基于优化神经网络预报的原油含水率测量  

Prediction method of water content ratio of crude oil based on optimized artificial neural network

作　　者：张勇军[1] 庄欣莉[1] 宋佳民[2] 

机构地区：[1]北京科技大学冶金工程研究院,北京100083 [2]大庆钻探工程公司,大庆163412

出　　处：《系统工程理论与实践》2011年第6期1112-1117,共6页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(60705017);北京市科委重大项目(D09030303780902)

摘　　要：在线密度法在原油含水率测量中有很强的实用价值,但存在着受现场不确定因素影响测量误差波动较大的缺点.为了提高含水率的测量精度和稳定性,将误差反向传播神经网络用于密度法计算含水率数学模型中,针对该算法收敛速度缓慢和易陷入局部极小点的缺点,提出了将模拟退火算法用于该模型的全局寻优,改进后的误差反向传播神经网络的误差预报值对密度法模型计算值进行修正.通过对离线实验数据的训练,该方法能够有效地提高在线快速含水率测定结果的准确性.On-line density method has a strong practical value in water content ratio measurement of crude oil,but the shortcoming is that the greater volatility of measurement error is affected by the uncertain factors in the field.In order to improve the accuracy of water content ratio and stability,the back-propagation neural network is used in density mathematical method of calculating water content.For the algorithm convergence speed is slow and easily get into the local minimum points,it is proposed simulated annealing algorithm for the optimization of the model.The error prediction of the improved back-propagation neural network is amended in the calculated value of density method model.By the training of the off-line experimental data,this method can effectively improve the on-line rapid determination accurate results of the water content ratio.

关 键 词：含水率测量 误差反向传播神经网络 模拟退火算法 误差预报 

分 类 号：TP216[自动化与计算机技术—检测技术与自动化装置]