Seismic Magnitude Forecasting Through Machine Learning Paradigms: A Confluence of Predictive Models
Seismic Magnitude Forecasting Through Machine Learning Paradigms: A Confluence of Predictive Models
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
Seismic Magnitude Forecasting through
Machine Learning Paradigms: A Confluence of
Predictive Models
1 2
Kakarla Sri Chandana Upputuri Someswara Sandeep
Department of Computer Science and Engineering, Department of Computer Science and Engineering,
Koneru Lakshmaiah Education Foundation, Vaddeswaram, Koneru Lakshmaiah Education Foundation, Vaddeswaram,
Andhra Pradesh, India Andhra Pradesh, India
3 4
Pujala Asritha Radha Mothukuri
Department of Computer Science and Engineering, Associate Professor,
Koneru Lakshmaiah Education Foundation, Vaddeswaram, Department of Computer Science and Engineering,
Andhra Pradesh, India Koneru Lakshmaiah Education Foundation, Vaddeswaram,
Andhra Pradesh, India
5
Mula Deepak Reddy
Department of Computer Science and Engineering,
Koneru Lakshmaiah Education Foundation, Vaddeswaram,
Andhra Pradesh, India
Abstract:- This study focuses largely on earthquake B. The Value of Earthquake Forecasting - Reducing
prediction, which is a crucial element of geoscience and Uncertainty via Early Detection :
emergency and disaster management. We apply state-of- The intrinsically unpredictable character of earthquake
the-art machine learning methods, most notably the occurrences underscores the vital need for credible prediction
Random Forest Regression approach, to examine the models. Early seismic activity detection and evaluation is
intricate link between geographical data analysis and vital for allowing speedy reactions, expediting evacuation
earthquake prediction. Once we have patiently traversed processes, and minimizing the amount of deaths and property
the challenges of seismic data processing, we create damage. Early warning systems are particularly critical
prediction models that deliver insights via sophisticated because they allow emergency responders the vital time they
visualization of earthquake occurrences. The research need to deploy resources properly and carry out life-saving
offers confirmation that machine learning approaches measures.
perform exceptionally well for forecasting earthquakes.
These results show the relevance of these paradigms for
enhancing, among other things, early warning systems
and catastrophic preparedness measures.
Keywords:- Seismic Forecasting; Machine Learning;
Predictive Modeling; Algorithmic Discernment; Complexity
Analysis.
I. INTRODUCTION
A. Context and Background - Necessities for Sturdy
Prediction Models :
Being emblems of the strength of nature, earthquakes
pose a severe danger to infrastructure and human habitation
all over the world. The rising need for proactive disaster
response measures to limit the devastating impacts of
seismic occurrences on communities and infrastructure is
pushing the development of effective prediction models. The
history of devastating earthquakes underlines how crucial it
is to develop prediction abilities in order to minimize deaths Fig 1 Seismic Magnitude Distribution: Unveiling Patterns
and property loss. through Histogram Analysis
IJISRT24JUN2025 www.ijisrt.com 2606
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
C. Conventional Method’s Limitations - Handling resources and prioritize activities.
Complexity in Seismic Analysis :
Standard seismic prediction systems, however H. Research Focus - Preprocessing utilizing Predictive
beneficial, are intrinsically insufficient to capture the Models :
complicated dynamics and stochastic character of earthquake The study has a wide scope and involves demanding
occurrences. These limits motivate the exploration of fresh, data preparation to assure data quality, dynamic seismic
data-driven strategies to increase forecast reliability and pattern display for exploratory analysis, and intensive
accuracy. Nonlinear seismic waves, the impact of geological training of the Random Forest Regression model. The
formations, and the unpredictable nature of fault rupture research stresses how vital local context is to enhance
processes pose obstacles that standard approaches are unable predictive modeling's accuracy and reliability. Large and
to successfully handle. important findings may arise from the focus on the complete
data lifecycle, from preprocessing to modeling, which gives a
D. Machine Learning's Function - Using Data Insights to holistic approach to seismic research.
Unlock Predictive Potential :
Machine learning technologies show considerable I. Disaster Management's Function - Creating Preventive
potential for increasing earthquake prediction skills owing to Resilience :
their capacity to identify patterns and insights from massive The study's results have a substantial influence on
datasets. A paradigm change in seismic research is now disaster management techniques and offer stakeholders with
conceivable thanks of the advent of machine learning essential information for proactive catastrophe planning.
