Velocity Model Building From Raw Shot Gathers Using Machine Learning Method, & More

Introduction

Introduction

Seismic information translation is a foundation of subsurface investigation, significant in fields like oil and gas extraction, ecological examinations, and geotechnical designing. Making precise speed models, which depict seismic wave speeds inside the World’s subsurface, is fundamental for deciphering seismic information really. Generally, this interaction included relentless manual translation and broad calculations. Be that as it may, propels in AI (ML) are changing speed model structure, making it more productive, precise, and versatile. This article digs into the most common way of building speed models from crude shot accumulates utilizing AI strategies, featuring the advantages and difficulties related with these advanced techniques.

Seismic imaging requires a speed model to make an interpretation of seismic signs into subsurface pictures. Tomography and FWI are traditional strategies used to build these models. Tomography includes iterative updates to the speed model, frequently requiring human contribution for each update. FWI, while exact, is computationally concentrated and furthermore requests impressive human exertion for readiness.

AI has shown huge outcome in different fields like PC vision, discourse acknowledgment, and bioinformatics. Motivated by these triumphs, there have been endeavors to apply AI to seismic imaging (Yang and Mama, 2019; Zheng et al., 2019). While some headway has been made, especially in the immediate utilization of AI to seismic shot assembles, current models don’t yet match the accuracy of conventional strategies like reflection tomography and FWI (Øye and Dahl, 2019). A more viable methodology may be to involve AI to aid explicit pieces of the model-building process that are presently work escalated.

One potential application is in reflection tomography, where remaining moveout (RMO) picking is a basic however relentless step. AI could help via computerizing the recognizable proof and evacuation of terrible RMO picks, which would some way or another require huge manual remedy. This paper investigates a technique for incorporating AI into this interaction.

Method

Reflection tomography customarily includes a few stages: RMO picking, gamma filtering, occasion picking, speed refreshing, beam following, and movement (Figure 1). Robotizing these means can smooth out the model-building process. The strategy introduced mechanizes the RMO picking step utilizing AI to diminish manual mediation.

In this methodology, AI is utilized to group RMO picks as either legitimate or invalid. A subset of the movement accumulates is at first handled with predefined edges to naturally pick RMOs. Human mediation is then used to address these picks, and this rectified information is utilized to prepare an AI model. The prepared model is then applied to the full dataset, mechanizing the RMO picking process minus any additional human info.

The AI model is prepared utilizing elements, for example, spatial directions, filtered gamma esteem, top similarity, and nearby reflector plunge. These highlights are standardized and utilized in regulated characterization to distinguish awful RMO picks.

Field Examples

Example 1: Small North Sea Dataset

A little dataset from the North Ocean, with 12 inlines and 841 crosslines, was utilized for starting testing. Arbitrary Woodland and Brain Organization classifiers were assessed for their exhibition in distinguishing terrible RMO picks.

  • Arbitrary Woods Classifier: Accomplished 100 percent exactness on the preparation information. The model’s component significance showed that spatial directions and gamma values were generally compelling. The classifier’s outcomes were steady when applied to the full dataset, really eliminating profound area RMOs that were delegated terrible.
  • Brain Organization Classifier: Tried with different hyperparameters, the Brain Organization accomplished almost 100% precision however battled with order execution in more profound areas. The outcomes were practically identical to Irregular Woods in the shallow districts yet less solid in general.

Example 2: Larger North Sea Dataset

A bigger dataset with 721 inlines, 1057 crosslines, and 40 balances was utilized for additional testing. The Arbitrary Backwoods classifier again performed well, with 100 percent exactness on the preparation information and powerful characterization of RMO picks. Notwithstanding, the Brain Organization classifier, in spite of its underlying commitment, detailed just 86% precision and performed inadequately in the profound districts.

Understanding Raw Shot Gathers

Shot accumulates are major to seismic information handling. They address the assortment of seismic information recorded at different geophones or recipients following a solitary seismic source or “shot” occasion. These accumulates catch the reflected seismic waves from various subsurface layers, giving a depiction of the seismic reaction to a specific source.

Understanding Raw Shot Gathers

Crude shot accumulates, nonetheless, are innately loud and complex. They incorporate different wellsprings of commotion like ecological circumstances, hardware blunders, and surface waves, which can darken the significant seismic signs. Removing helpful data from these uproarious datasets generally required critical manual mediation by geophysicists. With the coming of AI, this preprocessing step can now be mechanized and streamlined, working with additional proficient and exact information examination.

Importance of Velocity Models

Speed models are pivotal for deciphering seismic information since they portray how seismic waves travel through various subsurface materials. By understanding the speed of seismic waves, geophysicists can gather the sort and design of subsurface arrangements, including the presence of various stone sorts, liquids, and geographical designs.

Exact speed models lead to better seismic imaging, which is fundamental for:

Asset Investigation: Distinguishing possible areas for oil, gas, and minerals.

Quake Chance Evaluation: Understanding subsurface designs to foresee seismic movement.

Boring Choices: Settling on informed decisions about where to penetrate and how to oversee boring tasks.

