en
×

分享给微信好友或者朋友圈

使用微信“扫一扫”功能。
作者简介:

吴君达(1996—),男,安徽合肥人,在读硕士研究生,从事油气田开发研究。E-mail:wujd510@163.com。

通讯作者:

李治平(1963—),男,四川泸州人,教授,博导。E-mail:2002011671@cugb.edu.cn。

中图分类号:TE319

文献标识码:A

文章编号:1009-9603(2020)04-0085-09

DOI:10.13673/j.cnki.cn37-1359/te.2020.04.010

参考文献 1
阳晓燕.非均质油藏水驱开发效果研究[J].特种油气藏,2019,26(2):152-156.YANG Xiaoyan.Waterflood development effect study of heteroge⁃ neous reservoir[J].Special Oil & Gas Reservoirs,2019,26(2):152-156.
参考文献 2
李明,戚楠,陈朝兵,等.甘谷驿油田L2区特低渗储层非均质性及其对油层分布的影响[J].油气藏评价与开发,2019,9(2):1-6.LI Ming,QI Nan,CHEN Chaobing,et al.Extra-low permeability reservoir heterogeneity and its effect on the distribution of reser⁃ voirs in L2 area of Ganguyi Oilfield[J].Reservoir Evaluation and Development,2019,9(2):1-6.
参考文献 3
陈宇航,朱增伍,王喆,等.鄂尔多斯盆地东南部长7油页岩时空分布及控制因素——来自沉积环境和沉积速率的制约[J].石油实验地质,2018,40(2):200-209,217.CHEN Yuhang,ZHU Zengwu,WANG Zhe,et al.Time-space dis⁃ tribution of Chang7 oil shale in southeastern Ordos Basin:Con⁃ trolled by sedimentary environments and deposition rates[J].Pe⁃ troleum Geology & Experiment,2018,40(2):200-209,217.
参考文献 4
郁文.考虑平面连通差异的杏六区西部注水细分标准[J].大庆石油地质与开发,2019,38(4):96-104.YU Wen.Subdividing standards of the water injection considering the planar connectivity difference for West Block Xing-6[J].Pe⁃ troleum Geology & Oilfield Development in Daqing,2019,38(4):96-104.
参考文献 5
朱丽红,杜庆龙,姜雪岩,等.陆相多层砂岩油藏特高含水期三大矛盾特征及对策[J].石油学报,2015,36(2):210-216.ZHU Lihong,DU Qinglong,JIANG Xueyan,et al.Characteristics and strategies of three major contradictions for continental facies multi-layered sandstone reservoir at ultra-high water cut stage [J].Acta Petrolei Sinica,2015,36(2):210-216.
参考文献 6
邹存友,王国辉,窦宏恩,等.油田开发效果评价方法与关键技术[J].石油天然气学报,2014,36(4):125-130,147.ZOU Cunyou,WANG Guohui,DOU Hong’en,et al.Method and key techniques for evaluating the effect of oilfield development [J].Journal of Oil and Gas Technology,2014,36(4):125-130,147.
参考文献 7
吕端川,吴伟.水驱开发断块型油藏注采参数调整——以大庆油田北二西F断裂两侧断块为例[J].断块油气田,2017,24(6):813-816.LÜ Duanchuan,WU Wei.Injection-production parameter adjust⁃ ment of water flooding development:taking two fault-blocks on both sides of F Fault of Daqing Oilfield Bei’erxi as examples[J].Fault-Block Oil and Gas Field,2017,24(6):813-816.
参考文献 8
严科,张俊,王本哲,等.平面非均质油藏均衡水驱调整方法研究[J].特种油气藏,2015,22(5):86-89.YAN Ke,ZHANG Jun,WANG Benzhe,et al.Balance water flood⁃ ing adjustment of plane heterogeneous reservoirs[J].Special Oil & Gas Reservoirs,2015,22(5):86-89.
参考文献 9
赵继勇,樊建明,何永宏,等.超低渗—致密油藏水平井开发注采参数优化实践——以鄂尔多斯盆地长庆油田为例[J].石油勘探与开发,2015,42(1):68-75.ZHAO Jiyong,FAN Jianming,HE Yonghong,et al.Optimization of horizontal well injection-production parameters for ultra-low permeable-tight oil production:A case from Changqing Oilfield,Ordos Basin[J].Petroleum Exploration and Development,2015,42(1):68-75.
参考文献 10
王锋.临盘油田复杂断块油藏注采井网优化[J].西安石油大学学报:自然科学版,2014,29(5):58-60,69.WANG Feng.Optimization of reasonable injection-production well pattern in complex fault block of Linpan Oilfield[J].Journal of Xi’an Shiyou University:Natural Science Edition,2014,29(5):58-60,69.
参考文献 11
杜殿发,巴忠臣,刘洋,等.稠油油藏活性水驱注采参数优化 [J].特种油气藏,2014,21(1):88-91.DU Dianfa,BA Zhongchen,LIU Yang,et al.Optimization of injec⁃ tion-production parameters for activated waterclooding in heavy oil reservoirs[J].Special Oil & Gas Reservoirs,2014,21(1):88-91.
参考文献 12
唐海,何娟,荣元帅,等.塔河断溶体油藏典型断溶体注水驱替规律及剩余油分布特征[J].油气地质与采收率,2018,25(3):95-100.TANG Hai,HE Juan,RONG Yuanshuai,et al.Study on water drive law and characteristics of remaining oil distribution of typi⁃ cal fault-karst in fault-karst reservoirs,Tahe Oilfield[J].Petro⁃ leum Geology and Recovery Efficiency,2018,25(3):95-100.
参考文献 13
王瑞,袁士宝,王建,等.复杂断块油藏注采耦合技术提高采收率机理[J].大庆石油地质与开发,2018,37(6):38-42.WANG Rui,YUAN Shibao,WANG Jian,et al.Mechanism of the injection-production coupling technique for complex fault-block oil reservoirs[J].Petroleum Geology & Oilfield Development in Daqing,2018,37(6):38-42.
参考文献 14
房娜,张占女,程明佳,等.基于不同裂缝发育程度下周期注水注采参数优化研究[J].特种油气藏,2019,26(1):131-135.FANG Na,ZHANG Zhannü,CHENG Mingjia,et al.Injection-pro⁃ duction parameter optimization study of cyclic water injection at different development levels of fractures[J].Special Oil & Gas Reservoirs,2019,26(1):131-135.
参考文献 15
张建国,刘锦华,何磊,等.水驱砂岩气藏型地下储气库长岩心注采实验研究[J].石油钻采工艺,2013,35(6):69-72.ZHANG Jianguo,LIU Jinhua,HE Lei,et al.Long core injectionproduction experiments study on water flooding sandstone gas res⁃ ervoir type underground gas storage[J].Oil Drilling & Production Technology,2013,35(6):69-72.
参考文献 16
周涌沂,汪勇,田同辉,等.改善平面非均质油藏水驱效果方法研究[J].西南石油大学学报:自然科学版,2007,29(4):82-85.ZHOU Yongyi,WANG Yong,TIAN Tonghui,et al.Research on the method of improving water-flooding effect for the reservoirs with plane heterogeneity[J].Journal of Southwest Petroleum Uni⁃ versity:Science & Technology Edition,2007,29(4):82-85.
参考文献 17
潘伟义,王友启,张诗洋,等.晚期注水开发实验研究[J].油气地质与采收率,2018,25(3):122-126.PAN Weiyi,WANG Youqi,ZHANG Shiyang,et al.An experimen⁃ tal study on late water flooding[J].Petroleum Geology and Recov⁃ ery Efficiency,2018,25(3):122-126.
参考文献 18
王东英,任熵,蒋明洁,等.非均质油藏井网注采参数优化的可视化模拟驱替实验[J].油气地质与采收率,2013,20(2):95-98.WANG Dongying,REN Shang,JIANG Mingjie,et al.Visual simu⁃ lation study on optimization of well pattern injection and produc⁃ tion parameters in heterogeneous reservoirs[J].Petroleum Geolo⁃ gy and Recovery Efficiency,2013,20(2):95-98.
参考文献 19
陈霆.非均质储层模型微观水驱油实验[J].油气地质与采收率,2007,14(4):72-75.CHEN Ting.Experimental study of water-oil displacement with microscopic models of heterogeneous reservoirs[J].Petroleum Ge⁃ ology and Recovery Efficiency,2007,14(4):72-75.
参考文献 20
ANIFOWOSE F,LABADIN J,ABDULRAHEEM A.Improving the prediction of petroleum reservoir characterization with a stacked generalization ensemble model of support vector machines[J].Ap⁃ plied Soft Computing,2015,26:483-496.
参考文献 21
CHITHRA Chakra N,SONG K Y,GUPTA M M,et al.An innova⁃ tive neural forecast of cumulative oil production from a petroleum reservoir employing higher-order neural networks(HONNs)[J].Journal of Petroleum Science and Engineering,2013,106:18-33.
参考文献 22
侯春华.基于长短期记忆神经网络的油田新井产油量预测方法[J].油气地质与采收率,2019,26(3):105-110.HOU Chunhua.New well oil production forecast method based on long-term and short-term memory neural network[J].Petroleum Geology and Recovery Efficiency,2019,26(3):105-110.
参考文献 23
ANIFOWOSE F A,LABADIN J,ABDULRAHEEM A.Ensemble machine learning:An untapped modeling paradigm for petroleum reservoir characterization[J].Journal of Petroleum Science and Engineering,2017,151:480-487.
参考文献 24
白玉湖.不确定性致密气产能预测技术[J].中国石油勘探,2018,23(4):107-113.BAI Yuhu.Uncertain productivity prediction technology for tight gas[J].China Petroleum Exploration,2018,23(4):107-113.
参考文献 25
GRAVES A,LIWICKI M,SANTIAGO Fernández,et al.A novel connectionist system for unconstrained handwriting recognition [J].IEEE Transactions on Pattern Analysis and Machine Intelli⁃ gence,2009,31(5):855-868.
参考文献 26
FARABET C,COUPRIE C,NAJMAN L,et al.Learning hierarchi⁃ cal features for scene labeling[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,35(8):1 915-1 929.
参考文献 27
ZEILER M D,KRISHNAN D,TAYLOR G W,et al.Deconvolu⁃ tional networks[C].San Francisco:Computer Society Conference on Computer Vision and Pattern Recognition,2010.
参考文献 28
ZEILER M D,TAYLOR G W,FERGUS R.Adaptive deconvolu⁃ tional networks for mid and high-level feature learning[C].Barce⁃ lona:International Conference on Computer Vision,2012.
参考文献 29
STORN R,PRICE K.Differential evolution-A simple and efficient heuristic for global optimization over continuous spaces[J].Jour⁃ nal of Global Optimization,1997,11(4):341-359.
参考文献 30
汪慎文,丁立新,张文生,等.差分进化算法研究进展[J].武汉大学学报:理学版,2014,60(4):283-292.WANG Shenwen,DING Lixin,ZHANG Wensheng,et al.Survey of differential evolution[J].Journal of Wuhan University:Natural Science Edition,2014,60(4):283-292.
目录contents

