MA Gang,LI Jun-fei,BAI Rui,DAI Zheng.Prediction of Corrosion Rate in Oil and Gas Pipelines Based on PSO-SVM Model[J],48(5):43-48 |
Prediction of Corrosion Rate in Oil and Gas Pipelines Based on PSO-SVM Model |
Received:August 12, 2018 Revised:May 20, 2019 |
View Full Text View/Add Comment Download reader |
DOI:10.16490/j.cnki.issn.1001-3660.2019.05.007 |
KeyWord:PSO-SVM model oil and gas pipeline internal corrosion rate prediction error analysis |
Author | Institution |
MA Gang |
1.School of Petroleum Engineering, Xi'an Shiyou University, Xi'an , China |
LI Jun-fei |
2.China Petroleum and Natural Gas Pipeline Bureau, Langfang , China |
BAI Rui |
1.School of Petroleum Engineering, Xi'an Shiyou University, Xi'an , China |
DAI Zheng |
1.School of Petroleum Engineering, Xi'an Shiyou University, Xi'an , China |
|
Hits: |
Download times: |
Abstract: |
The work aims to establish a new prediction model of the corrosion rate in oil and gas pipelines for the operational safety problem of oil and gas pipelines, so as to accurately predict the internal corrosion conditions of pipelines. The principle of internal corrosion was firstly analyzed and the main causes of corrosion in pipelines were discussed. The principles and structures of PSO (particle swarm optimization), SVM (support vector machine) and PSO-SVM models were discussed. The parameters C and g of the SVM algorithm were optimized by PSO algorithm in combination with the internal corrosion data obtained in the pipeline. On the basis of this, the three kernel functions of the Sine function, the Sigmoidal function and the Radial basis function were compared and optimized. Finally, the prediction errors of the GA-SVM model, CV-SVM model, LS-SVM model and FOA-SVM model were compared to prove the advanced nature of the PSO-SVM model. When the SVM algorithm parameters C=83.9243, g=0.6972, and the kernel function was the Sine function, the average absolute error and root mean square error of the PSO-SVM model were the smallest, respectively 0.58% and 0.000 618, but the time for training data was 11.26 s. Compared with the GA-SVM model, CV-SVM model, LS-SVM model and FOA-SVM model, the prediction error was small, but the training data took a long time. It is feasible to predict the corrosion rate in oil and gas pipelines by PSO-SVM model. The prediction error is relatively small, but due to the limitation of data training speed, it is still necessary to conduct in-depth research in this field. |
Close |
|
|
|