. The second experiment ( N  = 172) revealed that trust in a journalist writing about the focal domain of GM food was even influenced by him expressing a congruent attitude in an unrelated domain. This result supports a general similarity account of the congruence effect on trust, as opposed to a confirmatory bias account.     

Computer-aided development of flexible batch production recipes

In this paper an approach to model-based batch process quality control and optimization using flexible production recipes is presented. Unlike the fixed recipes traditionally used in process industries, a flexible recipe depends on externally provided data from the process as well as the market. In order to manipulate a flexible recipe in a proper and convenient way, the information system FRIS—containing modules for experiment design, recipe initialization and correction, on-line process and off-line experiment evaluation, model development and recipe improvement—is being developed. Each of these functions will be discussed further.