BP神经元网络在炼油厂低温腐蚀预测模型上的应用  

APPLICATION OF CORROSION RATE PREDICTION MODEL BASED ON BP NEURAL NETWORK IN REFINERY

作　　者：赵林[1] 李博文[1] 郑丽群[1] 杨富淋[1] 宋以常[2] 付伟[2] 

机构地区：[1]中国科学院金属研究所,辽宁沈阳110016 [2]中国石油化工股份有限公司北京燕山分公司,北京102500

出　　处：《石油化工设备技术》2013年第1期59-62,1,共4页Petrochemical Equipment Technology

摘　　要：将神经元网络建模应用于炼油厂腐蚀速率预测,整理分析现场数据,对神经元网络模型进行训练,建立了低温腐蚀速率预测模型。采用改进的Levenberg-Marquardt算法训练模型,较传统的梯度下降法可以更好地完成算法收敛。神经元网络建模后,可以将现有的腐蚀数据涵盖在腐蚀模型中,便于腐蚀数据的管理和查询,同时对现场参数变化影响腐蚀速率的情况起到了预测作用。预测结果与实际值之间的平均方差为0.013 2,可以满足现场需要。The neural network model is applied to predict the corrosion rate in the refinery.The field data is sorted out and analyzed.The neural network model is trained and the mathematical model for corrosion rate prediction is established.The improved Levenberg-Marquardt algorithm is used to train the model and this algorithm is better than the traditional gradient-descent algorithm for the algorithm convergence.Corrosion data can be included in the existing corrosion models by neural network modeling to facilitate data management and inquiry;meanwhile,the model can predict the relationship between the on-site parameters and corrosion rate.The average variance between the predicted value and the actual value is 0.013 2 which can meet the field needs.

关 键 词：神经元网络 LEVENBERG-MARQUARDT算法 腐蚀速率 现场数据 

分 类 号：TE986[石油与天然气工程—石油机械设备]