    摘要

    针对注水开发过程中注采参数的优化问题,提出采用神经网络代替数值模拟对剩余油分布进行预测,并结合无梯度差分进化算法对注采参数进行优化。该模型不仅建立了注采参数与目标函数的非线性关系,还能准确预测不同生产阶段剩余油分布。其预测原理是将注采参数和生产时间视为剩余油分布图像的高级特征,利用卷积层提取特征、转置卷积层进行上采样,通过多个卷积与转置卷积的组合逐级恢复原图像,从而达到准确预测的效果。在神经网络构建过程中,选择多个3×3的小卷积核来代替大卷积核,在不影响感受野的情况下减少了参数量,节约了计算成本,有效提高了模型训练时的迭代效率。以某区块4口注入井、5口生产井的五点井网为例,将不同阶段生产井的井底压力、注入井的注入量以及生产时间作为输入参数,建立了基于神经网络的预测模型,以净现值作为目标函数,通过差分优化算法对 4 个阶段的注采参数进行了优化。相比于基础方案,优化后的方案净现值提高了约 21%。

    Abstract

    Aiming at the optimization of injection-production parameters in the process of waterflooding development,it is proposed to use neural network instead of numerical simulation to predict the distribution of remaining oil,and combine the gradient-free differential evolution algorithm to optimize injection and production parameters. This model not only estab- lishes the nonlinear relationship between injection-production parameters and the objective function,but also accurately predicts the distribution characteristics of remaining oil in different production stages. The prediction principle of the model is that the injection-production parameters and production time are regarded as advanced features of the remaining oil dis- tribution image. Then,the convolution layer is used to extract image features and the transposed convolution layer is used to carry out up-sampling on the image. The original image is recovered step by step through a combination of multi-layer “convolution+transpose convolution”,so as to achieve the effect of accurate prediction. During the construction of the neu- ral network,multiple 3×3 small convolution kernels are selected to replace the large convolution kernels. Without affecting the receptive field,the amount of parameters is reduced,and the calculation cost is saved. The iteration efficiency during model training is effectively improved. Taking the five-point well pattern of 4 injection wells and 5 production wells in a block as an example,a neural network prediction model was established by taking the bottom hole pressure of the produc- tion well,the injection volume of the injection well and production time as input parameters. Taking the net present value (NPV)as the objective function,the injection-production parameters of the four stages were optimized through the differen- tial optimization algorithm. Compared with the basic plan,the NPV of the optimized plan has increased by about 21%.

