摘要
储层属性参数预测是油藏描述的一项重要基础工作,克里金或神经网络插值法是较为有效的简便预测方法。针对克里金法具有平滑性、神经网络难以反映变量的空间相关性等缺点,用变异函数修正了神经网络的目标函数,并利用遗传算法对神经网络进行全局优化,形成了一种遗传神经克里金混合插值方法。应用实例表明,该方法具有神经网络较好的自学习、自适应能力;利用变异函数能较好地恢复数据的空间相关性;通过遗传算法全局优化,克服了神经网络目标函数容易陷入局部极小的缺点;对不同空间分布的样本点均能得到稳定的插值结果,具有比其他插值方法更高的精度和稳定性。
Reservoir property prediction is an important basic work in reservoir description. Kriging or neural network (NN) interpolation is a facile and effective method to do this work. As the Kriging method has smoothing effect and the NN method can 't guarantee the spatial correlation structure, an integrated genetic-neural-Kriging interpolation method was introduced. The objective function of NN was modified by variation function and the NN was trained of global optimization by genetic algorithm (GA) in this method. The application results show that the integrated GA-NN-Kriging algorithm has spatial correlation structure reconstruction ability of variation, retains the selfadaptaion of NN and can overcome the NN's shortcoming that is easy to fall into local minimum with global optimization of GA. For different spatial distribution simples, stable interpolation maps can be obtained by the method. The method shows high precision and stability compared with other interpolation algorithms.
出处
《中国石油大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2007年第5期35-40,共6页
Journal of China University of Petroleum(Edition of Natural Science)
关键词
储层属性
神经网络
变异函数
遗传算法
插值
预测
reservoir property
neural network
variation function
genetic algorithm
interpolation
prediction