Abstract Pipeline is an important and valuable infrastructure for transporting oil and gas which expanding a long distance and working in a corrosive environment. Consequently, corrosion becomes one of the most critical threads for metal material pipeline. The high internal pressure in an oil and gas pipeline is the additional factor leading to the high risk of bursting. Various models predicting the burst pressure of defected pipeline have been developed in literature. However, evaluating burst pressure of defected pipe is a nonlinear mechanical problem with the appearance of the stress concentration, accuracy of the existing models is not high and the issue still open. The application of data-driven approach with soft computing and machine learning has been a potential and promising approach. This paper investigates the application of Adaptive Neuro Fuzzy Inference System (ANFIS) and a data transforming technique for dimension reduction and noise elimination, the Principal Component Analysis (PCA). The PCA has demonstrated its ability in noise removal for the database and ANFIS provides an improvement in the accuracy of the prediction. The developed model is the combination of ANFIS and PCA, the ANFIS-PCA model, has overwhelmed other existing models by archiving the correlation of determination at 0.9919 and the Root Mean Square Error decreases to 0.9883 MPa. Observations on the difference network configurations and number of epochs also provided.

中文翻译：

---

用主成分分析和自适应神经模糊推理系统预测缺陷管道的爆破压力

摘要 管道是输送油气的重要基础设施，具有长距离扩展和腐蚀性环境的特点。因此，腐蚀成为金属材料管道最关键的螺纹之一。油气管道内的高内压是导致爆裂风险高的附加因素。文献中已经开发了各种预测缺陷管道爆破压力的模型。然而，缺陷管道爆破压力的评估是一个非线性力学问题，会出现应力集中，现有模型精度不高，问题仍未解决。数据驱动方法与软计算和机器学习的应用一直是一种潜在且有前途的方法。本文研究了自适应神经模糊推理系统 (ANFIS) 和用于降维和消除噪声的数据转换技术、主成分分析 (PCA) 的应用。PCA 已经证明了它在消除数据库噪声方面的能力，并且 ANFIS 提高了预测的准确性。开发的模型是ANFIS和PCA的结合，ANFIS-PCA模型通过将确定的相关性归档在0.9919并且均方根误差降低到0.9883 MPa来压倒其他现有模型。还提供了对不同网络配置和时期数的观察。PCA 已经证明了它在消除数据库噪声方面的能力，并且 ANFIS 提高了预测的准确性。开发的模型是ANFIS和PCA的结合，ANFIS-PCA模型通过将确定的相关性归档在0.9919并且均方根误差降低到0.9883 MPa来压倒其他现有模型。还提供了对不同网络配置和时期数的观察。PCA 已经证明了它在消除数据库噪声方面的能力，并且 ANFIS 提高了预测的准确性。开发的模型是ANFIS和PCA的结合，ANFIS-PCA模型通过将确定的相关性归档在0.9919并且均方根误差降低到0.9883 MPa来压倒其他现有模型。还提供了对不同网络配置和时期数的观察。