Deep Learning Microseismic Magnitude Calculation Using DAS Strain Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current microseismic magnitude calculation methods are inaccurate due to reliance on conventional geophone seismic data, lacking the ability to effectively utilize microseismic strain data from Distributed Acoustic Sensors (DAS) and failing to capture the complexity of microseismic events.
Innovation Solution
A real-time microseismic magnitude calculation method based on deep learning that directly inputs DAS strain data into a magnitude calculation module, utilizing a 3-layer convolution structure to extract frequency and waveform characteristics, and employing a model fusion with fully connected layers to output calculated magnitudes, without converting strain data to conventional data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional geophone seismic data and empirical formulas are used for magnitude calculation, then the calculation process is simple, but the accuracy of magnitude estimation is insufficient
Solution Approach 1:
The patent replaces traditional mechanical/empirical magnitude calculation methods with a deep learning-based neural network system. The neural network automatically learns complex patterns from raw DAS strain data, substituting the need for manual feature extraction and empirical formulas, thereby improving accuracy while managing complexity through automated processing.
Solution Approach 2:
The patent transforms the input data representation from conventional geophone seismic data to DAS strain data, and changes the calculation approach from empirical formulas to a trained neural network model. This parameter change enables the system to capture more nuanced characteristics of microseismic events, improving magnitude estimation accuracy.
2Measurement precision
If multiple characteristic parameters are combined to improve magnitude estimation accuracy, then the estimation accuracy improves, but the calculation complexity increases
Solution Approach 1:
The patent merges multiple characteristics (amplitude, frequency, duration, and other waveform features) into a unified deep learning model that processes them simultaneously. The neural network automatically integrates these multiple parameters through its layered architecture, achieving improved accuracy without the manual complexity of separate parameter processing.
Solution Approach 2:
The neural network performs self-service by automatically extracting and processing multiple characteristic parameters from the input data without requiring manual intervention. The model learns the optimal combinations of parameters during training, eliminating the need for researchers to manually select and process multiple parameters.
3Measurement precision
If traditional methods are used to comprehensively consider multiple factors, then the magnitude determination is thorough, but the process is time-consuming and slow
Solution Approach 1:
The patent performs preliminary action by pre-training the neural network model on extensive datasets before actual magnitude calculations. The model learns complex patterns and relationships during the training phase, enabling it to rapidly process new data in real-time without needing to perform complex calculations during the actual measurement process.
Solution Approach 2:
The patent replaces the manual, step-by-step process of considering multiple factors with an automated neural network system. The model simultaneously processes multiple characteristics through its architecture, eliminating the sequential nature of traditional methods and achieving both thoroughness and speed.
4Productivity
If DAS strain data is directly used without conversion to conventional data, then the processing speed improves, but the compatibility with existing methods is reduced
Solution Approach 1:
The patent creates a universal deep learning model that can directly process DAS strain data while maintaining the ability to handle various input formats. The model's architecture is designed to be adaptable, allowing it to process raw strain data directly without conversion, thus achieving high processing speed while maintaining versatility through its flexible input handling capabilities.
Data Source
AI summary
Embodiments of the present disclosure provide a real-time microseismic magnitude calculation method based on deep learning and a corresponding device. The method includes: constructing a DAS-based horizontal well microseismic monitoring system; constructing a training data set; constructing a magnitude calculation module, wherein the magnitude calculation module comprises two input branches of frequency spectrum and time waveform, the two input branches use a 3-layer convolution structure to extract frequency characteristic and waveform characteristic of a microseismic event, and then a model fusion is performed, and then 2 fully connected layers are used, and finally a calculated magnitude is outputted; training the magnitude calculation module; and analyzing and processing field data. The microseismic magnitude calculation method in the present disclosure improves the ability to quickly estimate the microseismic magnitude, without the need for converting the strain data, and improves the accuracy of the microseismic magnitude estimation.