西南石油大学学报(自然科学版) ›› 2017, Vol. 39 ›› Issue (5): 120-128.DOI: 10.11885/j.issn.16745086.2016.01.19.02

• 石油与天然气工程 • 上一篇    下一篇

粗糙集和神经网络相融合的钻井安全评价模型

李建, 李珂, 王兵   

  1. 西南石油大学计算机科学学院, 四川 成都 610500
  • 收稿日期:2016-01-19 出版日期:2017-10-01 发布日期:2017-10-01
  • 作者简介:李建,1960年生,男,汉族,四川绵阳人,教授,主要从事石油工程计算、数据仓库与数据挖掘等方面的研究。E-mail:lijian2835@sina.com;李珂,1990年生,男,汉族,河南南阳人,硕士研究生,主要从事计算机工程与应用、数据挖掘等方面的研究。E-maill:like4070@163.com;王兵,1977年生,男,汉族,四川南充人,副教授,主要从事钻井安全评价、数据分析与数据挖掘等方面的研究。E-mail:w9521423@sina.com
  • 基金资助:
    国家科技重大专项(2011ZX05021-006);四川省安全生产监督局项目(sichuan-0009-2016AQ)

An Evaluation Model of Drilling Safety Based on Combined Rough Set and Neural Network

LI Jian, LI Ke, WANG Bing   

  1. School of Computer Science, Southwest Petroleum University, Chengdu, Sichuan 610500, China
  • Received:2016-01-19 Online:2017-10-01 Published:2017-10-01
  • Contact: 李建,E-mail:lijian2835@sina.com

摘要: 针对动态性、随机性和不确定性较强的钻井作业现场,开展了安全评价的研究。提出了一种基于粗糙集和BP神经网络对作业现场进行安全评价的方法。首先以粗糙集为基础来构建BP神经网络的前置系统,对采集到的样本数据进行属性约简。其次,根据约简结果以及作业当天的事故情况完成了BP神经网络输入层和输出层的设计,并根据输入层和输出层神经元的个数通过试凑法确定网络隐含层的神经元数量范围,并采用训练样本对不同神经元个数所对应的网络模型进行训练,选择网络误差最低的网络作为所构建的网络模型。最后,选取16 d的测试样本对网络进行验证,将网络的输出同作业现场的实际结果进行比较,有14 d的网络结果与实际结果相符,测试准确率达到了87.5%。

关键词: 安全评价, 粗糙集, BP神经网络, 属性约简, 钻井作业

Abstract: A safety evaluation study was carried out for a drilling site with strong dynamics, randomness, and uncertainty. A safety evaluation method based on a rough set and a BP neural network is proposed for an operational field. First, the pre-system of the BP neural network is constructed based on the rough set, and a simplification of the attributes of the collected sample data is performed. Second, the input and output layers of the BP neural network are constructed based on the simplification results and the accident scenario on that particular operational day. Furthermore, the number of neurons in the hidden layer of the network is determined through a trial and error method based on the number of neurons in the input and output layers. The training samples are used to train the network models with different number of neurons. The network with the lowest error is selected as the constructed network model. Finally, test samples for 16 days are selected to verify the network. The network results are consistent with the actual results for 14 days, indicating that the test accuracy is 87.5%.

Key words: safety evaluation, rough set, BP neural network, attribute simplification, drilling operation

中图分类号: