人工神经网络模型在测井中实例.docx

1、人工神经网络模型在测井中实例八 应用实例分析81 有监督学习的前馈神经网络反向传播算法BP应用实例分析811 实例1对陆9井区块储层油气水层的识别在利用神经网络识别储层的油气水层和预测新井的油气水层的特征时，是以实际的测井资料和试油结论为依据来建立样本集。根据陆9井区块10口井的测井资料和试油结果，建立样本集34组如下表8-1：表8-1 陆9井区块测井与试油结果的油气水层样本数据井号井 段厚度RTRXOSPGRACCNLDEN试油结论期望值(m)(m).m.m(mv)(API)s/ft(%)g/cm3陆91031.3 - 1034.02.77.98.6-56.958.9107.9272.11气

2、层0 0陆91037.0 - 1042.05.14.76.1-57.660104.7352.15水层0 1陆91122.5 - 1127.04.65.47-53.766105.832.92.17水层0 1陆91186.0 - 1192.06.19.29.4-54.461.6104.231.82.16油水同层1 0陆91233.0 - 1237.04.16.78.3-59.164105.934.32.13油水同层1 0陆91295.0 - 1299.14.14.77.2-58.464.9102.132.42.16水层0 1陆91323.1 - 1328.057.49-5867.4103.733.

3、12.14油层1 1陆91415.0 - 1418.03.111.48.3-5863.999.933.12.15油层1 1陆91424.0 - 1426.12.15.28.3-58.267.4104.733.82.14油层1 1陆91434.6 - 1437.32.84.77.5-60.167.410232.32.14水层0 1陆1011180.9 - 1187.76.865.2-20.262106.332.22.12油水同层1 0陆1011273.0 - 1275.02.16.910.8-22.66594.930.42.22油水同层1 0陆1011409.1 - 1413.84.87.15.

4、9-24.5689830.92.15油层1 1陆1021164.0 - 1168.04.16.94.9-5.264.2106.3332.12油水同层1 0陆1021216.0 - 1219.03.15.24.6-5.464.6101.834.12.16油水同层1 0陆1021400.1 - 1403.33.35.15.4-7.869.3104.932.72.14油水同层1 0陆1031267.0 - 1270.03.18.711.2-11.265.4103.833.92.13油水同层1 0陆1031352.0 - 1356.04.113.415.1-1670.596.930.92.2油水同层1

5、 0陆1041002.0 - 1010.085.113.5-20.965.8104.533.42.16水层0 1陆1041192.2 - 1194.01.95.15.7-19.967.8101.633.62.16水层0 1陆1041202.0 - 1208.66.65.56-17.982.19832.12.3油水同层1 0陆1041214.0 - 1218.954.55-20.360.9103.533.72.13水层0 1陆1081030.0 - 1036.06.15.26.7-49.664106.835.32.15水层0 1陆1081122.0 - 1126.14.15.47-49.869.

6、210634.42.16水层0 1陆1081215.0 - 1217.02.15.97.9-52.477.2102.534.82.17水层0 1陆1091346.7 - 1350.03.47.98.6-25.66597.531.92.18油水同层1 0陆1111166.8 - 1172.05.211.213.3-74.461.1111.231.372.14油层1 1陆1111197.0 - 1202.05.17.99.7-72.464.1119.630.692.12油水同层1 0陆1111276.2 - 1278.01.96.89.3-71.164.211131.122.12油水同层1 0陆1

7、121133.0 - 1136.13.166.2-14.877.4101.532.332.2水层0 1陆1121225.9 - 1228.02.15.46.2-18.769.5103.533.522.14水层0 1陆1131264.5 - 1266.01.65.86.8-39.574.199.231.42.23水层0 1陆1131334.0 - 1335.01.16.87.2-43.167.2101.933.42.19油层1 1陆1131338.0 - 1340.02.15.87.3-48.264.697.930.62.19水层0 1期望值中00为气层，01为水层，10为油水同层，11为油层。

8、输入计算矢量为 RT、RXO、SP、GR、AC、CNL、DEN ，网络拓扑结构为：7772，即输入节点为7个，第一隐层为7个节点，第二隐层为7个节点，输出节点为2个，用来识别气层、油层、油水同层、水层。为了消除测井数据不同量纲在数值上的差异，需要对样本集数据进行归一化处理，经归一化处理后，各物理量的测井数据既不失去自身的性质，各测井数值间又具有统一的度量尺度，量纲的影响自然就消除了，从而为神经网络提供可靠的基础数据。为此可以利用本论文开发的人工神经网络测井解释软件系统的预处理功能模块的归一化处理工具完成。归一化处理结果如下表8-2：表8-2 陆9井区块测井与试油结果的归一化训练数据输入参数及网

