| Autor: |
薛禹胜 Xue, Yusheng, 黄天罡 Huang, Tiangang, 陈国平 Chen, Guoping, 郑玉平 Zheng, Yuping, 文福拴 Wen, Fushuan, 徐岩 Xu, Yan, 赵俊华 Zhao, Junhua |
| Přispěvatelé: |
School of Electrical and Electronic Engineering |
| Jazyk: |
čínština |
| Rok vydání: |
2019 |
| Předmět: |
|
| Popis: |
算例筛选环节通过定性的机器学习或者定量的近似评估，快速识别出尽量多的稳定算例及失稳算例，以减少需要详细分析的算例数及总计算量。文中评述暂态稳定分析中的算例筛选所采用的假设、特征变量、分类规则及泛化能力。讨论数据驱动与模型驱动的融合，包括将因果元素引入分类器，将数据统计范式与模型仿真范式深度融合来提取知识。在稳定性理论指导下，建议两层框架的分类器:下层由若干个并行环节组成，各自采用不同近似程度的定量算法，取其输出数据作为上层分类器的输入。这些数据已经按近似的因果关系整合了大部分原始数据对暂态稳定性的影响，不但可提高分类器的正确识别率和强壮性，并可揭示误差机理、评估可信度及可接受度。In the case filtering link, as many stable and unstable cases as possible can be identified rapidly through qualitative machine learning or quantitative approximate analysis so that the number of cases requiring detailed analysis and the total computational burden can be reduced. This paper discusses the adopted assumptions, characteristic variables, classification rules and generalization ability of case filtering in transient stability analysis. The integration of data driven and model driven is analyzed, which includes the introduction of causal elements, the deep integration of data statistic paradigm and model simulation paradigm for knowledge extraction. Based on the stability mechanism, a two-layer classifier is proposed: the lower layer contains several parallel links, and each link utilizes quantitative algorithm with different degrees of approximation. Their output data are regarded as the input of the upper layer. According to the approximate causality, it can reflect the influence of most original data on transient stability. Therefore not only the correct recognition rate and robustness of the classifier can be both enhanced, but also the error mechanism, evaluation credibility and acceptability can be revealed. Published version |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
