遥测数据时间序列滑动窗口动态分割技术

经由传感器采集的航天器遥测数据是产品验证、算法检验、数据挖掘等学科和方向重要的数据来源之一。由于原始测量数据采集密度高、信息量大,通常采用滑动窗口技术进行逐段处理和算法应用。但当滑动窗口的宽度固定时,数据的局部信息不能被充分提取出来;为此,提出遥测时序滑动窗口的动态分割流程。仿真结果表明:提出的时序分割和窗口宽度计算方法,可以根据数据自身的特点得到更加准确合理的非等长窗口集合;在保持曲线线段曲率的基础上,滑动窗口动态计算方法可以极大地提取拐点信息,合理地给出窗口宽度,对数据进行了简化与提纯,降低了建模难度,为下一步聚类分析、曲线匹配等工作提供了技术支持,提高了遥测数据的诊断与分析效率。

A Dynamic Slide Windows Segmentation Technology for Telemetry Time Series

Telemetry data collected by sensors is an important data source for product validation, algorithm test and data mining. A widely used technology is slide windows to process data piecewise considering the mass amount and density of data. But it is difficult to extract local information once the length of windows is fixed. To solve this problem, a dynamic slide windows segmentation procedure is introduced. It is found that the procedure obtains more accurate and rational results using dynamic length of windows. While maintaining curve curvature, the slide window dynamic calculation method is capable of more powerful extraction of inflection point information and it gives reasonable window width. Data simplification and purification is performed to reduce the difficulty of modeling and to provide technical preparation for next step of clustering analysis and curve matching. As a result, it increases the efficiency of telemetry diagnosis and analysis.