  • 在注水开发过程中,注水井和采油井注采参数的确定一直是个难题,不同注采参数对生产井的产量、见水时间及最终采收率等具有极大影响[1],在非均质性较强的储层中,由于油层分布不均[2-3],储层渗透率差异较大[4] 等原因,注采参数的影响尤为突出[5-6]

  • 常见的确定注采参数的方法有经验法[7-9]、数值模拟法[10-14] 及室内实验法[15-17],然而这些方法都存在着各自的局限性。经验法只能定性确定注采参数,具有很强的主观性,且对于不同地质特征的油藏适应性较差;数值模拟法对较小模型的模拟效果较好,较大模型的模拟需耗费大量计算时间,不便于现场应用;室内实验法可以直观的观测注水开发过程中的渗流规律[18-19],但受实验设备限制,很难模拟真实油藏的开发过程。

  • 近年来,许多学者引入数学和机器学习方法对实际的注采参数进行了预测和优化,如支持向量机[20]、神经网络[21-22]、集成学习[23]、不确定性分析[24] 等。通常这些预测模型在经过大量的油田实际数据训练后,可以代替数值模拟对产量、采收率等生产数据进行较为准确且快速预测,但是大多数预测模型只建立了注采参数与待优化目标的单一非线性关系,无法预测剩余油分布,也无法得到地层压力等动态参数,仍存在一定局限性。为此,引入图像处理的方法,运用神经网络代替数值模拟对研究区的生产动态进行预测,可以较为准确地描述不同开发方式下剩余油的分布规律,再通过进化差分算法对注采参数进行优化。

  • 1 神经网络理论基础

  • 卷积神经网络是目前机器学习领域中极具代表性的神经网络之一,其在图像特征的提取分类[25]、场景识别[26] 等领域取得了众多进展。卷积神经网络相较于传统的神经网络,其最大的优势在于神经元能通过共享卷积核的方式来共享权重,大大减少了需要计算的参数数量,简化了网络结构。以卷积对图像的处理方法作为核心框架,通过卷积层及转置卷积层完成了神经网络的架构。

  • 1.1 卷积层

  • 卷积层由多个卷积单元即卷积核构成,采用权值共享的方法,通过反向传播得到每个卷积单元最佳参数,其目的是提取输入图像或数据的不同特征。每个卷积核按照由左到右、由上到下的滑动方式对图像或数据进行卷积操作,从而得到输入图像的全部特征。一般单层的卷积只能提取一些较低级的图像特征,若要发掘输入图像的高级特征可通过设计多层卷积来实现。

  • 1.2 转置卷积层

  • 转置卷积是 ZEILER 提出的一种上采样方法,可以视为卷积的逆过程,一般将其运用于对图像的像素级别分类或语义分割[27-28]。通常的卷积运算会形成多输入对单输出的关系,而转置卷积的思想则是通过一个转置卷积核使单个输入与多个输出形成对应关系,这也是转置卷积的核心思想,即将抽象特征还原为原图像的过程。与其他如双线性插值等上采样方法最大的不同在于其卷积核参数可以经过反向传播学习调整,从而使转置卷积的上采样参数更加合理。

  • 2 数据处理及预测模型构建

  • 利用注采制度(不同生产阶段的注采参数组合)及生产时间来预测剩余油的分布规律,将井底压力、注入量、产液量等注采制度及生产时间视为剩余油分布的高级图像特征,通过卷积层提取学习图像特征后,再应用转置卷积层进行上采样,即对卷积学习到的特征进行像素级别的分类,从而将多个图像特征还原到原图大小。

  • 2.1 输入数据处理

  • 在注水开发过程中,不合理的注采参数会导致生产井快速见水,从而影响最终采收率与经济效益。注采参数通常会根据现场经验及生产情况进行调整,因此在模拟中将其设置为多阶段的注采制度。假定共有 N 个生产阶段,每个生产阶段共有 M 个注采参数 α。则第 k个生产阶段的注采参数矩阵为:

  • Sk=α1k,α2k,,αhk,,αM-1k,αMk
    (1)
  • 在训练神经网络过程中,当训练生产时间满足 tk< t< tk+1时,为防止tk+1后的注采参数与输出的剩余油分布产生无意义的关联,将第 k 个注采制度之后的注采参数矩阵取O,得到不同生产时刻、不同参数组合下的输入参数,其表达式为:

  • Ik=S1,S2,,Sk,Sk+1O,,Sn-1O,SnO,tk
    (2)
  • 2.2 数据归一化处理

  • 由于注采参数的数量级各不相同,为避免由于数量级引起的数值和梯度问题,对输入样本的注采参数进行归一化处理,其处理步骤如下:

  • a-=1nis=1n ais
    (3)
  • σ=1n-1is=1n ais-a-2
    (4)
  • bis=ai-a-σis=1,2,,n
    (5)
  • 2.3 神经网络预测模型构建

  • 本文构建的神经网络模型如图1所示。该模型的第一个隐藏层为全连接层,有效避免了不同生产参数仅对复原图像局部区域产生影响,解决了预测结果缺乏空间一致性的问题。卷积层选择3×3的卷积核,通过多个 3×3 卷积核堆叠来代替大尺寸的卷积核,在不影响感受野的情况下可以有效的减少需要计算的参数个数,节约计算成本,加快模型训练迭代效率。在卷积提取了图像特征后,通过转置卷积进行上采样恢复图像。与常规的双线性差值上采样相比,转置卷积预测的总体质量更好。

  • 由于油田面积较大,网格数较多,单一的卷积与转置卷积组合预测效果不佳,所以采用三层卷积与转置卷积组合来加深神经网络层数,逐步提取图像特征并恢复原图像,提高预测准确度。

  • 3 注采参数优化

  • 通过数值模拟计算出不同注采参数下的方案样本,将其代入神经网络模型中进行训练,首先建立预测模型代替数值模拟从而节约计算成本,减少计算时间,再选取合适的目标函数,最后采用差分进化算法对注采参数进行优化。

