摘要
在分析前向BP神经网络基本原理的基础上,对3种混油建立了人工神经网络混油粘度预测模型,该模型结构为1-7-1的三层BP网络模型。运用实测数据对BP网络进行训练和仿真。结果表明,三种模型预测误差全在2.5%以内,比前苏联学者提出的混油粘度计算公式——克恩达尔-莫恩罗埃公式和兹达诺夫斯基公式更具有计算精度高、适用性强的特点,可完全满足工程实际需要。
The model of forecasting the viscosity of oil mixture is set up respectively to three different mixtures based on analysis of the basic principle of forward back propagation (BP neural network. The structure of model is 1-7-1 three-layer BP network.. The three BP neural networks are trained and simulated respectively by using practical measuring data. The results show that the errors of three models are all less than 2.5%. It also indicates that the present method has higher accuracy and wider applicability than Kerndal-Munnloe formula and Zdanowski formula proposed by former Soviet scholar and it can well meet the needs of engineering.
出处
《石油与天然气化工》
CAS
CSCD
北大核心
2007年第4期335-337,共3页
Chemical engineering of oil & gas
基金
中石化集团公司项目X504007
江苏省油气储运重点实验室资助ZDK0602004
关键词
管道
BP神经网络
混油粘度
预测模型
pipeline, BP neural network, viscosity of oil mixture, forecasting model