methods, which allow the design of complicated prediction Effective earthquake prediction benefits in resource
models based on the analysis of real data. Machine learning allocation, decision-making, and community
models' versatility and scalability allow continual study and resiliencebuilding programs. Predictive models minimize
development, which finally enhances prediction accuracy. dependency on reactive techniques and increase response
operations' efficacy in disaster management frameworks,
E. Random Forest Regression Overview - A Sturdy Structure ultimately saving lives and reducing damage.
for Predictive Models :
In the realm of machine learning approaches, Random
Forest Regression looks to be a practical framework for
forecasting the depth and magnitude of earthquakes. Its
ensemble learning architecture and feature relevance ranking
algorithms allow it to execute predictive modeling and
analysis on enormous volumes of geographical data. In
seismic research, Random Forest Regression is a helpful tool
because it can handle nonlinear connections, account for
feature interactions, and give insights on variable relevance.
F. The Study's Objectives - Linking Innovation and
Traditions :
This article presents a detailed investigation of
Random Forest Regression's relevance to earthquake
prediction, combined with sophisticated spatial data
processing approaches to increase predictive insights. By
merging stateof-the-art machine learning methods with the
tried-and- true seismic analysis approach, the project intends
to develop a mutually beneficial interaction between domain
knowledge and computer intelligence. The major objective Fig 2 Geospatial Analysis of Earthquake Magnitudes:
is to make earthquake forecasts more trustworthy, accurate, Mapping Latitudinal and Longitudinal Trends
and timely so that better strategies for preparation and
responding to catastrophes may be implemented. J. Paper Structure - Steer Clear of Discussions that Might
Spark thoughts :
G. Geographic Data Analysis's Significance - The article's next portions contain a full description of
Contextualizing Predictive Precision : the approach employed, real data and debate, and closing
Our predictive modeling approach is based on our opinions on prospective routes for additional study into
geographic data, which covers essential components like earthquake prediction models. Every chapter rigorously
depth, latitude, and longitude. Our technique attempts to adheres with conference standards and highlights the
deliver more thorough and contextually rich forecasts by utilization of empirical research and innovative
adding regional data, which will allow personalized problemsolving strategies to enhance the area of seismic
catastrophic response plans. The integration of geographic analysis and disaster resilience. The scientific community and
data boosts the accuracy of forecasts made during seismic the disaster management sectors gain from the clarity,
occurrences and gives decision-makers with vital uniformity, and relevance that are supplied by the systematic
geographical context, allowing them to properly allocate presentation of facts and ideas.
IJISRT24JUN2025 www.ijisrt.com 2607
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
II. LITERATURE SURVEY performance and highlights the potential of datadriven
approaches to pass defined boundaries in seismic inversion
As the research project's author, I believe it's vital to methods.
undertake a complete analysis of current advances in deep
learning approaches connected to resource finding and By bringing out a knowledge-embedded close-looped
seismic data processing. Wang et al. [1] proved the potential deep learning framework for the intelligent inversion of
of transfer learning in resource identification challenges by multi-solution situations, Zhang et al. [11] enlarged our
offering a novel approach for predicting the attributes of shale research field and demonstrated how machine learning
gas deposits using a layered transfer learning network. In a approaches and domain expertise may be employed to address
similar line, Chen et al. [2] looked at the application of hard geophysical problems. Similar to this, Gouda et al. [12]
dictionary learning for single-channel passive seismic created a technique for lithology and fluid content analysis
denoising and offered novel signal processing approaches to that accurately calculates the litho-fluid facies distribution
enhance the quality of seismic data. employing zero-offset sound and shear impedances.
Parallel to this, Sui et al. [3] developed a Park et al. [13] proposed DeepNRMS, an unsupervised
comprehensive self-attention network specifically designed deep learning model for noise-robust CO2 tracking in
for weak seismic signal recovery in vertical seismic profile timelapse seismic photographs, in response to the necessity
data from distributed acoustic sensing, emphasising the for environmental monitoring. This gives a practical
significance of sophisticated machine learning algorithms in technique for enhancing environmental tracking skills.
tackling the particular difficulties associated with seismic Additionally, based on well-logging data, Sun et al. [14]
data analysis. To further research this issue, Zheng et al. [4] established a novel strategy for classifying fluid kinds
enhanced seismic elastic parameter inversion using a multi- combining gate recurrent unit networks and the Adaboost
task learning framework that incorporated Gated Recurrent algorithm, underscoring the prospects of hybrid machine
Units (GRU) and Fully Convolutional Regression Networks learning techniques in geological classification issues.