  • 3.1 目标函数

  • 不同注采制度对油田的实际开发具有较大影响,通常采用经济效益指标作为衡量标准。为获取最优的注采参数,选取净现值作为目标函数,注采制度为待优化参数:

  • NPVα1,α2,,αM-1,αM=t=1n (CI-CO)t1+ic-t=t=1n Q0tvo-Qw1tvw1-Qw2tvw21+ict
    (6)
  • 通过神经网络预测不同注采制度、不同生产时间的剩余油分布,可以得到不同时间段的产油量与产水量,从而建立了注采参数和目标函数的非线性函数关系。

  • 3.2 差分进化算法

  • 由于净现值目标函数参数较多、复杂程度较高、局部最优解较多,为此引入无梯度的差分进化算法(DE)。该算法是 STORN等提出的一种基于群体差异的启发式并行搜索方法[29]。作为一种基于群体导向的随机搜索技术,与其他优化算法如遗传算法PSO相比,DE算法的优点在于进化个体的扰动是通过多个个体的差分信息来体现,其个体间的差异大、扰动大,使算法的搜索范围更广,不容易陷入局部最优解[30]。DE算法主要包括初始化、变异、交叉和选择。

  • 图1 神经网络预测模型构建流程

  • Fig.1 Construction process of neural network prediction model

  • 第一步,初始化。为了初始化种群,随机产生第0代种群:

  • xi(0)xi,jLxi,j(0)xi,jUi=1,2,,NP;j=1,2,,D
    (7)
  • 其中:

  • xi,j(0)=xi,jL+rand(0,1)xi,jU-xi,jL
    (8)
  • 第二步,变异。DE算法通过差分策略实现个体变异,常见的差分策略是随机选取种群中 2 个不同的个体,将其向量差缩放后与待变异个体进行向量合成:

  • Vi(g+1)=xr1(g)+Fxr2(g)-xr3(g)
    (9)
  • 第三步,交叉。通过随机选取个体的方式来提高种群的多样性,其表达式为:

  • Ui,j(g+1)=Vi,j(g+1)rand(0,1)CRxi,j(g)rand(0,1)>CR
    (10)
  • 第四步,选择。DE 算法采用贪婪选择的策略,即选择适应度较高的个体作为新个体,对于最大化优化问题,其选择算子表达式为:

  • xi(g+1)=Ui(g+1) fUi(g+1)fxi(g)xi(g) fUi(g+1)<fxi(g)
    (11)
  • 通过选择算子使子代个体的适应度始终优于父代的适应度,种群始终向最优解的位置逼近,直到寻找到最优解。其优化设计步骤如图2所示。

  • 4 应用实例

  • 4.1 预测模型训练

  • 选取某区块地质模型作为建模依据,其渗透率 (K)、孔隙度分布场如图3所示。建立四注五采的五点井网模型作为预测和优化的研究对象。该模型由 60×60×1 个网格组成,每个网格的尺寸为 10 m× 10 m×10 m。生产井定井底流压生产,注入井定注入量注水,以200 d为一个生产制度共进行4个阶段为期 800 d 的生产。油藏原始地层压力为 25 MPa,油藏初始含油饱和度为 0.8,地层原油黏度为 1.2 mPa·s,原油密度为 0.8 g/cm3,体积系数为 1.15。原油价格为 1 800 元/m3,产出水处理价格为 330 元/m3,注入水价格为 165 元/m3。随机选取 1 500 组不同的注采制度,以25 d为步长,运用MRST数值模拟软件计算得到不同生产时间、不同工作制度下的 48 000个数据样本,选取38 400个样本作为训练集, 9 600个样本作为测试集,代入神经网络进行训练。

  • 图2 基于神经网络的差分进化算法优化流程

  • Fig.2 Optimization process of differential evolution algorithm based on neural network

  • 图3 渗透率及孔隙度分布场

  • Fig.3 Distribution fields of permeability and porosity

  • 在实际生产过程中,近井地带的含油饱和度变化较大,如果选取常见的均方误差(MSE)函数或者平均绝对误差(MAE)函数,模型对于远井地带饱和度变化的敏感性会大大降低。为此,模拟中选取平滑平均绝对误差(Huber loss)函数作为损失函数,其函数表达式为:

  • Lδ(y,f(x))=12(y-f(x))2 |y-f(x)|δδ|y-f(x)|-12δ2 |y-f(x)|>δ
    (12)
  • Huber loss函数在结合了MSE和MAE两种损失函数优点的基础上,降低了对离群点的惩罚程度,增强了模型对近井地带离群点的鲁棒性,进而能够更准确地预测研究区的剩余油分布规律。

  • 由于数据样本较多,仅在神经网络中传递一次完整的数据集是远远不够的,多轮次数据集的传递可以使神经网络中的权重更新次数增加,并避免欠拟合的现象。与此同时,训练过程中学习率的选择也会极大地影响模型的训练效果。学习率过低会使训练过慢,学习率过高虽然能让模型在前期更容易接近局部或全局最优解,但在后期会在最优解附近波动,难以达到最优效果。随训练过程的进行,学习率逐渐降低,当评价指标不再提升,即学习停滞时,减少2~10倍的学习率,从而获得更好的效果。由递减学习率与固定学习率的训练过程(图4)可以看出,使用递减学习率在训练初期能够较快速收敛,在训练后期也能有效避免梯度在最小值附近震荡,训练效果较好,最终模型的训练及测试损失约为6×10-5

  • 图4 不同学习率下的神经网络预测模型训练及测试损失

  • Fig.4 Neural network prediction model training and test loss under different learning rates

  • 4.2 差分进化算法寻优

  • 建立研究区剩余油分布的预测模型后,将净现值作为待优化的目标函数,将每个阶段 9 口井的生产和注水制度作为待优化变量(4 个阶段共计 36个),采用差分进化算法找到最大经济效益对应的各阶段的注采参数。

  • 在优化过程中,交叉和变异决定了收敛速度的大小和优化效果的优劣。本次模拟中的缩放因子取值为 0.7,交叉概率取值为 0.5,种群规模设为 50,进化代数为150代,由差分进化算法寻优过程(图5) 可以看出,当差分进化约为100代时,NPV基本达到最优值,约为9.45×107 元。

