Task-driven data mining in the formation evaluation field

摘要: In the traditional data-driven data mining process, there are huge gaps between efficient algorithms and intelligent tools as well invalidity of knowledge, which is obtained by mining. Meanwhile, each in earth science field contains a solid physical meaning. If no corresponding domain knowledge involved information explored will be lack practicability not able to effectively solve problems area. Therefore, task-driven proposed. Additionally, concepts principles elaborated with help techniques. It divided into seven elements such warehousing, preprocessing, feature subset selecting, modeling, model evaluating, updating releasing. Those constitute cyclic iterative process until appearance predictive model, capable achieving objectives. The applied recognizing complex lithologies low resistivity oil layer, whole elaborated. Their accuracy rates more than 90%. Finally, paper puts forward understandings, development prospects key challenges facing.