田晓东,魏海峰,朱宝君
TIAN Xiao-dong1,2,WEI Hai-feng3

• 1:中国地质大学
• 2:中国石化勘探开发研究院
• 3:大庆油田有限责任公司采油四厂 北京100083
• 4:大庆油田有限责任公司勘探开发研究院

摘要(Abstract)：

目前,很多油田都已进入了特高含水期,此阶段的自然递减率变化受多种因素影响,所以很难用传统的数学方法表达其变化规律,并对其变化趋势进行预测。在分析了遗传算法和基本反向传播算法各自的优势和原理的基础上,针对前向网络反向传播算法收敛速度缓慢和易陷入局部极值点的缺点,将有全局寻优特性的遗传算法与反向传播算法有效地结合,提出了一种快速、高效的前向网络学习算法,即GA-BP算法。此方法在特高含水期油田自然递减率预测中的应用结果表明,此方法比基本BP算法具有更好的适应性,预测精度较高,能够较好地反映自然递减率与其影响因素之间的内在关系,所以利用改进BP神经网络方法预测特高含水期自然递减率是有效的、可行的。
At present,several oilfields have entered into ultrahigh water cut stage.During this period,natural decline variation is affected by many factors.So it is difficult to express and predict variation rules by conventional mathematics methods.Based on the analysis on advantage and theory of GA method and BP method,aiming at low convergence rate and easily falling into the local extreme point of BP method,GA with global optimum characteristics is effectively combined with BP method,and then GA-BP method is proposed which is a fast and high-effective forward network study method.The results of application in prediction for natural decline rate show that,improved BP method not only has better self-adapting and higher forecasting accuracy than former,but also can better reflect the inner relationship between natural decline rate and its affected factors,thus it is a effective and feasible method to predict natural decline rate during ultra-high water cut stage.

关键词(KeyWords)： 自然递减率;神经网络;遗传算法;反向传播算法
natural decline rate;neural network;genetic algorithm(GA);back-propagation(BP)

Abstract:

Keywords:

基金项目(Foundation): 中国石油股份公司“大庆喇萨杏油田特高含水期水驱优化调整配套技术”项目(040115)部分成果

作者(Author): 田晓东,魏海峰,朱宝君
TIAN Xiao-dong1,2,WEI Hai-feng3

参考文献(References)：

• [1]王俊魁.油田产量递减类型的判别与预测[J].大庆石油地质与开发,1991,10(4):27-34.
• [2]计秉玉.油田开发指标预测方法综述[J].大庆石油地质与开发,1999,18(2):19-22.
• [3]田晓东,吴晓慧,冮明立.水驱油田基于液量的分结构开发指标预测方法[J].大庆石油地质与开发,2002,21(3):35-37.
• [4]曲艳玲,杨昱杰.小层剩余油定量化分析方法[J].大庆石油地质与开发,2005,24(2):37-40.
• [5]李孝安.神经网络与神经计算机导论[M].西安:西北工业大学出版社,1994.
• [6]Holland J.Adaptation in natural and artificial systems[M].Universityof Michigan Press,1975.
• [7]Demuth H.Neural network toolbox[R].The Math works Inc,1997.
• [8]Ahn Ki-Hwan,Choi Yoon Kyung,Lee Soo-young.Pruned feed forwardnetworks for efficient implementation of multiple F IR filters with arbi-trary frequency responses[R].Neural processing letters 4,1998:221-227.
• [9]Bertels KL,Neuberg,Vassiliadis S,et al.Chaos and neural networklearn-ing some observations[R].Neural precede letters 7,1998:69-50.
• [10]Cho Siu-Yeung,CroowTommy W S.A fast neural learning vision sys-tem for crowd estimation at underground stations platform[R].Neuralprocess leters10,1999:111-120.
• [11]Yutaka Fukuoka,Hideo Matsuki,Haruyuki,et al.A modified back-propagation method to avoid false local minima[R].Neuralnetworksll,1998:1059-1072.
• [12]李士勇.模糊控制、神经控制和智能控制论[M].哈尔滨:哈尔滨工业大学出版社,1994.
• [13]王俊魁,王克成.神经网络建模、预报与控制[M].哈尔滨:哈尔滨工程大学出版社,1996.
• [14]Holland J H.Adaptation in nature and artificial systems[M].MITpress,1992.
• [15]陈国良,王煦法,庄镇泉,等.遗传算法及其应用[M].北京:人民邮电出版社,1996.