基于参数优化神经网络的海底油气管道腐蚀泄漏预测
DOI:
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

TE832

基金项目:

国家工信部高技术船舶项目(项目编号:2018GXB01-02-003)。


Research on corrosion leakage prediction of submarine oil and gas pipeline based on neural network for parameter optimization
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    为有效预测海底管道因腐蚀导致的泄漏风险,提出了一种海底管道腐蚀泄漏预测模型,首先采用斯皮尔曼相关系数分析各影响因素间的相关性,随后基于随机森林袋外数据进行各因素的重要性排序,剔除掉相关性较高且重要性较小的因素,利用筛选出的数据建立前馈神经网络和随机森林回归预测模型,并利用粒子群算法对神经网络预测模型的权值、阈值进行了优化,构建粒子群优化下的神经网络预测模型。经分析结果表明:神经网络预测模型在5组随机模型训练中平均绝对误差(MAE)、均方误差(MSE)的平均值分别为1.59、 3.37,均高于随机森林回归预测模型,说明该模型误差较大,但决定系数(R2)较随机森林回归预测模型高0.13,因决定系数越接近于1,模型拟合越好,故随机森林回归预测模型较神经网络预测模型拟合度较差,长期预测误差较高,因此可采用粒子群算法对神经网络进行优化,优化后的模型MAE为0.79,MSE为0.7293,R2为0.9151,可见优化后的神经网络预测模型在保证精度的基础上提高了稳定性,预测效果更优。最后编制了集随机森林回归、神经网络及粒子群优化下的神经网络为一体的多模型管道腐蚀预测软件。为海底管道泄漏风险的精准预测以及高效控制提供了依据,在海洋油气运输安全方面具有重要意义。

    Abstract:

    A prediction model for submarine pipeline corrosion leakage is proposed in order to effectively predict the leakage risk of submarine pipelines caused by corrosion. Firstly, the correlation between the influencing factors is analyzed by the Spearman correlation coefficient, and then the importance of each factor is ranked based on the data outside the random forest bag to eliminate the factors with high correlation and low importance. Using the selected data, the feedforward neural network and random forest regression prediction model are established. The neural network prediction model's weight and threshold are optimized using particle swarm optimization, and the neural network prediction model is built using particle swarm optimization. The analysis results show that the neural network prediction model's average absolute error(MAE) and mean square error (MSE) in the training of five groups of random models are 1.59 and 3.37, respectively, which are higher than the random forest regression prediction model, indicating that the model's error is large, but the determination coefficient (R2) is 0.13 higher than the random forest regression prediction model, because the closer the determination coefficient is to “1”, the better the model. As a result, the random forest regression prediction model has a lower fitting degree and a higher long-term prediction error than the neural network prediction model. Therefore, particle swarm optimization algorithm can be used to optimize the neural network. The optimized model MAE is 0.79, MSE is 0.7293 and R2 is 0.9151. It can be seen that the optimized neural network prediction model improves the stability and better prediction effect on the basis of ensuring the accuracy. Finally, a multi model pipeline corrosion prediction software integrating random forest regression, neural network and neural network under particle swarm optimization is compiled. It provides a basis for accurate prediction and efficient control of submarine pipeline leakage risk, and is of great significance in the safety of offshore oil and gas transportation.

    参考文献
    相似文献
    引证文献
引用本文

鲁中歧,肖文生,崔俊国,等. 基于参数优化神经网络的海底油气管道腐蚀泄漏预测[J]. 科学技术与工程, 2022, 22(20): 8673-8682.
Lu Zhongqi, Xiao Wensheng, Cui Junguo, et al. Research on corrosion leakage prediction of submarine oil and gas pipeline based on neural network for parameter optimization[J]. Science Technology and Engineering,2022,22(20):8673-8682.

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2022-01-14
  • 最后修改日期:2022-04-15
  • 录用日期:2022-03-16
  • 在线发布日期: 2022-08-04
  • 出版日期:
×
律回春渐,新元肇启|《科学技术与工程》编辑部恭祝新岁!
亟待确认版面费归属稿件,敬请作者关注