Neural Network VSP Data Correction for Corrupt Seismic Sections

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Vertical seismic profiling (VSP) data often contains noisy, corrupt, anomalous, and/or missing sections due to poor cementation between casing and the well wall, limited access of seismic receivers, and VSP acquisition system failures, which render these data unsuitable for converting surface seismic (SS) data from a time domain to a depth domain for accurate subterranean feature characterization.

Innovation Solution

A method and system utilizing a neural network to correct VSP data by determining attributes and spectra from valid sections, training the network with VSP and SS data, and predicting corrected VSP spectra for corrupt sections using the trained neural network.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If VSP data is collected using traditional methods, then data acquisition is straightforward, but the data contains noisy, corrupt, anomalous, and missing sections due to poor cementation, limited receiver access, and system failures

Engineering Contradiction:

Improvedata qualityVSAvoidcorrupt data sections

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The patent uses surface seismic (SS) data as an intermediary to correct VSP data. The SS data, which covers the same depth range and geological formations, serves as a reference to identify and correct corrupt sections in VSP data through attribute comparison and spectral analysis

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent creates a corrected version of the corrupt VSP data by copying information from the valid VSP sections and SS data. The correction process involves copying spectral characteristics and attributes from reliable sources to reconstruct the missing or corrupt information

Inventive Principle:

Principle #26Copying

2Measurement precision

If neural network training is performed using only valid VSP sections, then the model can learn from clean data, but the training dataset size is reduced

Engineering Contradiction:

Improveattribute determination accuracyVSAvoidtraining data volume

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent transforms the training approach by changing from using raw VSP data directly to using derived attributes (spectral ratios, coherence, frequency content) as training inputs. This parameter transformation allows the neural network to learn from more subtle patterns while maintaining data quality

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12455392B2Method to correct VSP data

Publication Date: 2025.10.28 SAUDI ARABIAN OIL CO

AI summary

Systems and methods are disclosed. The method includes obtaining vertical seismic profiling (VSP) data and surface seismic (SS) data for a subterranean region of interest. The VSP data includes a corrupt section and a valid section. The method further includes determining a VSP attribute and a VSP spectrum using the VSP data, determining an SS attribute using the SS data, and determining a corrected VSP attribute for the corrupt section. The method still further includes training a neural network using the VSP attribute, the SS attribute, and the VSP spectrum for the valid section, predicting a corrected VSP spectrum for the corrupt section by inputting the corrected VSP attribute and the SS attribute for the corrupt section into the trained neural network, and determining corrected VSP data for the corrupt section using the corrected VSP attribute and the corrected VSP spectrum.