9、络拓扑结构如下：Temperature = 1.000000 ；神经元温度ETA = 0.250000 ；学习速率ALPHA = 0.020000 ；动量常数MAXITER = 940000 ；最大迭代次数ERRTOL = 0.020000 ；最坏误差的最小容许误差Number of layers=4 ；网络层数Number of neurons in layer 0 = 7 ；输入层节点数Number of neurons in layer 1 = 7 ；第一隐层节点数Number of neurons in layer 2 = 7 ；第二隐层节点数Number of neurons in

10、layer 3 = 2 ；输出层节点数经过网络训练收敛后，神经元之间的权值和各神经元阈值为：8.727776 -4.440123 9.305367 -3.788903 5.700017 -6.647582 7.313374-6.078639 5.644599 3.452872 -0.972144 5.722620 -2.032519 -2.189986-3.890867 -2.025399 3.054030 1.962049 0.223342 2.564262 -1.235002-0.568113 -2.290304 -1.470534 7.280448 -2.077451 0.886878

11、0.3015156.177826 -1.106736 -6.932556 -2.540787 15.417054 6.679054 -7.197594-1.818464 -1.959654 1.100705 -3.614675 -0.439954 6.579241 -0.8249681.722311 0.746903 -2.035187 0.609210 -2.013068 -2.475797 3.4140292.391842 -0.469176 3.742926 0.353154 -1.548412 -2.616227 -1.446883-1.785754 -0.407602 -0.0044

12、58 -1.139143 -1.922753 1.070846 2.4699861.421194 -1.642308 -0.444225 -1.181446 -0.373724 -2.662882 -2.273508-0.998508 -0.298864 -7.840540 2.662292 1.202784 -3.962256 -12.430850-1.118792 1.127342 5.060939 -4.330868 -2.209113 -2.214474 0.202059-3.956062 -2.255609 -1.939415 -1.247700 -10.597683 6.65665

13、2 3.222105-6.296951 -8.128192 2.523240 2.860141 3.567064 -4.093600 -0.893888-0.220077 -2.210967 -3.004974 -0.991026 -0.677851 -0.837868 -2.634454-0.856080 -1.211024 -1.042876 -2.741504 -1.728751 -1.039846 -1.5389063.699656 1.792121 -1.801712 -0.192390 -0.757207 -0.161089 1.074828-0.966101 -0.137877

14、0.918620 -0.348751 1.719734 1.432033 -1.6595210.097559 0.038501BP算法收敛曲线为：经过9700次的训练后，网络收敛，平均误差为0.000061，最差误差为 0.002072，这个误差范围完全可以满足解释工作的实际需要。学习样本集的归一化数据测试结果如下表8-3所示：表8-3 陆9井区块测井检验样本集的归一化测试结果原始样本数据及测试结果见表8-4，从表8-4可以看出，测试结果与期望值完全吻合，测试样本的回判准确率为100%。表8-4 陆9井区块油气水层识别的样本数据与神经网络测试结果井号 井 段厚度RTRXOSPGRACCNLDE

15、N试油结论期望值网络测试结果(m)(m)(.m)(.m)(mv)(API)(s/ft)(%)(g/cm3)陆91031.3 - 1034.02.77.98.6-56.958.9107.9272.11气层000.0048520.003651陆91037.0 - 1042.05.14.76.1-57.660104.7352.15水层01-0.0010421.009907陆91122.5 - 1127.04.65.47-53.766105.832.92.17水层010.013640.999245陆91186.0 - 1192.06.19.29.4-54.461.6104.231.82.16油水同层1

16、01.0016530.007476陆91233.0 - 1237.04.16.78.3-59.164105.934.32.13油水同层101.0062610.017213陆91295.0 - 1299.14.14.77.2-58.464.9102.132.42.16水层01-0.0007991.005666陆91323.1 - 1328.057.49-5867.4103.733.12.14油层111.0036680.997893陆91415.0 - 1418.03.111.48.3-5863.999.933.12.15油层111.003681.003336陆91424.0 - 1426.12.

17、15.28.3-58.267.4104.733.82.14油层111.0017421.000637陆91434.6 - 1437.32.84.77.5-60.167.410232.32.14水层01-0.0125191.010433陆1011180.9 - 1187.76.865.2-20.262106.332.22.12油水同层101.0028030.003288陆1011273.0 - 1275.02.16.910.8-22.66594.930.42.22油水同层101.0023650.010802陆1011409.1 - 1413.84.87.15.9-24.5689830.92.15油

18、层110.9992330.999196陆1021164.0 - 1168.04.16.94.9-5.264.2106.3332.12油水同层100.852671-0.016625陆1021216.0 - 1219.03.15.24.6-5.464.6101.834.12.16油水同层101.003622-0.015764陆1021400.1 - 1403.33.35.15.4-7.869.3104.932.72.14油水同层101.0058780.014551陆1031267.0 - 1270.03.18.711.2-11.265.4103.833.92.13油水同层101.001565-0.

