引入重叠度指标的FPPC油气管道管段划分方法

为了准确、合理地划分油气管道的管段数目,有针对性地维护具有不同风险的管段,降低管道风险评价的成本,提出了一种改进的、引入重叠度指标的模糊投影寻踪聚类(FPPC)管段划分方法。从管道样本集的数据类特征出发,构造了同时考虑到类间稀疏度、重叠度和类内紧密度的投影指标函数,据此建立了改进的FPPC算法管段划分模型,并以我国某输气管道的管段划分为例,对比分析了改进算法与传统FPPC算法的管段划分效果。研究结果表明:(1)改进的FPPC模型无需制订相应的管道风险等级标准,克服了传统模型随机、模糊等主观性缺陷,将管段划分为4类更加合理准确;(2)设计的投影指标函数能够识别管道样本中的小类,提高了管段划分的聚类精度和可信度;(3)与投影寻踪聚类算法和传统的FPPC算法对比结果表明,改进的FPPC算法收敛速度更快、迭代次数更少。结论认为,该研究成果给出了更为科学的管段划分方法,为油气管道的风险评价提供了理论依据。

An improved FPPC algorithm for oil & gas pipeline segmentation by introducing an overlap index

In order to divide a pipeline accurately and reasonably, specifically maintain pipeline segments with different risks and reduce the pipeline risk assessment cost, we proposed an improved fuzzy projection pursuit clustering algorithm (FPPC) for pipeline segmentation in this paper. In this method, the overlap index is introduced. A new projection index function which takes inter-class sparseness, overlap degree and intra-class compactness into account was constructed with the data class characteristics of the pipeline sample set as the beginning point. Based on this, an improved FPPC algorithm segmentation model was established. Finally, the pipeline segmentation result of the improved FPPC algorithm was compared with that of traditional FPPC algorithm with the segmentation of one gas pipeline in China as an example. And the following research results were obtained. First, when the improved FPPC model is adopted, the subjective defects of traditional model (e.g. stochastic and fuzzy) are overcome while the corresponding pipeline risk level standard is not needed. In this improved FPPC model, the pipeline is divided into 4 classes, which is more reasonable and accurate. Second, by virtue of the designed projection index function, the sub-classes of the pipeline samples can be identified and the clustering accuracy and reliability of pipeline segmentation are improved. Third, compared with the projection pursuit clustering (PPC) algorithm and the traditional FPPC algorithm, the improved FPPC algorithm has a higher convergence rate and less iteration. In conclusion, the research results provide a more scientific method for pipeline segmentation, as well as a theoretical basis for oil & gas pipeline risk assessment.