Skip to main content
— Presentation at CIKM2016 —
March 24, 2017
The 25th ACM International Conference on Information and Knowledge Management (CIKM) was held in Indianapolis, United States, during October 24-28, 2016. CIKM is a premier interdisciplinary conference bringing together researchers and practitioners from the knowledge management, information retrieval and database communities. The research track of CIKM 2016 received 935 full paper submissions out of which 165 were accepted, with an acceptance rate of 17.6%. The industry track of CIKM 2016 received 111 full paper submissions out of which 22 were accepted, with an acceptance rate of 19.8%.
In CIKM2016, Research & Development Group, Hitachi, Ltd. made a presentation titled "Deep Match between Geology Reports and Well Logs Using Spatial Information". This paper was accepted as a full paper in the Industry track. This work is one of core techniques in Hitachi's oil and gas solution that aims at reducing costs of oil operators in a data-driven perspective.
Fig. 1 Framework of non-linear model
In the shale oil and gas industry, operators are looking toward big data and new analytics tools and techniques to optimize operations and reduce cost. Formation evaluation is an essential step in evaluating the subsurface by using the information commonly gathered during the drilling process. An appropriate formation evaluation will assist operators in making optimal operations that dramatically reduce the cost of operations. To assist engineers in understanding the subsurface and in turn make optimal operations, this work focused on learning semantic relations between geology reports and well logs, which are collected during down-hole drilling. By using this technique, it allows engineers to interpret well logs more rapidly, and assist engineers in issuing geology reports.
The challenges of this work are how to represent the features of the geology reports and the well logs collected at measured depths and how to effectively embed them into a common feature space. The feature representations for both geology reports and well logs, which capture the characteristics of depth-series data, are extracted. Both linear and nonlinear (artificial neural network) models were proposed to achieve such an embedding. The framework of a model using an artificial neural network is illustrated in Figure 1.
Extensive validations were conducted on public well data of North Dakota in the United States. It is discovered that the feature representation of both geology reports and well logs are correlated with the geological distance. It is shown that this spatial information is highly effective in both the linear and nonlinear models. The proposed nonlinear model with the spatial information performs the best among the state-of-the-art methods. This is the first work that formally defines the problem of learning semantic relations among heterogeneous data in the oil and gas industry.
(By TONG Bin, IWAYAMA Makoto)