BP神经网络模型的改进及其在海底管道外腐蚀速率预测中的应用  

network model and its application in the prediction of the external corrosion rate of submarine pipeline

作　　者：张梁[1] 高源 杨光 李民强 ZHANG Liang;GAO Yuan;YANG Guang;LI Minqiang(School of Mechanical Engineering,Southwest Petroleum University,Chengdu 610500,China;Offshore Engineering Technology Service Center of Sinopec Shengli Oilfield Offshore Oil Production Plant,Dongying 257237,Shandong,China)

机构地区：[1]西南石油大学机电工程学院,成都610500 [2]中石化胜利油田海洋采油厂海洋工程技术服务中心,山东东营257237

出　　处：《安全与环境学报》2023年第11期3882-3888,共7页Journal of Safety and Environment

基　　金：国家重点研发计划项目(2016YFC0802102-03)。

摘　　要：针对海底油气管道外腐蚀问题构建模型预测其腐蚀速率,建立基于改进鹈鹕优化算法(Improvement Pelican Optimization Algorithm,IPOA)的BP神经网络(Back Propagation Neural Network,BPNN)腐蚀速率预测模型。通过Logistic-Tent混沌映射初始种群与收敛因子非线性化的方法提高鹈鹕算法(POA)的全局搜索能力和寻优精度,采用IPOA算法优化BPNN的权值和阈值,提升模型的预测精度与鲁棒性。以实海挂片试验数据为基础,建立POA-BPNN和BPNN模型作为对比。结果表明:IPOA-BPNN模型的决定系数R2为0.9664,均方误差为0.2353,平均相对误差为3.16%,均优于其余两个模型,模型的鲁棒性较未改进有较大的提升。这表明IPOA-BPNN模型能够为海底管道的维修与更换提供决策支持。The external corrosion of submarine oil and gas pipelines is one of the main causes of pipeline failure.The external corrosion rate is closely related to marine environmental factors.In this paper,by analyzing the relationship between the marine environment and the external corrosion rate of pipelines,theImprovement tPelicanOptimization Algorithm-Back Propagation Neural Network(IPOA-BPNN)model is constructed to predict the external corrosion rate of submarine oil and gas pipelines.The global search ability and optimization accuracy of the POA are improved by the initial population of the chaotic map and the nonlinearization of the convergence factor.The IPOA algorithm is used to optimize the weight threshold of BPNN to improve the prediction accuracy and robustness of the model.This paper mainly improves the pelican optimization algorithm from two aspects:the initial population of the pelican optimization algorithm is used to increase the diversity of the initial population by Logistic-Tent chaotic mapping,and the convergence factor is nonlinearized by the sine trigonometric function.The idea increases the ability of global optimization in the early stage of the algorithm and improves the accuracy of the later optimization of the algorithm.Collecting and adjusting the corrosion test data of real sea hanging pieces can be applied to study the corrosion rate of submarine pipelines.Based on the experimental data of real sea coupons,the prediction results of POA-BPNN and BPNN models are established in MATLAB software by the Hold-out verification method.The results show that the square of the determination coefficient of the IPOA--BPNN model is 0.9664,the mean square error is 0.2353,and the average relative error is 3.16%,better than the other two models.The model has better prediction accuracy and fitting ability,and the robustness of the model is greatly improved compared with the unimproved model.The application of the model in engineering practice can reduce the calculation and analysis time,improve work efficienc

关 键 词：安全工程 海洋油气管道 管道腐蚀速率 改进鹈鹕优化算法(IPOA) BP神经网络(BPNN) 

分 类 号：X937[环境科学与工程—安全科学]