基于PSO-SVM模型的油气管道内腐蚀速率预测

Prediction of Corrosion Rate in Oil and Gas Pipelines Based on PSO-SVM Model

目的针对油气管道的运行安全问题,建立油气管道内腐蚀速率预测新模型,对管道的内腐蚀状况进行准确预测。方法首先对内腐蚀的原理进行简单分析,探讨引起管道内腐蚀的主要原因。对PSO(粒子群算法)、SVM(支持向量机)以及PSO-SVM模型的原理及结构进行探讨,结合文献中获取的管道内腐蚀数据,使用PSO算法对SVM算法的参数C和g进行寻优。在此基础上,对Sine函数、Sigmoidal函数和Radial basis函数三种核函数进行对比优选。最终将PSO-SVM模型与GA-SVM模型、CV-SVM模型、LS-SVM模型和FOA-SVM模型四种模型进行预测误差对比,以此证明PSO-SVM模型的先进性。结果当SVM算法的参数C=83.9243、g=0.6972,核函数选择Sine函数时,PSO-SVM模型的平均绝对误差和均方根误差最小,平均绝对误差和均方根误差分别为0.58%和0.000618,但是该模型在使用的过程中,其训练数据所使用的时间为11.26 s,与GA-SVM模型、CV-SVM模型、LS-SVM模型和FOA-SVM模型四种模型相比,其预测误差较小,但训练数据所使用的时间较长。结论利用PSO-SVM模型对油气管道内腐蚀速率进行预测是可行的,预测误差相对较小,但是由于受限于数据训练速度问题,今后仍需要对该领域进行深入研究。

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 was11.26 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.