Blunders in speed models can bring about exorbitant boring errors, confused information, or botched open doors. Hence, working on the precision and productivity of speed model structure is fundamentally important, which AI procedures can essentially upgrade.

Traditional Methods of Velocity Model Building

All things considered, speed model structure was a manual and iterative cycle. Geophysicists would physically decipher seismic information, build beginning models, and afterward utilize these models to reproduce seismic wave spread. This interaction included a few phases:

Information Obtaining: Gathering crude seismic information through field overviews.

Information Handling: Cleaning and arranging the information to eliminate commotion and right for twists.

Manual Translation: Master geophysicists would decipher the handled information, frequently requiring broad experience and information.

Model Structure: Building speed models in light of the deciphered information and running reenactments to refine these models.

This approach was successful however tedious and inclined to human mistake. It likewise battled with adaptability as the size and intricacy of datasets expanded with current seismic reviews.

Challenges in Velocity Model Building

Building exact speed models from seismic information includes a few difficulties:

Information Clamor: Seismic information is frequently boisterous due to ecological and gear factors, requiring broad preprocessing.

Computational Force: The reversal interaction to get speeds from seismic information can be computationally costly and complex.

Subjectivity: Manual understanding can be emotional, with various specialists possibly offering differing translations of similar information.

Volume of Information: Present day seismic overviews create huge measures of information, making it hard to process and decipher physically.

These difficulties have driven the investigation of AI as a way to mechanize and upgrade the speed model structure process.

Types of Machine Learning Used in Seismic Data

Types of Machine Learning Used in Seismic Data

A few AI methods are utilized in seismic information handling, each offering various benefits:

  1. Administered Realizing: This strategy includes preparing models on named information where the results are known. With regards to speed demonstrating, administered learning can be utilized to anticipate speed models for new shot assembles in view of recently named models. Strategies, for example, brain organizations and backing vector machines are ordinarily utilized.
  2. Solo Learning: Unaided learning doesn’t need marked information. It is utilized for assignments like bunching seismic information into various locales in light of closeness. Strategies, for example, k-implies bunching and various leveled grouping are instances of solo learning procedures applied to seismic information.
  3. Support Realizing: This strategy includes preparing a specialist to pursue choices in view of criticism from the climate. In seismic information handling, support learning can upgrade dynamic cycles, for example, picking the best boundaries for model structure.

Raw Shot Gathers to Velocity Model: Process Overview

The most common way of building a speed model from crude shot accumulates utilizing AI includes a few key stages:

  1. Information Preprocessing: Crude shot accumulate information is cleaned to eliminate commotion and right for contortions. This step is essential as it guarantees that the information took care of into the AI models is precise and solid.
  2. Highlight Extraction: Significant elements are removed from the shot accumulates. These elements could incorporate travel time, adequacy, recurrence content, and other seismic traits that are important for demonstrating.
  3. Model Preparation: AI models are prepared on a dataset of named shot assembles where the speed models are known. This preparing permits the models to gain proficiency with the connections between seismic properties and speed values.
  4. Expectation and Approval: When prepared, the models can anticipate speed models for new, unlabeled shot accumulates. These forecasts are approved against extra information or contrasted with results from conventional strategies to guarantee precision.

Data Preprocessing for Machine Learning

Information preprocessing is a basic move toward applying AI to seismic information. It includes:

Commotion Expulsion: Wiping out clamor from the crude information to work on the nature of the contribution for AI models.

Standardization: Changing the information to a typical scale to guarantee reliable handling.

Highlight Designing: Extricating and choosing important elements from the shot assembles that will be utilized by the AI models.

Compelling preprocessing guarantees that the AI models are prepared on top notch information, which is fundamental for accomplishing exact and solid outcomes.

Feature Engineering from Shot Gathers

Include designing includes getting significant properties from crude shot accumulates that can be utilized by AI models. Normal elements include:

Seismic Qualities: Like recurrence, stage, and envelope abundancy.

Dimensionality Decrease: Procedures like head part investigation (PCA) can diminish the intricacy of the information while holding significant examples.

Highlight designing aides in working on the information and working on the proficiency of AI models, particularly while managing enormous and complex seismic datasets.

Labeling in Machine Learning for Velocity Models

Labeling in Machine Learning for Velocity Models

In managed picking up, naming includes appointing the right speed model to each shot assemble in the preparation set. This cycle can be testing and normally requires:

Manual Translation: Specialists physically decipher seismic information to make marked datasets.

Engineered Information: Utilizing forward displaying to create manufactured shot assembles with realized speed models.

Precise marking is vital for preparing compelling AI models, as it gives the premise to learning and expectation.

How Machine Learning Transforms Velocity Model Building

AI presents a critical change in speed model structure via robotizing and improving the cycle. Here is an itemized see how AI is changing this field:

1. Automated Feature Extraction

AI calculations succeed in robotizing the extraction of highlights from seismic shot accumulates. Not at all like manual picking, ML models can recognize and remove significant elements from crude information with high exactness. This robotization diminishes the time and exertion expected for include extraction and limits human blunder, prompting more dependable and reliable outcomes.

