Wellbore Planning via Overlapping Local Subsurface Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting subsurface properties in geospatial data often result in biased parameter estimates due to spatial autocorrelation, leading to inaccurate predictions when grouping is done at large geographic levels, and are limited by the heterogeneity and incompleteness of datasets from previously drilled wells.
Innovation Solution
The method involves creating a global model by combining overlapping local models from smaller, heterogeneous geospatial datasets, using machine learning and statistical models to interpolate values and limit local biases, and incorporating depth information to create depth layers for more accurate subsurface property predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If grouping is done at large geographic level, then the model covers broader area, but the prediction accuracy decreases due to spatial autocorrelation bias
Solution Approach 1:
The patent divides the large geographic area into multiple smaller modelling areas (local regions) and creates separate local models for each area. This segmentation allows the system to capture local spatial autocorrelation patterns while maintaining broad coverage through the aggregation of multiple local models into a global model.
Solution Approach 2:
The patent applies different modelling approaches and parameters to different local areas based on their specific characteristics. Each local model is tailored to the specific geospatial data patterns of its region, allowing for more accurate local predictions that respect local spatial autocorrelation effects.
2Measurement precision
If local models are created for each modelling area, then the prediction accuracy improves, but the model complexity increases
Solution Approach 1:
The patent combines multiple local models into a single global model that integrates the strengths of each local model. This merging process reduces overall complexity by creating a unified modelling framework while preserving the accuracy benefits of local specialised models through the use of overlapping modelling areas and data sharing.
3Reliability
If overlapping modelling areas are used, then the spatial autocorrelation bias is reduced, but the computational resources required increase
Solution Approach 1:
The patent uses partial overlap between modelling areas rather than complete coverage by each area. This approach reduces computational redundancy while maintaining the benefits of spatial autocorrelation reduction. The overlapping regions are sufficient to provide reliable parameter estimates without requiring excessive computational resources for complete redundancy.
Data Source
AI summary
A method of predicting a subsurface property value includes obtaining a plurality of geospatial datapoints; selecting a plurality of modelling areas of the geospatial datapoint, wherein each modelling area overlaps at least a portion of a neighboring modelling area; determining a local model for each modelling area; assigning a local model value for the subsurface property to the local cells based on values of the plurality of geospatial datapoints within the modelling area; determining a global model for the plurality of geospatial datapoints; assigning at least one global model value for the subsurface property to each global cell of the global model; predicting a calculated value of the subsurface property at a selected location based at least partially on the global model value of a selected global cell of the global model; and planning a subsurface well based at least partially on the calculated subsurface property.