  • 图5 差分进化算法寻优过程示意

  • Fig.5 Optimization process of differential evolution algorithm

  • 4.3 优化结果对比

  • 优化模拟以采油井井底压力为15 MPa,注入井注入量为 150 m3 /d 作为基础方案,其中真实的剩余油分布场由 mrst数值模拟软件计算得出,预测的剩余油分布场由预测模型计算得出。分析各生产阶段的经济效益和剩余油的分布规律。

  • 最优方案的注采参数如表1所示,从图6和图7 可以看出,最优方案和基础方案各阶段预测和真实的剩余油分布相似,单个网格的含油饱和度预测的最大误差不超过 3%,说明该模型可以较好地代替数值模拟对剩余油分布进行预测,可以较准确地反映不同注采制度、不同生产时间下的剩余油分布规律,并且大大减少了计算量。生产 800 d后,最优方案与基础方案的研究区平均含油饱和度分别为 59.77%和56.76%。与基础方案相比,经差分进化算法寻优后所得到的最优注采方案虽然采收率较低,但总注水量仅为基础方案的1/2,NPV提高了约21% (图8),在投入更小的情况下提高了经济效益,改善了开发效果。

  • 表1 最优方案各阶段注采参数

  • Table1 Injection-production parameters of each stage of optimal scheme

  • 图6 不同生产时间下最优方案的真实(左)及预测(右)剩余油分布对比

  • Fig.6 Comparison of actual(left)and predicted(right)remaining oil distributions of optimal scheme at different production times

  • 图7 不同生产时间下基础方案的真实(左)及预测(右)剩余油分布对比

  • Fig.7 Comparison of actual(left)and predicted(right)remaining oil distributions of base scheme at different production times

  • 图8 基础方案和最优方案净现值对比

  • Fig.8 NPV comparison of basic scheme and optimal scheme

  • 5 结论

  • 基于神经网络的剩余油分布预测模型,通过输入注采制度和生产时间得到不同生产阶段的剩余油分布。引入无梯度差分进化算法,将注采制度作为优化变量,净现值作为优化目标,基于神经网络的剩余油分布预测模型,对注采参数进行了优化。研究结果表明,该预测模型可以较好的代替数值模拟对剩余油分布进行预测,单个网格的含油饱和度预测的最大误差不超过 3%,解决了传统数值模拟运算量过大的问题,大大节约了运算时间。

  • 经实例验证,差分进化算法对注采参数优化问题适用性较高,收敛和寻优速度较快,减少了参数选取的计算时间。与基础方案对比,寻优后的最优方案以较小的注水量获得了更高的经济效益,并延缓了注水时间,改善了开发效果。

  • 符号解释

  • a-——输入样本的平均值;

  • ais ——第is个输入样本;

  • bis——第is个归一化处理后的输入样本;

  • CI——现金流入,107 元;

  • CO——现金流出,107 元;

  • CR ——交叉概率;

  • D ——个体维度;

  • fx)——样本预测值;

  • fUig+1fxig——个体 Uig+1xig 的适应度;

  • F ——缩放因子;

  • h ——第h个注采参数;

  • i ——第i个个体;

  • ic ——基准收益;

  • is ——第is个样本;

  • Ik ——神经网络输入参数;

  • j ——第j个维度;

  • k ——第k个生产阶段;

  • K ——渗透率;

  • Lδyfx——样本实际值和误差值的平滑平均绝对误差;

  • M——注采参数的总数;

  • n ——输入样本数;

  • N ——总生产阶段;

  • NPV ——净现值,107 元;

  • NP ——种群规模;

  • O ——零矩阵;

  • Qot——t-1到t时间内的产油量,m3

  • Qw1t——t -1到t时间内的产水量,m3

  • Qw2t——t-1到t时间内的注水量,m3

  • rand(0,1)——在[ 0,1 ]区间上的随机数;

  • Sk ——第k个生产阶段的注采参数矩阵;

  • t ——时间步长,25 d;

  • tk ——第k个生产阶段的生产时间,d;

  • Uig+1——第g + 1代种群中经交叉的第i个个体;

  • Uijg+1——第g + 1代种群中第i个个体经交叉后的第j个维度;

  • v o——原油价格,元/m3

  • v w1——产出水处理价格,元/m3

  • v w2——注入水价格,元/m3

  • Vig+1——第g + 1代种群中第i个变异后的个体;

  • Vijg+1——第g + 1代种群中第i个变异后个体的第j 个维度;

  • xi( 0)——第0代的第i个个体;

  • xi,j(0)——第0代的第i个个体的第j个维度;

  • xigxig+1——第gg + 1代种群中的第i个个体;

  • xi,jg)——第g代种群中第i个个体的第j个维度;

  • xr1gxr2gxr3g——第g代种群中随机的3个个体;

  • r1r2r3——在[ 1,NP ]区间上的随机数;

  • xijL——第i个个体第j维的下界;

  • xijU——第i个个体第j维的上界;

  • y ——样本实际值;

  • α——注采参数;

  • αhk——第k个生产阶段的第h个注采参数;

  • δ——超参数;

  • σ——输入样本的标准差。

  • 参考文献

    • [1] 阳晓燕.非均质油藏水驱开发效果研究[J].特种油气藏,2019,26(2):152-156.YANG Xiaoyan.Waterflood development effect study of heteroge⁃ neous reservoir[J].Special Oil & Gas Reservoirs,2019,26(2):152-156.

    • [2] 李明,戚楠,陈朝兵,等.甘谷驿油田L2区特低渗储层非均质性及其对油层分布的影响[J].油气藏评价与开发,2019,9(2):1-6.LI Ming,QI Nan,CHEN Chaobing,et al.Extra-low permeability reservoir heterogeneity and its effect on the distribution of reser⁃ voirs in L2 area of Ganguyi Oilfield[J].Reservoir Evaluation and Development,2019,9(2):1-6.

    • [3] 陈宇航,朱增伍,王喆,等.鄂尔多斯盆地东南部长7油页岩时空分布及控制因素——来自沉积环境和沉积速率的制约[J].石油实验地质,2018,40(2):200-209,217.CHEN Yuhang,ZHU Zengwu,WANG Zhe,et al.Time-space dis⁃ tribution of Chang7 oil shale in southeastern Ordos Basin:Con⁃ trolled by sedimentary environments and deposition rates[J].Pe⁃ troleum Geology & Experiment,2018,40(2):200-209,217.

    • [4] 郁文.考虑平面连通差异的杏六区西部注水细分标准[J].大庆石油地质与开发,2019,38(4):96-104.YU Wen.Subdividing standards of the water injection considering the planar connectivity difference for West Block Xing-6[J].Pe⁃ troleum Geology & Oilfield Development in Daqing,2019,38(4):96-104.