19、000174陆1031352.0 - 1356.04.113.415.1-1670.596.930.92.2油水同层101.0000970.003677陆1041002.0 - 1010.085.113.5-20.965.8104.533.42.16水层010.0024390.997946陆1041192.2 - 1194.01.95.15.7-19.967.8101.633.62.16水层010.0103861.007813陆1041202.0 - 1208.66.65.56-17.982.19832.12.3油水同层101.001080.005502陆1041214.0 - 1218.95

20、4.55-20.360.9103.533.72.13水层010.0020721.00359陆1081030.0 - 1036.06.15.26.7-49.664106.835.32.15水层010.0021670.989958陆1081122.0 - 1126.14.15.47-49.869.210634.42.16水层010.0117671.007604陆1081215.0 - 1217.02.15.97.9-52.477.2102.534.82.17水层01-0.0009321.012274陆1091346.7 - 1350.03.47.98.6-25.66597.531.92.18油水同

21、层101.0033080.000074陆1111166.8 - 1172.05.211.213.3-74.461.1111.231.3712.14油层111.0008951.004737陆1111197.0 - 1202.05.17.99.7-72.464.1119.630.6932.12油水同层101.00143-0.017952陆1111276.2 - 1278.01.96.89.3-71.164.211131.1232.12油水同层101.0078590.00695陆1121133.0 - 1136.13.166.2-14.877.4101.532.3282.2水层010.0004421

22、.002218陆1121225.9 - 1228.02.15.46.2-18.769.5103.533.5232.14水层01-0.0091621.008353陆1131264.5 - 1266.01.65.86.8-39.574.199.231.42.23水层01-0.0003181.003866陆1131334.0 - 1335.01.16.87.2-43.167.2101.933.42.19油层111.000641.007691陆1131338.0 - 1340.02.15.87.3-48.264.697.930.62.19水层010.007860.997355812 实例2 某油田储集

23、层的油气评价在利用神经网络识别储层的油气水层和预测新井的油气水层的特征时，是以实际的测井资料和试油结论为依据来建立样本集。根据该油田试油或生产证实的油层有23层，油水同层有7层，水层17层，将岩性系数(LTH)、孔隙度(POR)、侵入系数(Di)、含油饱和度(So)作为油气评价参数，输入矢量为 LTH，POR，Di，So，以其中30层作为已知层，供学习之用(表8-5)，其余17层作检验之用(表8-6)：表8-5 学习样本原始数据表表8-6 检验测试样本原始数据表表8-5中期望值1 0为油层，0 1为油水同层，0 0为水层。网络拓扑结构为：4442，即输入节点为4个，第一隐层为4个节点，第二隐层

24、为4个节点，输出节点为2个，用来识别油层、油水同层、水层。为了消除测井数据不同量纲在数值上的差异，需要对样本集数据进行归一化处理，经归一化处理后，各物理量的测井数据既不失去自身的性质，各测井数值间又具有统一的度量尺度，量纲的影响自然就消除了，从而为神经网络提供可靠的基础数据。为此可以利用本论文开发的人工神经网络测井解释软件系统的预处理功能模块的归一化处理工具完成。训练样本归一化处理结果如下表8-7：表8-7 该油田训练样本归一化处理结果输入参数及网络拓扑结构如下：Temperature = 1.000000 ；神经元温度ETA = 0.300000 ；学习速率ALPHA = 0.050000

25、；动量常数MAXITER = 10000000 ；最大迭代次数ERRTOL = 0.001000 ；最坏误差的最小容许误差Number of layers=4 ；网络层数 Number of neurons in layer 0 = 4 ；输入层节点数Number of neurons in layer 1 = 4 ；第一隐层节点数Number of neurons in layer 2 = 4 ；第二隐层节点数Number of neurons in layer 3 = 2 ；输出层节点数经过网络训练收敛后，神经元之间的权值和各神经元阈值为：-10.018420 -0.028980 13.463097 -2.739568 2.107397 -13.526530 1.53651524.335915 -15.529228 9.512752 -3.739093 6.385159 10.988239 -9.099783-1.946299 25.293393 -6.355862 -21.526247 -16.627912 -2.731663 -1.461133-9.275027 -4