基于神经网络的蒙脱土悬浮液流变参数预测  

Prediction of montmorillonite suspensions rheological parameters based on neural network

作　　者：陶怀志[1] 蒲晓林[1] 王贵[1] 潘丽娟[1] 项朝刚[1] 

机构地区：[1]西南石油大学油气藏地质与开发工程国家重点实验室,四川成都610500

出　　处：《计算机与应用化学》2011年第11期1481-1484,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(50774065)

摘　　要：应用BP神经网络方法,研究高粘聚阴离子纤维素、羧甲基纤维素、磺化酚醛树脂以及磺化褐煤树脂四种常用处理剂对蒙脱土悬浮液表观粘度与塑性粘度的影响。采用弹性算法和最优停止法对神经网络进行优化,避免了过拟合现象,提高了神经网络的训练速度和泛化能力。以实际数据作为验证样本对神经网络模型进行检验,模型计算结果与实际结果比较,表观粘度的最大绝对误差率为8.75%,平均绝对误差率为1.12%;塑性粘度的最大绝对误差率为8.82%,平均绝对误差率为1.42%。最后运用该模型分析了单一处理剂对蒙脱土悬浮液流变参数的影响。A model was used to research four kinds of additives（PAC-HV,CMC,SMP and SPNH） effecting on apparent viscosity（AV） and plastic viscosity（PV） of montmorillonite suspensions,which was based on a backpropagation（BP） neural network.The methods of resilient propagation（RP） algorithm and optimal stopping rule were used to avoid overfitting and improve training speed as well as generalization ability.A test on real data showed the maximum absolute error of AV was 8.75%,the average absolute error of AV was 1,12%;the maximum absolute error of PV was 8.82%, the average absolute error of PV was 1.42%.At last,the model was used to analysis the single additive effecting on rheological parameters.

关 键 词：表观粘度 塑性粘度 神经网络 神经网络优化 泛化能力 

分 类 号：TG221[金属学及工艺—铸造]