    • [5] 朱丽红,杜庆龙,姜雪岩,等.陆相多层砂岩油藏特高含水期三大矛盾特征及对策[J].石油学报,2015,36(2):210-216.ZHU Lihong,DU Qinglong,JIANG Xueyan,et al.Characteristics and strategies of three major contradictions for continental facies multi-layered sandstone reservoir at ultra-high water cut stage [J].Acta Petrolei Sinica,2015,36(2):210-216.

    • [6] 邹存友,王国辉,窦宏恩,等.油田开发效果评价方法与关键技术[J].石油天然气学报,2014,36(4):125-130,147.ZOU Cunyou,WANG Guohui,DOU Hong’en,et al.Method and key techniques for evaluating the effect of oilfield development [J].Journal of Oil and Gas Technology,2014,36(4):125-130,147.

    • [7] 吕端川,吴伟.水驱开发断块型油藏注采参数调整——以大庆油田北二西F断裂两侧断块为例[J].断块油气田,2017,24(6):813-816.LÜ Duanchuan,WU Wei.Injection-production parameter adjust⁃ ment of water flooding development:taking two fault-blocks on both sides of F Fault of Daqing Oilfield Bei’erxi as examples[J].Fault-Block Oil and Gas Field,2017,24(6):813-816.

    • [8] 严科,张俊,王本哲,等.平面非均质油藏均衡水驱调整方法研究[J].特种油气藏,2015,22(5):86-89.YAN Ke,ZHANG Jun,WANG Benzhe,et al.Balance water flood⁃ ing adjustment of plane heterogeneous reservoirs[J].Special Oil & Gas Reservoirs,2015,22(5):86-89.

    • [9] 赵继勇,樊建明,何永宏,等.超低渗—致密油藏水平井开发注采参数优化实践——以鄂尔多斯盆地长庆油田为例[J].石油勘探与开发,2015,42(1):68-75.ZHAO Jiyong,FAN Jianming,HE Yonghong,et al.Optimization of horizontal well injection-production parameters for ultra-low permeable-tight oil production:A case from Changqing Oilfield,Ordos Basin[J].Petroleum Exploration and Development,2015,42(1):68-75.

    • [10] 王锋.临盘油田复杂断块油藏注采井网优化[J].西安石油大学学报:自然科学版,2014,29(5):58-60,69.WANG Feng.Optimization of reasonable injection-production well pattern in complex fault block of Linpan Oilfield[J].Journal of Xi’an Shiyou University:Natural Science Edition,2014,29(5):58-60,69.

    • [11] 杜殿发,巴忠臣,刘洋,等.稠油油藏活性水驱注采参数优化 [J].特种油气藏,2014,21(1):88-91.DU Dianfa,BA Zhongchen,LIU Yang,et al.Optimization of injec⁃ tion-production parameters for activated waterclooding in heavy oil reservoirs[J].Special Oil & Gas Reservoirs,2014,21(1):88-91.

    • [12] 唐海,何娟,荣元帅,等.塔河断溶体油藏典型断溶体注水驱替规律及剩余油分布特征[J].油气地质与采收率,2018,25(3):95-100.TANG Hai,HE Juan,RONG Yuanshuai,et al.Study on water drive law and characteristics of remaining oil distribution of typi⁃ cal fault-karst in fault-karst reservoirs,Tahe Oilfield[J].Petro⁃ leum Geology and Recovery Efficiency,2018,25(3):95-100.

    • [13] 王瑞,袁士宝,王建,等.复杂断块油藏注采耦合技术提高采收率机理[J].大庆石油地质与开发,2018,37(6):38-42.WANG Rui,YUAN Shibao,WANG Jian,et al.Mechanism of the injection-production coupling technique for complex fault-block oil reservoirs[J].Petroleum Geology & Oilfield Development in Daqing,2018,37(6):38-42.

    • [14] 房娜,张占女,程明佳,等.基于不同裂缝发育程度下周期注水注采参数优化研究[J].特种油气藏,2019,26(1):131-135.FANG Na,ZHANG Zhannü,CHENG Mingjia,et al.Injection-pro⁃ duction parameter optimization study of cyclic water injection at different development levels of fractures[J].Special Oil & Gas Reservoirs,2019,26(1):131-135.

    • [15] 张建国,刘锦华,何磊,等.水驱砂岩气藏型地下储气库长岩心注采实验研究[J].石油钻采工艺,2013,35(6):69-72.ZHANG Jianguo,LIU Jinhua,HE Lei,et al.Long core injectionproduction experiments study on water flooding sandstone gas res⁃ ervoir type underground gas storage[J].Oil Drilling & Production Technology,2013,35(6):69-72.

    • [16] 周涌沂,汪勇,田同辉,等.改善平面非均质油藏水驱效果方法研究[J].西南石油大学学报:自然科学版,2007,29(4):82-85.ZHOU Yongyi,WANG Yong,TIAN Tonghui,et al.Research on the method of improving water-flooding effect for the reservoirs with plane heterogeneity[J].Journal of Southwest Petroleum Uni⁃ versity:Science & Technology Edition,2007,29(4):82-85.

    • [17] 潘伟义,王友启,张诗洋,等.晚期注水开发实验研究[J].油气地质与采收率,2018,25(3):122-126.PAN Weiyi,WANG Youqi,ZHANG Shiyang,et al.An experimen⁃ tal study on late water flooding[J].Petroleum Geology and Recov⁃ ery Efficiency,2018,25(3):122-126.

    • [18] 王东英,任熵,蒋明洁,等.非均质油藏井网注采参数优化的可视化模拟驱替实验[J].油气地质与采收率,2013,20(2):95-98.WANG Dongying,REN Shang,JIANG Mingjie,et al.Visual simu⁃ lation study on optimization of well pattern injection and produc⁃ tion parameters in heterogeneous reservoirs[J].Petroleum Geolo⁃ gy and Recovery Efficiency,2013,20(2):95-98.

    • [19] 陈霆.非均质储层模型微观水驱油实验[J].油气地质与采收率,2007,14(4):72-75.CHEN Ting.Experimental study of water-oil displacement with microscopic models of heterogeneous reservoirs[J].Petroleum Ge⁃ ology and Recovery Efficiency,2007,14(4):72-75.

    • [20] ANIFOWOSE F,LABADIN J,ABDULRAHEEM A.Improving the prediction of petroleum reservoir characterization with a stacked generalization ensemble model of support vector machines[J].Ap⁃ plied Soft Computing,2015,26:483-496.

