Research Article
A Deep Neural Network Based Feature Learning Method for Well Log Interpretation
@INPROCEEDINGS{10.1007/978-3-030-67514-1_43, author={Liyuan Bao and Xianjun Cao and Changjiang Yu and Guanwen Zhang and Wei Zhou}, title={A Deep Neural Network Based Feature Learning Method for Well Log Interpretation}, proceedings={IoT as a Service. 6th EAI International Conference, IoTaaS 2020, Xi’an, China, November 19--20, 2020, Proceedings}, proceedings_a={IOTAAS}, year={2021}, month={1}, keywords={Well logging interpretation Deep neural network Feature learning Autoencoder}, doi={10.1007/978-3-030-67514-1_43} }- Liyuan Bao
Xianjun Cao
Changjiang Yu
Guanwen Zhang
Wei Zhou
Year: 2021
A Deep Neural Network Based Feature Learning Method for Well Log Interpretation
IOTAAS
Springer
DOI: 10.1007/978-3-030-67514-1_43
Abstract
Well log interpretation is an important task in the process of petroleum logging. It is able to help the researchers to determine the residual oil volume and to improve the petroleum productivity efficiency. Well log interpretation requires the synthesis of a large amount of data, and it is difficult to manually browse the data from a global perspective. It is urgent to introduce big data analysis methods to deal with the complex oil well logs data. The accuracy of logging interpretation greatly depends on the logging features selection and representation. However, the conventional methods using expert experiences easily lead to feature incomplete problem and affects the interpretation results. In this paper, we propose a deep neural network based feature learning method for well log interpretation. Firstly, we select original features of the well log data according to the physical characteristics of well logging sensors. And then, we formulate a deep neural network based autoencoder model to explore the intrinsic representation of original features. At last, we utilize linear SVM classifier on well log interpretation problem to evaluate the proposed feature learning method. The experimental results demonstrate that the classification accuracy by using learned feature representation increase to(99.8\%)compared with that of(74.6\%)by using original feature representation.
