基于改进SADE算法的神经网络预测储层物性  

A New Method Predicting Reservoir Properties with Neural Network Based on SADE Algorithm

作　　者：李虎[1,2] 范宜仁[1,2] 丛云海[1,2] 胡云云[1,2] 刘智中 

机构地区：[1]中国石油大学地球资源与信息学院,山东青岛266580 [2]中国石油大学CNPC测井重点实验室,山东青岛266580 [3]中国石油玉门油田分公司,甘肃酒泉735210

出　　处：《测井技术》2012年第6期585-589,共5页Well Logging Technology

基　　金：中国石油天然气集团公司科学研究与技术开发项目(2011D-4101);中国石油国家重大专项(2011ZX05020-008);国家自然基金资助项目(41174099)联合资助

摘　　要：为准确计算孔隙度、渗透率等储层物性参数,结合模拟退火和差分进化算法的主要优点,提出一种改进的模拟退火差分进化(SADE)算法,将复杂储层物性预测过程中神经网络权值的训练转化为无约束优化问题,并建立新目标函数,进而利用改进的SADE算法进行求解,并与传统方法计算结果进行比较。新目标函数使得神经网络权值的调整不受样本期望输出大小的影响,更适用于变化范围较大的样本数据训练;改进的SADE算法利用退火温度控制差分进化的选择过程和差分策略的选用,前期具有很好的多样性,后期有较好的收敛能力,克服了经典算法早熟的缺点,提高了全局搜索能力和鲁棒性。利用该算法对现场实际资料进行计算,取得了很好的效果。In order to accurately calculate reservoir properties, the improved Simulated Annealing Differential Evolution Algorithm （SADE） is proposed by combining simulated annaling with differential evolution algorithm. The training of neural network weights in the process of predicting complicated reservoir properties is transformed into an unconstrained optimization problem, and also a new objective function is offered. Then this problem can be solved by SADE algorithm. Compared with other traditional methods, the new objective function is independent of the desired output during the training of neural network, and thus is more suitable for large range of sample data. At the same time, the annealing temperature is used in the algorithm to control the selection process of differential evolution and the differential strategy. In the early stage, the algorithm is of good diversity, while in the late stage, it is of good convergence, overcoming the shortcoming of prematurity in the classical algorithm, and improving the general search ability and robustness. Finally we calculate the field reservoir properties with this algorithm, and obtain good effect.

关 键 词：测井评价 模拟退火 差分进化 神经网络 目标函数 储层物性预测 

分 类 号：TE122[石油与天然气工程—油气勘探]