Abstract: Soil corrosivity is an important factor for the design and management of long-distance pipelines. Up to now, none of the methods for predicting soil corrosivity is completely successful in both reliability and flexibility. In recent Years, pattern recognition has become a powerful tool to handle acquisition,trangformation,display and interpretation of experimental data. This paper demonstrates the principle and the procedure of Principal Component Analysis(PCA)for predicting soil corrosivityalong long-distance pipelines. Eleven data groups of weight loss tests along the Cang-ling and the Pu-ling pipelines were gathered as the "learning set" for PCA. Three main characteristics of contents of CaCO₃, content of total dissolved salts, and soil resistivity were taken as the key parameters.In the planar display generated by the first and third components of PCA, the data set had obvious class separability. The decision rules were also given to predict the corrosivity of soils.The PCA prediction for other 51 soil samples was made andthe results were found in good acco rd with that of the NLM prediction and were proved true by practical records of the pipelines breakdown caused by soil corrosion during the past 10 Years.

摘要： 现有的预测土壤腐蚀性方法,如电阻率法和多因子综合评分法,存在可靠性差;或缺乏应变性能及因子过多影响其实用性等缺点.近年迅速发展的计算机模式识别方法是一种处理多元变量的有效方法.本文重点介绍主分量的原理和分析法(PCA)对长输管道沿线土壤腐蚀等级的预测.根据濮临,沧临两线沿线少量的土壤腐蚀埋片数据作为训练样本集,采用土壤电阻率,含盐量和CaCO₃量三项特征指标为关键参数,用1,3主分量坐标平面显示,可使样本明显地按土壤腐蚀等级分类.以此推导出相应的超平面界面方程和特征指标判别公式.对同管线的另外51个土壤点(需要测定三个特征指标)进行预测,画出管线沿线的土壤腐蚀等级图,经检验,与现实情况基本相符.文中还讨论了处理土壤腐蚀数据的具体技术和实用的预测方法.

关键词: 管道, 土壤腐蚀, 腐蚀等级预测, 主分量分析, 计算机识别