(FCRN). This illustrated the advantages of combining diverse
activities for improved predicted accuracy. In their conclusion, Wang et al. [15] underlined the
relevance of pre-training approaches in boosting model
Luo et al.'s [5] investigation into the application of performance and proved the usefulness of deep learning based
deep learning for seismic sound impedance inversion and on self-supervised pre-training for sandstone content
lowfrequency extension in the area of seismic data processing prediction. Numerous research studies illustrate the
highlighted the technology's potential for detecting essential tremendous potential of deep learning technology in
characteristics from seismic data. Meanwhile, Li et al. [6] increasing our knowledge of seismic occurrences and offering
focused on convolutional neural networks (CNNs) for seismic geoscientists with fresh tools for resource finding and
profile denoising, underscoring the need of employing environmental monitoring.
sophisticated deep learning approaches for successful seismic
data preprocessing. III. METHODOLOGY
Regarding the challenges with reservoir evaluation, Lu A. Preparing Data to Guarantee Accuracy :
et al. [7] illustrated how deep learning may be used to the Initially, we extensively preprocess the seismic data to
identification of subsurface rocks by recognising ultradeep find the main properties required for predictive modeling.
carbonate reservoir lithofacies using deep convolutional This approach includes the extraction of crucial variables like
neural networks. Additionally, Choi and Oh [8] devised the date, time, latitude, longitude, magnitude, and depth in
elastic-band transform, a cutting-edge approach for addition to the construction of powerful algorithms to cope
discovering and visualising seismic features that provides with missing values and assure data consistency. In addition
researchers a straightforward tool for comprehending to the data context-based imputation and exclusion
complicated seismic data patterns. procedures, timestamp conversion is crucial for standardizing
temporal data in order to assure consistent analysis. These
In order to investigate seismic data from distributed strategies also add to our dataset's comprehensiveness and
acoustic sensing (DAS), Wang et al. [9] created a multi-scale reliability. By applying tight preprocessing, we assure data
interaction network. They underlined the significance of quality and give a strong platform for subsequent study and
blending varied information scales in order to adequately model development.
portray the intricate dynamics of subsurface formations. This
work enhances the approach for managing DAS seismic data, B. Feature Engineering - Improving Forecasting Efficiency
resulting to a more detailed representation of subsurface To increase our model's prediction performance,
settings. feature engineering approaches are utilized on top of the
preprocessed input. Statistical analysis, criteria based on
Zhu et al. [10] proposed a data-driven technique based domain expertise, and sophisticated machine learning
on a multi-scale method for seismic impedance inversion to algorithms like recursive feature removal are examples of
manage concerns with seismic inversion, enabling a practical feature selection procedures. These techniques try to identify
way of identifying subterranean parameters. By incorporating and prioritize essential components that greatly effect
multi-scale data, their strategy enhances prediction meaningful information for model calibration and localized
IJISRT24JUN2025 www.ijisrt.com 2608
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
predictive modeling. This complete approach to exploratory incr ease interpretability, decrease dimensionality
earthquake forecast accuracy. Feature engineering seeks to repeatedly, leveraged to give helpful insights for our
and compress data representation. Finally, crucial traits are prediction model.
Fig 3 Geospatial Impact Analysis: Mapping Affected Areas of Earthquakes
C. Spatial Illustration - Highlighting Organizations in Space E. Selecting a Model using Random Forest Regression :
: We particularly picked Random Forest Regression as
Our work largely depends on geographic visualization, our predictive modeling framework thanks of its inherent
which gives meaningful data on the distribution of earthquake scalability, durability, and adaptability to a broad variety of
locations and magnitudes across space. With the use of datasets. Because of its ensemble learning approach, Random
complicated spatial patterns, scatter plots, global heatmaps, Forests are an ideal candidate for forecasting the depth and
and histograms, among other visualization approaches, we magnitude of earthquakes based on the evaluation of
may locate possible seismic hotspots and identify grouping geographic data. They can also manage nonlinear
phenomena. Our visualizations are strengthened by relationships and identify the importance of characteristics.
integration with Basemap technology, which offers a After a detailed examination and comparison with numerous
complete picture of seismic activity across geographic areas modeling methodologies, Random Forest Regression is
and allows extensive spatial analysis essential for predictive demonstrated to be the most successful strategy, giving the
modeling and catastrophe risk assessment. greatest possible balance between prediction accuracy, model
interpretability, and processing efficiency.
