人工神经网络在石油分析中的应用研究(Ⅰ)——BP神经网络预测石油馏分临界性质  

Application and Study of Artificial Neural Network in Petroleum Analysis(I) ——Application of BP Neural Networks for Prediction of Petroleum Fraction Critial Property

作　　者：周山花[1,2] 张晓彤[1,2] 张素萍[1,2] 孙兆林 李梦龙[1,2] 

机构地区：[1]抚顺石油学院应用化学系 [2]四川联合大学化学系

出　　处：《石油化工高等学校学报》1998年第1期23-27,共5页Journal of Petrochemical Universities

基　　金：中石化总公司资助

摘　　要：基于面向对象程序设计的设计思想，定义３种新的神经元结构体类型变量，研究了４种变形ＢＰ神经网络模型，引入神经元非线性敏感度因子、动量因子和阈值的动态修正，在一定程度上克服了传统ＢＰ网络收敛速度慢、易陷于局部极小的缺陷，将４种模型应用于石油馏分临界性质的预测，取得较一般常规非线性处理方法更高的预测精度。Based on the object-oriented programming concept, three new structural variables are defined, which makes neuron become object-oriented variable. Four reductive models are studied. Nonlinear sensitive factor, memory factor and threshold value dynamic modification are introduced, which overcomes the shortcomings of BP network that are slow convergent speed and local minimization to a certain degree. The model is applied in the prediction of petroleum fraction critical properties and a better predictive accuracy is obtained than the general nonlinear fitting methods.

关 键 词：人工神经网络 石油馏分 临界性质 预测 石油分析 

分 类 号：TE622.1[石油与天然气工程—油气加工工程] TP18[自动化与计算机技术—控制理论与控制工程]