GBP神经网络在改性石蜡性质预测中的应用  

APPLICATION OF GBP NEURAL NETWORK IN PREDICTING THE PROPERTIES OF MODIFIED PARAFFIN WAX

作　　者：张晓彤[1] 李兴明[1] 代敏[1] 刘丹[1] 刘荣江 孙兆林[1] 孙挺[3] 

机构地区：[1]辽宁石油化工大学石油化工学院,辽宁抚顺113001 [2]抚顺石油一厂,辽宁抚顺113008 [3]东北大学理学院,辽宁沈阳110004

出　　处：《石油化工高等学校学报》2004年第4期1-5,共5页Journal of Petrochemical Universities

基　　金：国家自然科学基金项目 ( 2 0 3 4 3 0 0 6 )

摘　　要：在误差反向传播神经网络算法中引入动量因子和非线性敏感度因子 ,并实现其在学习过程中根据整体误差的变化对非线性敏感度因子进行动态调整。首次采用GBP模型对改性石蜡的 2 5℃针入度和 1 0 0℃运动粘度进行了预测 ,2 5℃针入度预测的绝对误差 (A .D .)不超过± 0 .6 ( 1 0 -1 mm) ,标准偏差为 0 .5 4 ;1 0 0℃运动粘度预测的绝对误差 (A .D .)不超过± 1 .0 76mm2 /s ,标准偏差为 0 .5 8。结果表明 ,GBP神经网络算法适用于改性石蜡 2 5℃针入度和 1 0 0℃运动粘度的预测 ,可用于指导特种蜡产品的研制和生产 ,从而减少石蜡调合试验及相应的改性石蜡性质测试的次数 ,节省人力物力 。Momentum factor and nonlinear sensitivity factor were introduced in the modified back propagation neural network algorithm. The nonlinear sensitivity factor was adjusted according to the change of the integral deviation in the development. For the first time, 25 ℃ penetration degree and 100 ℃ kinematic viscosity of modified paraffin wax were predicted by using the generalized back propagation(GBP) neural network. The absolute deviation(A.D.) of 25 ℃ penetration degrees of the modified paraffin wax is in the range of ±0.6 units(a unit is equal to 10^(-1) mm) and the standard deviation(S.D.) is 0.54. The A.D. of 100 ℃ kinematic viscosities of the modified paraffin wax is in the range of ±1.076 mm^2/s and the S.D. is 0.58. The results prove that the modified BPNN is applicable to predict the properties of the modified paraffin wax. This method can be used to instruct the research and the production of special paraffin wax, which can decrease the times of the experiments and the testing of properties, save the manpower and material resource and shorten the cycle of the special paraffin wax development.

关 键 词：人工神经网络 石蜡改性 针入度 运动粘度 

分 类 号：TQ645[化学工程—精细化工]