2. Advanced Pattern Recognition

Profound learning methods inside AI are especially adroit at perceiving complex examples in seismic information. These models can uncover unpretentious connections and elements that may be disregarded by customary techniques. Accordingly, ML calculations give more precise and nitty gritty speed models, upgrading the general nature of subsurface understanding.

3. Predictive Modeling

AI calculations can be prepared to anticipate speed models in light of verifiable information. By dissecting broad volumes of past seismic information, ML models figure out how to gauge speeds with high accuracy. This prescient capacity works on the exactness of speed models and supports better dynamic in investigation and asset the executives.

4. Handling Large Datasets

Seismic overviews produce huge measures of information, which can be trying to handle utilizing customary strategies. AI calculations are intended to productively deal with and investigate enormous datasets. This ability is critical in current seismic investigation projects, where information volumes are persistently expanding. ML’s capacity to process broad datasets takes into consideration thorough examination and exact model structure.

Case Studies and Real-World Applications

Case Study 1: Oil and Gas Exploration

An oil and gas investigation organization as of late integrated AI into their speed model structure process. By using ML calculations, the organization had the option to break down seismic information all the more rapidly and precisely. The outcome was a more exact speed model that superior the distinguishing proof of potential penetrating areas. This headway expanded the achievement pace of investigation as well as altogether decreased costs related with penetrating activities.

Case Study 2: Mineral Exploration

A mining organization applied AI to enhance mineral investigation. By utilizing ML models, the organization handled huge volumes of seismic information all the more effectively. This approach prompted better asset assessment and designated boring. The utilization of ML additionally helped in limiting investigation gambles and enhancing asset the executives, showing the innovation’s viability in different geophysical applications.

Facts:

  1. Seismic Information Translation: Accurate velocity models are crucial for interpreting seismic data in subsurface investigations, such as oil and gas extraction, environmental studies, and geotechnical engineering.
  2. Traditional Methods: Building velocity models traditionally involves manual interpretation and extensive calculations through techniques like tomography and full-waveform inversion (FWI). These methods are time-consuming and computationally intensive.
  3. Machine Learning (ML) Integration: ML techniques are being applied to streamline the velocity model-building process. ML can automate tasks such as picking residual moveout (RMO) and extracting features from seismic data.
  4. Advantages of ML:
    • Automated Feature Extraction: ML models can automatically identify and extract relevant features from seismic data, reducing manual effort and potential errors.
    • Advanced Pattern Recognition: Deep learning methods can recognize complex patterns in seismic data, leading to more accurate velocity models.
    • Predictive Modeling: ML models can predict velocity models based on historical data, improving accuracy and decision-making.
    • Handling Large Datasets: ML algorithms are designed to process and analyze large volumes of seismic data efficiently.
  5. Case Studies:
    • Oil and Gas Exploration: ML integration in velocity modeling has improved the accuracy and efficiency of identifying drilling locations, reducing costs.
    • Mineral Exploration: ML has enhanced the assessment of mineral resources and optimized drilling operations, demonstrating its effectiveness in geophysical applications.

Summary:

Velocity model building from raw shot gathers using machine learning represents a significant advancement in seismic data processing. Traditionally, this process involved labor-intensive methods such as tomography and FWI, which required substantial manual input and computational resources. Machine learning offers a transformative approach by automating feature extraction, improving pattern recognition, and handling large datasets efficiently. The integration of ML techniques enhances the accuracy of velocity models and supports better decision-making in fields like oil and gas exploration and mineral assessment. Real-world applications have demonstrated that ML can reduce costs and improve outcomes by optimizing the seismic data interpretation process.

FAQs:

1. What is a velocity model in seismic imaging?

  • A velocity model represents the speed at which seismic waves travel through different subsurface materials. It is essential for interpreting seismic data and understanding the structure and composition of the Earth’s subsurface.

2. How does machine learning improve velocity model building?

  • Machine learning improves velocity model building by automating tasks such as feature extraction and RMO picking, recognizing complex patterns in seismic data, and handling large datasets more efficiently. This leads to more accurate and reliable velocity models.

3. What are some traditional methods for building velocity models?

  • Traditional methods include tomography and full-waveform inversion (FWI). Tomography involves iterative updates to the velocity model based on seismic data, while FWI is a computationally intensive process that requires significant human input.

4. What are the benefits of using machine learning for seismic data processing?

  • Machine learning provides several benefits, including reduced manual effort, improved accuracy in pattern recognition, enhanced predictive modeling, and the ability to efficiently process large volumes of seismic data.

5. Can machine learning replace traditional methods entirely?

  • While machine learning offers significant advantages, it may not entirely replace traditional methods. Instead, it complements them by automating and optimizing specific tasks within the velocity model-building process, leading to improved efficiency and accuracy.

6. What are some real-world applications of machine learning in seismic data processing?

  • Real-world applications include improving oil and gas exploration by identifying better drilling locations and enhancing mineral exploration by optimizing resource assessment and drilling operations.

For more Information About Technology visit LatestSession.com

Leave a Reply

Your email address will not be published. Required fields are marked *