Machine Learning Noise Removal in Distributed Acoustic Sensing Seismic Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Vertical seismic profiling data sets contaminated with coupling noise, such as zigzag noise, from fiber optic distributed acoustic sensing systems, which obscures down-going signals and contaminates up-going reflection data, making it difficult to analyze subsurface formation properties effectively.
Innovation Solution
A machine learning model, specifically a generative adversarial network (GAN), is used to identify and eliminate coupling noise from seismic data sets, utilizing synthetic data for training to improve noise mitigation and enhance the quality of seismic records.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fiber optic distributed acoustic sensing is used for seismic profiling, then sensing capability and coverage are improved, but coupling noise contaminates the data
Solution Approach 1:
The patent applies extraction by isolating and removing the coupling noise component from the seismic data. The system identifies and extracts the noise pattern caused by fiber-optic cable coupling with the wellbore, then separates it from the valid seismic signals to produce cleaned data.
Solution Approach 2:
The patent introduces an intermediary processing layer between data acquisition and analysis. This intermediary system uses machine learning models and signal processing algorithms to mediate the relationship between the raw noisy data and the final interpretation, filtering out harmful coupling noise while preserving valid seismic information.
2Object-affected harmful factors
If traditional noise filtering methods are used, then some noise is reduced, but down-going signals are obscured and reflection data is contaminated
Solution Approach 1:
The patent employs feedback mechanisms where the system continuously monitors the processed data quality and adjusts its noise filtering parameters accordingly. The machine learning models learn from the feedback of processed results to optimize the balance between noise removal and signal preservation, preventing information loss.
Solution Approach 2:
The patent dynamically changes processing parameters based on the specific characteristics of the seismic data. Rather than applying fixed filtering parameters, the system adapts thresholds, window sizes, and filtering strengths to match the local signal-to-noise conditions, preserving down-going signals while removing coupling noise.
3Reliability
If machine learning models are used to remove noise, then data quality is improved, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models on synthetic seismic data with known noise patterns before deploying them on real data. This preliminary training phase allows the models to learn effective noise filtering strategies offline, reducing the computational burden during actual processing while maintaining high data quality.
Solution Approach 2:
The patent uses synthetic copies of seismic data with injected coupling noise as training data for the machine learning models. By creating and processing synthetic replicas rather than requiring extensive real noisy data, the system reduces computational complexity while still effectively training models to recognize and remove actual coupling noise patterns.
Data Source
AI summary
Systems and techniques are provided for processing wellbore data to generate denoised seismic data where a coupling noise is eliminated. An example method can include receiving wellbore data comprising one or more seismic measurements; generating a seismic input image based on seismic measurements; processing the seismic input image to remove a zigzag noise in the seismic input image. The zigzag noise represents noise in the corresponding seismic measurements. The example method can further include outputting a denoised seismic image. Systems and machine-readable media are also provided.