    • [21] CHITHRA Chakra N,SONG K Y,GUPTA M M,et al.An innova⁃ tive neural forecast of cumulative oil production from a petroleum reservoir employing higher-order neural networks(HONNs)[J].Journal of Petroleum Science and Engineering,2013,106:18-33.

    • [22] 侯春华.基于长短期记忆神经网络的油田新井产油量预测方法[J].油气地质与采收率,2019,26(3):105-110.HOU Chunhua.New well oil production forecast method based on long-term and short-term memory neural network[J].Petroleum Geology and Recovery Efficiency,2019,26(3):105-110.

    • [23] ANIFOWOSE F A,LABADIN J,ABDULRAHEEM A.Ensemble machine learning:An untapped modeling paradigm for petroleum reservoir characterization[J].Journal of Petroleum Science and Engineering,2017,151:480-487.

    • [24] 白玉湖.不确定性致密气产能预测技术[J].中国石油勘探,2018,23(4):107-113.BAI Yuhu.Uncertain productivity prediction technology for tight gas[J].China Petroleum Exploration,2018,23(4):107-113.

    • [25] GRAVES A,LIWICKI M,SANTIAGO Fernández,et al.A novel connectionist system for unconstrained handwriting recognition [J].IEEE Transactions on Pattern Analysis and Machine Intelli⁃ gence,2009,31(5):855-868.

    • [26] FARABET C,COUPRIE C,NAJMAN L,et al.Learning hierarchi⁃ cal features for scene labeling[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,35(8):1 915-1 929.

    • [27] ZEILER M D,KRISHNAN D,TAYLOR G W,et al.Deconvolu⁃ tional networks[C].San Francisco:Computer Society Conference on Computer Vision and Pattern Recognition,2010.

    • [28] ZEILER M D,TAYLOR G W,FERGUS R.Adaptive deconvolu⁃ tional networks for mid and high-level feature learning[C].Barce⁃ lona:International Conference on Computer Vision,2012.

    • [29] STORN R,PRICE K.Differential evolution-A simple and efficient heuristic for global optimization over continuous spaces[J].Jour⁃ nal of Global Optimization,1997,11(4):341-359.

    • [30] 汪慎文,丁立新,张文生,等.差分进化算法研究进展[J].武汉大学学报:理学版,2014,60(4):283-292.WANG Shenwen,DING Lixin,ZHANG Wensheng,et al.Survey of differential evolution[J].Journal of Wuhan University:Natural Science Edition,2014,60(4):283-292.

  • 参考文献

    • [1] 阳晓燕.非均质油藏水驱开发效果研究[J].特种油气藏,2019,26(2):152-156.YANG Xiaoyan.Waterflood development effect study of heteroge⁃ neous reservoir[J].Special Oil & Gas Reservoirs,2019,26(2):152-156.

    • [2] 李明,戚楠,陈朝兵,等.甘谷驿油田L2区特低渗储层非均质性及其对油层分布的影响[J].油气藏评价与开发,2019,9(2):1-6.LI Ming,QI Nan,CHEN Chaobing,et al.Extra-low permeability reservoir heterogeneity and its effect on the distribution of reser⁃ voirs in L2 area of Ganguyi Oilfield[J].Reservoir Evaluation and Development,2019,9(2):1-6.

    • [3] 陈宇航,朱增伍,王喆,等.鄂尔多斯盆地东南部长7油页岩时空分布及控制因素——来自沉积环境和沉积速率的制约[J].石油实验地质,2018,40(2):200-209,217.CHEN Yuhang,ZHU Zengwu,WANG Zhe,et al.Time-space dis⁃ tribution of Chang7 oil shale in southeastern Ordos Basin:Con⁃ trolled by sedimentary environments and deposition rates[J].Pe⁃ troleum Geology & Experiment,2018,40(2):200-209,217.

    • [4] 郁文.考虑平面连通差异的杏六区西部注水细分标准[J].大庆石油地质与开发,2019,38(4):96-104.YU Wen.Subdividing standards of the water injection considering the planar connectivity difference for West Block Xing-6[J].Pe⁃ troleum Geology & Oilfield Development in Daqing,2019,38(4):96-104.

    • [5] 朱丽红,杜庆龙,姜雪岩,等.陆相多层砂岩油藏特高含水期三大矛盾特征及对策[J].石油学报,2015,36(2):210-216.ZHU Lihong,DU Qinglong,JIANG Xueyan,et al.Characteristics and strategies of three major contradictions for continental facies multi-layered sandstone reservoir at ultra-high water cut stage [J].Acta Petrolei Sinica,2015,36(2):210-216.

    • [6] 邹存友,王国辉,窦宏恩,等.油田开发效果评价方法与关键技术[J].石油天然气学报,2014,36(4):125-130,147.ZOU Cunyou,WANG Guohui,DOU Hong’en,et al.Method and key techniques for evaluating the effect of oilfield development [J].Journal of Oil and Gas Technology,2014,36(4):125-130,147.

    • [7] 吕端川,吴伟.水驱开发断块型油藏注采参数调整——以大庆油田北二西F断裂两侧断块为例[J].断块油气田,2017,24(6):813-816.LÜ Duanchuan,WU Wei.Injection-production parameter adjust⁃ ment of water flooding development:taking two fault-blocks on both sides of F Fault of Daqing Oilfield Bei’erxi as examples[J].Fault-Block Oil and Gas Field,2017,24(6):813-816.

    • [8] 严科,张俊,王本哲,等.平面非均质油藏均衡水驱调整方法研究[J].特种油气藏,2015,22(5):86-89.YAN Ke,ZHANG Jun,WANG Benzhe,et al.Balance water flood⁃ ing adjustment of plane heterogeneous reservoirs[J].Special Oil & Gas Reservoirs,2015,22(5):86-89.

    • [9] 赵继勇,樊建明,何永宏,等.超低渗—致密油藏水平井开发注采参数优化实践——以鄂尔多斯盆地长庆油田为例[J].石油勘探与开发,2015,42(1):68-75.ZHAO Jiyong,FAN Jianming,HE Yonghong,et al.Optimization of horizontal well injection-production parameters for ultra-low permeable-tight oil production:A case from Changqing Oilfield,Ordos Basin[J].Petroleum Exploration and Development,2015,42(1):68-75.

    • [10] 王锋.临盘油田复杂断块油藏注采井网优化[J].西安石油大学学报:自然科学版,2014,29(5):58-60,69.WANG Feng.Optimization of reasonable injection-production well pattern in complex fault block of Linpan Oilfield[J].Journal of Xi’an Shiyou University:Natural Science Edition,2014,29(5):58-60,69.

