科研之友
微信
新浪微博
Facebook
分享链接
摘要
In the traditional data-driven data mining process, there are huge gaps between the efficient algorithms and intelligent tools as well as the invalidity of knowledge which is obtained by traditional data-driven data mining. Meanwhile, each data in the earth science field contains a solid physical meaning. If there is no corresponding domain knowledge involved in the mining process, the information explored by data-driven data mining will be lack of practicability and not able to effectively solve problems in the earth science area. Therefore, the task-driven data mining is proposed. Additionally, task-driven data mining concepts and principles are elaborated with the help of data mining concepts and techniques. It is divided into seven elements such as data warehousing, data preprocessing, feature subset selection, modeling, model evaluation, model updating and model release. Those constitute a cyclic and iterative process until a predictive model which is capable of effectively achieving the objectives. In order to accurately identify complex lithologies, this paper puts forward a self-organizing feature map neural network based on the task-driven data mining. With the attempt to solve the problem of complex reservoir identification, the decision tree and support vector machine are used to build the fluid predictive model. Meanwhile, the optimization algorithms inclusive of genetic, grid and quadratic are adopted to optimize the important parameters of C-SVC and upsilon-SVC, such as C, upsilon and gamma, so as to improve the classification performance and generalization ability of the predictive model of support vector machine. The conclusions of fine interpretation are compared with the core analysis data and well testing data. As a result, the accuracy of the complex lithology and reservoir identification is more than 90%. Finally, the paper puts forward the understandings, development prospects and key challenges of task-driven data mining facing.
- 出版日期2013-9
- 单位中国石油大学（北京）
