【作者】 张刚；

【导师】 王立欣；

【作者基本信息】 哈尔滨工业大学 ， 电气工程， 2014， 博士

【摘要】 作为电磁仿真模型有效性评估的核心内容，仿真结果可信度的定量评估可以消除主观目测评估的不确定性和不稳定性，为仿真模型修正及改进提供指导。特征选择验证（FSV：Feature Selective Validation）方法是IEEE标准1597.1所确立的电磁仿真有效性评估体系的核心方法，结合IEEE标准委员会的修订要求，对FSV方法本身的特性、有效性及功能拓展等关键问题进行研究对IEEE标准1597.1的进一步修订和推广具有重要意义。本课题首先对FSV方法主要特性进行了分析。通过与常用数据差异评估算法的对比说明了FSV方法在电磁仿真结果可信度评估中的适用性。对FSV方法数据敏感性的分析主要考察待评估数据的数据量变化对评估结果的影响，分析证明了当数据密度及数据丢失比例在一定范围内变化时FSV评价结果具有鲁棒性。参数敏感性分析则考察FSV方法中关键参数取值变化对评估结果的影响，通过分析确定了两个敏感参数并验证了这两个参数取值的合理性。在论证了FSV特性的基础上，为了使FSV方法更接近于专家评价，提出了对其评估结果定性表达方法的改进。首先，提出使用连续概率密度函数代替直方图来表示定性评价结果，从而可以提供更加细致的定性信息，并可以为多结果之间的交叉统计分析提供条件。其次，研究了FSV方法的定性结果与定量结果之间对应关系的模糊化问题。使用浮动边界代替原有的硬边界来重新定义定性评估与定量结果之间的对应关系，在结果表达层面减小了FSV方法的定性评估结果与专家调查结果之间的差异。此外，本课题通过组织新的数据可信度专家问卷调查，分析了专家目测评估的特点，为本课题对FSV方法的改进提供了参考数据。针对FSV方法在实际应用中出现的三种失效情况，分别研究了其失效原因及解决方案。首先，通过对直流差异评估量（ODM）算法的改进，解决了FSV方法在评估正负交替数据时的失效问题，且这一改进对常规数据的评估没有影响。其次，通过分段评估改进了FDM的计算，提出了改进的MAF-FSV算法，解决了对含有瞬态分量数据的评估失效问题。最后，探索了畸变数据的评估问题，提出使用Feature Selective Correction（FSC）技术进行数据校正，改进了全局差异评估量（GDM）计算方法。以上改进均通过与专家调查结果的对比证明了其有效性，增强了FSV方法的评估稳定性。针对FSV方法在实际应用中面临的多仿真结果可信度的综合评价问题，引入模糊综合评价模型和层次分析法，建立了基于FSV的多结果可信度综合评估方法，拓展了FSV方法的功能。在此基础上结合本课题对FSV提出的改进，研究高频多导体串扰简化模型的仿真结果可信度评估问题。通过对频域和时域仿真结果的综合评价证明了本课题提出的简化算法的有效性和优势。更多还原

【Abstract】 Quantifying the credibility of simulation result can eliminate the uncertainty and instability introduced by subjective visual assessment, which is the core content of validity assessment of electromagnetic simulation model. Further, quantitative assessment can provide a range of useful information for the rectification and improvement of the simulation model. The Feature Selective Validation (FSV) method is the core of the electromagnetic simulation validity assessment system established by the IEEE standard1597.1. So it is important to study the properties, validity, functional extension and other key issues of FSV method, which will also meet the requirement of IEEE Standards Association on the revision and promotion of IEEE standard1597.1.Properties of FSV method are first investigated in this project. The applicability of FSV method in the credibility assessment of electromagnetic simulation result is demonstrated by comparing with some frequently used data comparison algorithms. The influence of the quantity of data points on the assessment result is evaluated by data sensitivity analysis. It is demonstrated that the FSV result keeps robust when data density and integrity of data sets are changed to some extent. The parameter sensitivity analysis examines the influence of the changing values of key parameters on the FSV results. In this way, two sensitive parameters are identified and their values are proved to be reasonable.Based on the comprehensive discussion of the properties of FSV method, the improvement in the representation of FSV qualitative result is proposed, which aims to reduce the discrepancy between FSV result and experts’ assessment. First, the histogram result is replaced by the probability density function which can give the qualitative results in detail and makes it possible to do cross statistical analysis between multiple data. Further, fuzzification of the correspondence relationship between FSV qualitative and quantitative results is studied, which is realized by introducing floating boundaries of qualitative categories. It is shown that the agreement between FSV prediction and experts’ assessment can be improved by the floating boundaries. In addition, characteristics of visual assessment are also analyzed by performing a new data credibility survey. These visual assessments offer reference information for the following research on the improvement of FSV.Assessment failures of FSV method in three situations are also discussed and improved. First, the problem occured in the assessment of data oscillating between positive and negative half planes is solved by improving the algorithm of Offset Difference Measure (ODM). Data tests suggest that this modification can reduce the disagreement between FSV results and visual assessment without any pre-processing to original data sets. Secondly, the MAF-FSV method is proposed to improve the assessment of transient data sets, which divides the data sets under comparison into three regions and modifies the algorithm of Feature Difference Measure (FDM). Additionally, the assessment of distorted data sets is discussed. New Global Difference Measure (GDM) is calculated after the data sets are corrected by the Feature Selective Correction (FSC) technique. All of these improvements are validated by comparing with the survey results.To assess the credibility of multiple simulation results, the comprehensive assessment method is established based on the FSV method, fuzzy comprehensive assessment model and the Analytic Hierarchy Process (AHP), which extends the function of FSV method. Then the method is applied to the validation of the reduction model of multiconductor crosstalk problem in high frequency. In this process, improvements of FSV proposed in this project are also used. The validity and advantages of the proposed model are demonstrated by the comprehensive credibility assessments of simulation results in frequency and time domain.更多还原