Automated Resistivity Inversion and Formation Geometry Determination in High-Angle and Horizontal Wells Using Deep Learning Techniques
- Hu Li (Maxwell Dynamics, Inc) | Gang Liu (CNPC Logging, LWD Center) | Shansen Yang (CNPC Logging, LWD Center) | Ying Guo (CNPC Logging, LWD Center) | He Huang (CNPC Logging, LWD Center) | Mingzong Dai (CNPC Logging, LWD Center) | Yuanshi Tian (CNPC Logging, LWD Center)
- Document ID
- Society of Petrophysicists and Well-Log Analysts
- SPWLA 60th Annual Logging Symposium, 15-19 June, The Woodlands, Texas, USA
- Publication Date
- Document Type
- Conference Paper
- 2019. held jointly by the Society of Petrophysicists and Well Log Analysts (SPWLA) and the submitting authors
- 2 in the last 30 days
- 198 since 2007
- Show more detail
In High-Angle and Horizontal (HAHZ) wells, resistivity measurements are often complicated by environmental effects, such as approaching boundary, dip angle, and resistivity anisotropy. Abnormal responses, e.g., polarization horns, are frequently observed when the tool approaches or penetrates the formation boundary. Thus, before any quantitative petrophysical interpretation, one has to determine the geometric structure and remove the environmental effects. However, this can sometimes be a challenging task. On the one hand, it is questionable or incorrect to apply environmental correction algorithms to the measurements without identifying the geometric structure. On the other hand, it is not efficient for experts to determine these effects manually; besides, subjectivities may exist because of human intervention.
In this paper, we present a Deep Learning (DL), i.e., a deep Convolutional Neural Networks (CNNs) learning framework to automatically identify the controlling pattern, i.e., the geometric structure and the decisive environmental factor. The inputs are the raw logs from logging-while-drilling (LWD) measurements. The outputs are the locations of the boundaries and the controlling patterns, which are divided into two major categories: Boundary-dominated patterns and Environmental-effects-dominated patterns. Furthermore, the properties of the boundary or even the relationship between boundaries can be determined. Consequently, the decisive environmental factor can be identified.
The training phase takes well-defined features as class labels and numerical simulated LWD logs as the training set for machine learning. Then, the CNNs is performed to extract representations of the referenced features from the data for pattern identification and data interpretation. Meanwhile, a set of inversion-based correction algorithms are also developed to derive the formation “true” resistivity based on the geometric structure identified by the Deep CNN.
The proposed scheme has been tested on multiple data sets that were simulated or recorded in HAHZ wells. The deep CNNs gives a reliable identification results, based on which automated resistivity inversion can be implemented. Using the deep CNNs framework, feature extraction for geologic structure detection can be done via the DL method. Furthermore, the data set, which is recognized as a specific pattern by the user, can be added into the database for further training purposes.
|File Size||1 MB||Number of Pages||11|