D. Examining Information Through Investigation -
Identifying Angles : F. Preparing Training Data - Guaranteeing Model
Exploratory data analysis methods are utilized to Generalization :
unearth hidden insights in the seismic data to improve To achieve model generalization and increased
geographical visualization. While correlation analysis performance on unknown data, training data is carefully
exposes links between variables that feed feature selection prepared before the model is trained. Here, stratified sampling
and model refinement procedures, descriptive statistics strategies are utilized to eliminate biases and increase model
emphasize statistical patterns and distributions. DBSCAN dependability while retaining the distributional integrity of
and K-means are two examples of spatial clustering critical features across training and testing groups.
techniques that may locate coherent spatial clusters of Additionally, feature magnitudes and model convergence
earthquakes and give data analysis gives a detailed may be enhanced during training by applying data scaling
knowledge of the underlying patterns and dynamics driving and normalization techniques. A detailed model evaluation
seismic activity, which helps in the stages that follow model and validation is based on the training data, which is
creation and informed decision-making. correctly organized to permit an investigation of prediction
performance and generalization capabilities.
IJISRT24JUN2025 www.ijisrt.com 2609
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
G. Leveraging Training Models to Develop Predictive H. Model Evaluation - Appraisal of Performance Measures :
Effectiveness : Upon completion of the model training procedure, a
Using timestamp, latitude, and longitude as input data, detailed analysis is undertaken to assess the predictive
the Random Forest Regression model is updated constantly potential of the Random Forest Regression model. To
during the model training phase. Methods for hyperparameter evaluate prediction accuracy, precision, goodness of fit, and
tweaking that reduce overfitting, boost prediction accuracy, model stability, conventional performance metrics such as
and improve model performance include grid search, mean squared error, mean absolute error, R-squared score,
Bayesian optimization, and evolutionary algorithms. In order and root mean squared logarithmic error (RMSLE) are
to offer trustworthy predictions in earthquake situations created. Comparing actual and projected values using scatter
experienced in real life, crossvalidation processes assess the plots, regression curves, and residual plots makes it easy to
model's generalization and durability across numerous analyze model performance over a variety of magnitudes and
datasets. In line with best practices in machine learning model depths. In addition, these visualizations give greater insights
design, the iterative training technique attempts to increase on prediction strengths and opportunities for development.
predictive performance indicators, induce model
convergence, and minimize prediction errors.
Fig 4 Multivariate Analysis of Earthquake Characteristics: Insights from Pairwise Feature Relationships
IJISRT24JUN2025 www.ijisrt.com 2610
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
in decision-making and resource allocation. Advanced
interpretability approaches such as partial dependency
graphs, feature contribution studies, and SHAP (Shapley
Additive Explanations) values enable a more thorough
understanding of the link between input data and model
predictions. In earthquake-prone locations, these
interpretability tools offer stakeholders with essential
information to enhance risk assessment, proactive
decisionmaking, and disaster response preparedness.
IV. RESULT & DISCUSSIONS
A. Performance Metrics Analysis :
We acquire helpful information about the Random
Forest Regression model's prediction ability after
studying its performance indicators. The model offers a fair
Fig 5 Geospatial Correlation Analysis: Exploring degree of precision in forecasting earthquake depths and
Relationships among Earthquake Metrics magnitudes, with an accuracy score of 0.38. Despite the
high degree of predicted accuracy, there is still space for
I. Sensitivity Analysis - Sturdiness Evaluation : improvement, as evidenced by the 10.08 mean absolute
Sensitivity analysis explores how changes in input error (MAE) and the 1010.62 mean squared error (MSE).
properties impact model predictions, enabling us to test the These signals indicate out places where the model needs to
stability and robustness of our prediction model. Sensitivity be modified in order to enhance overall performance and
analysis approaches give information on how varied input reduce prediction mistakes. Moreover, the R-squared score
quality impact predicted outcomes. Sensitivity plots, partial (R2) of 0.38, showing that 38% of the variation in the data is
dependency graphs, and feature importance ranking are some explained by the model, indicates the model's performance
of these approaches. The model's resilience to a variety of in contrast to the observed earthquake dataset.
operating situations is enhanced, possible sources of model
uncertainty are uncovered, and the model's refinement B. Visual Assessment - Accurate versus Approximated
process is led by this robustness evaluation. A vital aspect of Magnitudes :
model validation that offers credible and intelligible A vivid image of the relationship between the
predictions under dynamic earthquake situations is sensitivity magnitudes of the actual and expected earthquakes can be
analysis. seen in Figure 6 owing to the violin plot utilized in the visual
analysis. While there are some apparent parallels between
J. Examining the Model and Drawing Useful Inferences : the two, particularly in specific magnitude ranges, there are
In order to give meaningful information for earthquake also significant differences that indicate to regions where the
prediction and disaster management, the third element of our model's predictions would need to be altered further. This
strategy focuses on model interpretation and insights visual depiction recommends techniques to increase forecast
extraction. The information offered by the Random Forest accuracy and gives helpful information about the model's
model's feature relevance rankings, which rank the important performance at different magnitude levels.
