储油罐液位时序数据模式发现

储油罐液位时序数据模式发现对油田生产管理、灾害预警有重要意义,由于目前油气田领域的数据体系繁杂,并未对这些数据加以分类和标识。已有方法借助图形化工具进行人工筛选与检查,这样的方法不适用于长时间不间断生产的石油工业。面对上述问题及已有方法的不足,针对储油罐液位时序数据的特点,提出基于层叠分段与层次聚类模式发现的处理方法。将观测序列转换为离散的线性分段序列,并对各线性分段进行基于DTW(距离的无监督层次聚类,可自动发现时序模式并分配标识符标注时序序列。以储油罐液位时序数据进行实验,发现了隐含的变化模式和变化规律。方法对液位时序变化模式有很好的识别及分类能力,无需人工筛选与检查,并可根据需要,查看不同粒度的变化模式,可为时序数据模式识别,异常检测提供参考和途径。

Pattern discovery of liquid level time series data in oil tank

The liquid level of oil tank time series data model is of great significance for oilfield production, management and disaster warning. Due to the miscellaneous data system in the field of oil and gas fields, these data are not classified and marked. There are some methods for manual screening and checking with graphical tools, which are not suitable for long time uninterrupted production of petroleum industry. In the face of the above problems and the shortcomings of the existing methods, a processing method which based on cascade piecewise linear representation and hierarchical clustering for the characteristics of reservoir tank level data is proposed. The observation sequence is transformed into discrete linear piecewise sequence, and each linear segment is clustered by unsupervised hierarchical clustering based on DTW distance, which can automatically discover the temporal pattern and assign identifiers to annotate the sequence. Based on the data of oil tank level sequence data, and the implied models and the changing rules were found. The method has a good ability to recognize and classify the time series change patterns of liquid level, without manual screening and inspection, and can view the changing patterns of different granularity according to the need, which can provide a reference and avenues for time series data pattern recognition and abnormality detection.