Fluid Optical Database Reconstruction With Validated Mapping from External Oil and Gas Information Source
- Dingding Chen (Halliburton Technology) | Christopher Jones (Halliburton Technology) | Bin Dai (Halliburton Technology) | Anthony van Zuilekom (Halliburton Technology)
- Document ID
- Society of Petrophysicists and Well-Log Analysts
- Publication Date
- December 2018
- Document Type
- Journal Paper
- 849 - 862
- 2018. Society of Petrophysicists & Well Log Analysts
- 0 in the last 30 days
- 64 since 2007
- Show more detail
Real-time downhole fluid analysis requires advanced modeling of downhole optical spectroscopic measurements of fluid to determine its compositions and properties. This type of modeling aims to build a multivariate correlation between optical-sensor measurements and fluid compositions or physical properties using statistical and machine-learning techniques. To obtain robust and accurate models, a representative and consistent database including both fluid optical measurements and compositions/properties information is needed. Conventionally, constructing such a database requires a significantly large number of reservoir fluid samples from different regions to be collected and analyzed in the in-house laboratory. Due to the high cost and long duration of sample procurement and data acquisition, building a comprehensive in-house fluid optical database can be a time-consuming and expensive endeavor.
The limitations of in-house database development can be overcome by augmenting them with external petroleum fluid-composition and property databases that are publicly available but often missing fluid optical measurement data. This paper presents a novel mutually complementing method that enables reconstruction of missing optical-measurement data in the external public databases. In this method, forward and inverse neural networks (NNs) are used to build multivariate correlation models between optical measurements and fluid-composition/property data in the in-house database. The NNs can then be applied to an external petroleum fluid database by alternatively performing forward and inverse computation to evaluate the data consistency and to reconstruct the missing fluid optical measurement data in the external database, hence augmenting the in-house database.
A case study is presented to exemplify the application of the proposed method for fluid optical-database reconstruction. Investigation includes model selection from all candidate NNs through evolutionary optimization to maximize the compatibility of real and reconstructed data. Compared to the original in-house database, the augmented database doubles the number of fluids, enhances sample diversity and distribution, and leverages the capability for utilization. After data integration, calibration models are refined based the reconstructed database, and there are measurable improvements to the predictive performance.
|File Size||10 MB||Number of Pages||14|