基于神经网络的杆泵抽油系统故障诊断.doc

基于神经网络的油故障诊断问题描述 抽油法当前国内外应用最广泛的传统的机械采油技术，目前世界上%到85%的采油井都采用此种采油方式。的钢抽油杆工作时在油管中上下反复运动，在这种交变载荷的作用下经常会发生断裂，接外蜡严重，泵效低，能耗高。油井供液变差，泵挂深度逐年加深，井下管柱更为复杂，管杆，产出液腐蚀性强，矿化度高，化学腐蚀及电化学腐蚀同时存在。恶劣的外界环境抽油系统及其容易出现故障，每年只是有杆泵抽油井故障诊断检修就得耗费数以亿计的资金。抽油系统 2、神经网络的输入与目标的设计 向量的确定实际上就是特征向量的提取。杆泵抽油系统来说：磨损）、化学腐蚀）、组合不合理）、最大拉应力破坏）、杆偏磨）、油采出高度）、油含水量）、井液不足）结蜡）、供液）等特征向量最能体现杆泵油系统的整体运行情况。因此以上的特征参数该神经元网络的输入向量个数为该神经元网络的输入一个向量，对数据进行归一化处理。表已经输入向量数据。数据见附录）样本数据 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 类型 1 79.5615 66.3013 57.2843 57.8147 56.2235 50.3890 59.9363 53.5714 50.3890 50.3890 无 故 障 2 71.6054 55.1627 47.7369 53.0410 54.6322 46.1457 76.9095 54.6322 47.2065 45.6153 3 76.7514 64.7100 55.1627 56.7539 53.0410 46.1457 56.7539 49.3282 46.6761 47.2065 4 80.0919 63.1188 54.6322 56.2235 54.6322 46.1457 62.5884 49.3282 46.6761 46.6761 5 82.2136 71.0750 55.1627 53.5714 53.5714 47.7369 58.3451 56.7539 47.7369 46.8963 6 86.4569 68.4229 56.2235 56.2235 54.6322 49.8586 63.6492 51.4498 48.2673 48.2673 7 73.1966 60.4668 50.9194 51.9802 52.5106 46.1457 54.6322 51.9802 45.6153 45.0849 8 84.8656 65.2405 54.6322 55.6931 55.1627 45.6153 55.1627 57.2843 45.0849 44.0240 9 74.7878 59.9363 48.7977 51.9802 55.1627 48.2673 78.5007 59.4059 48.7977 45.6153 10 75.8487 72.6662 53.0410 48.2673 46.1457 40.3112 45.0849 41.3720 63.1188 42.4328 11 72.1358 59.4059 48.2673 47.7369 44.0240 39.7808 43.4936 44.5545 39.7808 40.3112 12 68.9533 59.9363 46.1457 44.0240 46.6761 43.4936 41.3720 41.9024 44.5545 39.7808 13 89.1089 71.6054 59.9363 56.7539 51.9802 55.6931 76.9095 58.3451 50.3890 50.3890 14 91.7610 67.3621 58.8755 56.7539 51.4498 57.8147 61.5276 51.9802 50.3890 50.9194 1 49.8586 50.9194 47.7369 47.2065 57.8147 62.5884 57.8147 50.3890 46.6761 48.2673 双 凡 尔 漏 2 48.2673 48.2673 52.5106 51.4498 45.6153 52.5106 59.4059 58.8755 56.2235 49.8586 3 48.7977 47.2065 53.5714 54.6322 45.0849 48.7977 56.2235 57.8147 57.8147 52.5106 4 50.9194 48.7977 53.5714 54.1018 48.7977 56.2235 60.4668 62.0580 62.0580 56.2235 5 50.3890 49.8586 56.