    • [11] 杜殿发,巴忠臣,刘洋,等.稠油油藏活性水驱注采参数优化 [J].特种油气藏,2014,21(1):88-91.DU Dianfa,BA Zhongchen,LIU Yang,et al.Optimization of injec⁃ tion-production parameters for activated waterclooding in heavy oil reservoirs[J].Special Oil & Gas Reservoirs,2014,21(1):88-91.

    • [12] 唐海,何娟,荣元帅,等.塔河断溶体油藏典型断溶体注水驱替规律及剩余油分布特征[J].油气地质与采收率,2018,25(3):95-100.TANG Hai,HE Juan,RONG Yuanshuai,et al.Study on water drive law and characteristics of remaining oil distribution of typi⁃ cal fault-karst in fault-karst reservoirs,Tahe Oilfield[J].Petro⁃ leum Geology and Recovery Efficiency,2018,25(3):95-100.

    • [13] 王瑞,袁士宝,王建,等.复杂断块油藏注采耦合技术提高采收率机理[J].大庆石油地质与开发,2018,37(6):38-42.WANG Rui,YUAN Shibao,WANG Jian,et al.Mechanism of the injection-production coupling technique for complex fault-block oil reservoirs[J].Petroleum Geology & Oilfield Development in Daqing,2018,37(6):38-42.

    • [14] 房娜,张占女,程明佳,等.基于不同裂缝发育程度下周期注水注采参数优化研究[J].特种油气藏,2019,26(1):131-135.FANG Na,ZHANG Zhannü,CHENG Mingjia,et al.Injection-pro⁃ duction parameter optimization study of cyclic water injection at different development levels of fractures[J].Special Oil & Gas Reservoirs,2019,26(1):131-135.

    • [15] 张建国,刘锦华,何磊,等.水驱砂岩气藏型地下储气库长岩心注采实验研究[J].石油钻采工艺,2013,35(6):69-72.ZHANG Jianguo,LIU Jinhua,HE Lei,et al.Long core injectionproduction experiments study on water flooding sandstone gas res⁃ ervoir type underground gas storage[J].Oil Drilling & Production Technology,2013,35(6):69-72.

    • [16] 周涌沂,汪勇,田同辉,等.改善平面非均质油藏水驱效果方法研究[J].西南石油大学学报:自然科学版,2007,29(4):82-85.ZHOU Yongyi,WANG Yong,TIAN Tonghui,et al.Research on the method of improving water-flooding effect for the reservoirs with plane heterogeneity[J].Journal of Southwest Petroleum Uni⁃ versity:Science & Technology Edition,2007,29(4):82-85.

    • [17] 潘伟义,王友启,张诗洋,等.晚期注水开发实验研究[J].油气地质与采收率,2018,25(3):122-126.PAN Weiyi,WANG Youqi,ZHANG Shiyang,et al.An experimen⁃ tal study on late water flooding[J].Petroleum Geology and Recov⁃ ery Efficiency,2018,25(3):122-126.

    • [18] 王东英,任熵,蒋明洁,等.非均质油藏井网注采参数优化的可视化模拟驱替实验[J].油气地质与采收率,2013,20(2):95-98.WANG Dongying,REN Shang,JIANG Mingjie,et al.Visual simu⁃ lation study on optimization of well pattern injection and produc⁃ tion parameters in heterogeneous reservoirs[J].Petroleum Geolo⁃ gy and Recovery Efficiency,2013,20(2):95-98.

    • [19] 陈霆.非均质储层模型微观水驱油实验[J].油气地质与采收率,2007,14(4):72-75.CHEN Ting.Experimental study of water-oil displacement with microscopic models of heterogeneous reservoirs[J].Petroleum Ge⁃ ology and Recovery Efficiency,2007,14(4):72-75.

    • [20] ANIFOWOSE F,LABADIN J,ABDULRAHEEM A.Improving the prediction of petroleum reservoir characterization with a stacked generalization ensemble model of support vector machines[J].Ap⁃ plied Soft Computing,2015,26:483-496.

    • [21] CHITHRA Chakra N,SONG K Y,GUPTA M M,et al.An innova⁃ tive neural forecast of cumulative oil production from a petroleum reservoir employing higher-order neural networks(HONNs)[J].Journal of Petroleum Science and Engineering,2013,106:18-33.

    • [22] 侯春华.基于长短期记忆神经网络的油田新井产油量预测方法[J].油气地质与采收率,2019,26(3):105-110.HOU Chunhua.New well oil production forecast method based on long-term and short-term memory neural network[J].Petroleum Geology and Recovery Efficiency,2019,26(3):105-110.

    • [23] ANIFOWOSE F A,LABADIN J,ABDULRAHEEM A.Ensemble machine learning:An untapped modeling paradigm for petroleum reservoir characterization[J].Journal of Petroleum Science and Engineering,2017,151:480-487.

    • [24] 白玉湖.不确定性致密气产能预测技术[J].中国石油勘探,2018,23(4):107-113.BAI Yuhu.Uncertain productivity prediction technology for tight gas[J].China Petroleum Exploration,2018,23(4):107-113.

    • [25] GRAVES A,LIWICKI M,SANTIAGO Fernández,et al.A novel connectionist system for unconstrained handwriting recognition [J].IEEE Transactions on Pattern Analysis and Machine Intelli⁃ gence,2009,31(5):855-868.

    • [26] FARABET C,COUPRIE C,NAJMAN L,et al.Learning hierarchi⁃ cal features for scene labeling[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,35(8):1 915-1 929.

    • [27] ZEILER M D,KRISHNAN D,TAYLOR G W,et al.Deconvolu⁃ tional networks[C].San Francisco:Computer Society Conference on Computer Vision and Pattern Recognition,2010.

    • [28] ZEILER M D,TAYLOR G W,FERGUS R.Adaptive deconvolu⁃ tional networks for mid and high-level feature learning[C].Barce⁃ lona:International Conference on Computer Vision,2012.

    • [29] STORN R,PRICE K.Differential evolution-A simple and efficient heuristic for global optimization over continuous spaces[J].Jour⁃ nal of Global Optimization,1997,11(4):341-359.

    • [30] 汪慎文,丁立新,张文生,等.差分进化算法研究进展[J].武汉大学学报:理学版,2014,60(4):283-292.WANG Shenwen,DING Lixin,ZHANG Wensheng,et al.Survey of differential evolution[J].Journal of Wuhan University:Natural Science Edition,2014,60(4):283-292.