基于Lasso-PSO-BP神经网络的腐蚀管道失效压力的预测

Prediction of Failure Pressure of Corroded Pipeline Based on Lasso-PSO-BP Neural Network

管道是运输石油和天然气的重要工具,随着其腐蚀现象愈发严重,必须对腐蚀管道的失效压力进行预测。针对如何精确预测腐蚀管道失效压力的问题,提出一种Lasso-PSO-BP模型。首先,结合收集的腐蚀管道爆破试验数据,利用Lasso回归筛选出失效压力影响因素,确定BP神经网络的输入变量;然后,用粒子群算法(PSO)优化反向传播(BP)神经网络初始权值阈值;最后,将优化训练后的BP神经网络用于管道失效压力的预测。通过实例验证,对比分析2种模型的拟合效果,结果表明:相较Lasso-BP预测模型,Lasso-PSO-BP预测模型的平均误差(AE)从0.102 5减小到0.030 1,均方根误差(RMSE)由1.174 3减小到0.297 2,其各项指标都优于Lasso-BP模型,证明此方法具有较高的准确率,显示了PSO-BP神经网络模型更优的拟合度与预测精度,适用于腐蚀管道失效压力的预测。

Pipeline is an important tool for transporting oil and gas. With the increase of corrosion phenomenon,it is essential to predict the failure pressure of corroded pipeline. A Lasso-PSO-BP model was proposed to accurately predict the failure pressure of corroded pipelines.Firstly,combined with collecting experimental data of corrosion pipeline blasting,Lasso regression was used to screen out the influencing factors of failure pressure and determine the input variables of BP neural network. Then,particle swarm optimization (PSO) was used to optimize the initial weight threshold of BP neural network. Finally,the BP neural network after optimized training was applied to the prediction of pipeline failure pressure. Through the example verification,the two kinds of model fitting effect were contrasted and analyzed. Results showed that in the comparison of Lasso-BP predict model,the average error (AE) of Lasso-PSO-BP prediction model decreased from 0.102 5 to 0.0301,and the root mean square error (RMSE) decreased from 1.174 3 to 0.297 2,which indicated that all indexes were better than those of the Lasso-BP model. Therefore,this method had high accuracy and displayed better fitting degree and prediction precision,proving that it was suitable to be applied for the perdition of failure pressure of corroded pipeline.