摘要
对火电控制历史数据进行分析和建模,能有效地帮助用户实现更优的机组运行控制.由于这类控制数据的复杂性,传统系统辨识过程非常烦琐,甚至难以得到有效结果.为此,将可视分析技术引入系统辨识中,与自动建模算法集成,设计了面向系统辨识的可视分析系统imDCS,在时序数据特征分析、模型建立和筛选评估以及模型迭代优化等各阶段支持控制系统建模的全过程.通过多种可视化映射技术和交互联动视图支持多层次模型筛选过程;通过分组视图与堆叠视图展示高维多元模型结构;通过精度评估组合视图支持用户从不同侧面评估模型性能.与领域专家合作,基于电厂真实控制数据和运行优化需求进行了案例分析和评估,结果表明,该系统在工业控制数据分析和建模中具有更高的有效性和可用性.
The analysis and modeling of historical data of thermal power control can help users achieve better operation control. Due to the high data complexity, traditional system identification approaches are very complicated, and are difficult to obtain effective results. This paper integrates visual analytics into the process of system identification, called imDCS. The system supports the entire process of control system modeling in all stages including time series feature analysis, model establishment, selection and evaluation, and model iteration optimization. By using various visual mappings and coordinated views, it allows for multi-level model selection, and illustrates high-dimensional and multivariate model structure. By using accuracy evaluation combined views, it supports model performance evaluation from different perspectives. We worked with domain experts to case studies and evaluation based on real control data and operation optimization requirements of a power plant. The results verify the effectiveness and usability of our approach in industrial control data analysis and modeling.
作者
纪连恩
孔雨萌
王炎林
王中原
田彬
张东明
Ji Lianen;Kong Yumeng;Wang Yanlin;Wang Zhongyuan;Tian Bin;Zhang Dongming(Beijing Key Laboratory of Petroleum Data Mining, Beijing 102249;Department of Computer Science and Technology, China University of Petroleum, Beijing 102249;Beijing Guodian Zhishen Control Technology Co., Ltd., Beijing 102200)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2019年第10期1677-1686,共10页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60873093)
关键词
可视分析
系统辨识
火电控制
时序数据
visual analytics
system identification
thermal power control
time series data
