改进PSO-BPNN算法在管道腐蚀预测中的应用

Application of Improved PSO-BPNN Algorithm in Corroded Pipelines Prediction

输油管道由于埋藏环境、运输介质等影响,随着使用年限增加,管道会逐渐出现腐蚀,常规的腐蚀管道剩余强度计算方法有公式计算和有限元分析(FEA)等。针对常规方法中公式计算准确性较低和有限元分析过于复杂的问题,提出了一种改进的粒子群算法优化的神经网络模型(IPSO-BPNN)来预测腐蚀管道剩余强度。首先,在传统粒子群算法的基础上,提出了一种新的非线性递减惯性权重用于快速更新粒子速度和位置,并引入了遗传交叉算子增加粒子的多样性,形成了改进的粒子群算法(IPSO);其次,采用IPSO算法对神经网络的权重和阈值进行优化,并使用优化后的权重和阈值初始化神经网络,建立了IPSO-BPNN模型;最后,在2个真实的管道测试爆破数据集上进行实验,分别使用线性回归(LR)、FEA、前馈神经网络(BPNN)、粒子群算法前馈神经网络(PSO-BPNN)以及IPSO-BPNN模型对腐蚀管道剩余强度进行预测,使用平均绝对误差(MAE)、均方根误差(RMSE)和平均绝对百分比误差(MAPE)作为评估模型预测性的指标。在2个数据集的测试集上的结果表明:IPSO-BPNN模型的MAE分别为0.525 4、0.718 5,MAPE分别为3.77%、2.68%,RMSE分别为0.672 6、0.947 2,3项指标较LR、FEA、BPNN和PSO-BPNN有明显提升。改进PSO-BPNN算法可以提高腐蚀管道剩余强度预测的准确性,可以为管道检查提供较为准确的依据。

Due to the impact of buried environment and medium transported,oil pipelines will be gradually corroded with the increasing of service life.Traditional methods for calculating the residual strength of corroded pipelines included formula calculation and finite element analysis,etc.Aiming at the problems of low calculation accuracy of formulas and too complicated finite element analysis(FEA)in the prediction of the residual strength of corroded pipelines,an improved particle swarm optimization neural network model(IPSO-BPNN)was proposed to predict the residual strength of corroded pipelines.Firstly,based on the traditional particle swarm optimization,a new nonlinear decreasing inertia weight was proposed to update the velocity and location of elements quickly,and the genetic crossover operator was introduced to increase the diversity of particles,then form an improved particle swarm optimization algorithm(IPSO).Secondly,the IPSO algorithm was used to optimize the weights and thresholds of the neural network,and initialize the neural network with optimized weights and thresholds to establish the IPSO-BPNN model.Finally,the linear regression(LR),FEA,back-propagation neural network(BPNN),particle swarm optimization back-propagation neural network(PSO-BPNN)and IPSO-BPNN model were experimented on real pipelines test blasting data sets to predict the residual strength of the corroded pipelines.MAE,RMSE and MAPE were used as indicators to evaluate the predic-tability of the models.The results on the test set of two data sets showed that the MAE of the IPSO-BPNN model was 0.5254,0.7185;the MAPE was 3.77%,2.68%;the RMSE was 0.6726,0.9472,respectively.The three indicators were significantly improved compared to LR,FEA,BPNN and PSO-BPNN.It showed that this method could improve the accuracy of predicting the residual strength of corroded pipelines,and could provide a more accurate basis for pipeline inspection.