elements impacting earthquake depth and magnitude, assists
Fig 6 Comparative Analysis of Actual and Predicted Earthquake Magnitudes: A Violin Plot Examination
IJISRT24JUN2025 www.ijisrt.com 2611
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
C. Examining Results and Model Performance : Research in this discipline will concentrate on several
When reviewing the model's performance and going key challenges in the future. First, we seek to increase the
through the data, a few significant factors become obvious. resilience and prediction capabilities of the model by
First off, despite its low accuracy, the model has a tremendous integrating new relevant data, such as geology data,
lot of promise for forecasting the depths and magnitudes of infrastructure vulnerability indices, and previous seismic
earthquakes, especially for recognizing seismic patterns and activity patterns. This improved feature set will give more
trends. The R-squared number, which shows that the model thorough and nuanced forecast insights, making proactive
has a high degree of explanatory power, supports this. disaster management tactics viable.
Nonetheless, it's vital to realize the constraints of the In order to generate dynamic predictions in the future,
model, as the fairly high MSE and MAE data illustrate. These we also wish to incorporate real-time data streams into the
measures demonstrate a certain level of uncertainty and model. This real-time connection will considerably boost the
unpredictability in the prediction accuracy, which could be model's practical usability and relevance by allowing fast
impacted by variables such as the quality of the data, feature updates and revisions to earthquake calculations in response
selection criteria, and model hyperparameters. By addressing to changing seismic occurrences and environmental
these restrictions by enhanced feature engineering, intense variables.
hyperparameter tuning, and the inclusion of more data
sources, it would be feasible to greatly boost the model's We also underline how vital it is to operate in
resilience and prediction accuracy. collaboration with seismic research institutes and topic
specialists. To evaluate our model's efficacy and assure its
applicability in actual circumstances, rigorous validation and
validation against ground truth data will be needed. By
aiming to overcome the gap between domain-specific
knowledge and improvements in machine learning, we seek
to greatly enhance existing efforts in earthquake forecasting
and disaster preparation.
REFERENCES
[1]. Wang, Min, XinPing Guo, HongMing Tang,
WeiMing Yu, Peng Zhao, and XueWen Shi. "The
prediction of shale gas reservoir parameters through
a multilayer transfer learning network." Geophysical
Journal International 234, no. 2 (2023): 1463-1475.
[2]. Chen, Yangkang, Alexandros Savvaidis, and Sergey
Fomel. "Dictionary Learning for Single ‐ Channel
Fig 7 Temporal Evolution of Earthquake Magnitudes: A Passive Seismic Denoising." Seismological Research
Comprehensive Analysis Over Time Letters 94, no. 6 (2023): 28402851.
[3]. Sui, Jilei, Yanan Tian, Yue Li, and Ning Wu.
In order to develop the model, future techniques must "Complete perception self- attention network for
coordinate the efforts of domain experts and data scientists. weak seismic signal recovery in distributed acoustic
To achieve this, specialized feature engineering procedures sensing vertical seismic profile data." Geophysics 88,
customized to seismic data quality must be devised. Model no. 6 (2023): WC107-WC119.
hyperparameters must also be modified for best outcomes, [4]. Zheng, Qiqi, Chao Wei, Xinfei Yan, Housong Ruan,
and many data sources must be merged to acquire full and Bangyu Wu. "Seismic Elastic Parameter
prediction insights. These programs attempt to enhance Inversion via a FCRN and GRU Hybrid Network
disaster management techniques and community resilience by with Multi-Task Learning." Applied Sciences 13, no.
generating more reliable and accurate earthquake prediction 18 (2023): 10519.
models. [5]. Luo, Renyu, Jinghuai Gao, Hongling Chen, Zhiqiang
Wang, and Chuangji Meng. "Deep Learning for Low-
V. CONCLUSION & FUTURE WORK Frequency Extrapolation and Seismic Acoustic
Impedance Inversion." IEEE Transactions on
Finally, we emphasize how machine learning Geoscience and Remote Sensing (2023).
approaches, such as Random Forest Regression, could [6]. Li, Shuaishuai, Jiangjie Zhang, Qian Cheng, Feng
increase earthquake prediction accuracy by evaluating Zhu, and Linong Liu. "Seismic profile denoising
geographic data. The tiny but considerable accuracy scores based on common-reflection-point gathers using
and performance indicators show that the technique has convolution neural networks." Journal of Geophysics
promise for recognizing seismic patterns and trends. Our and Engineering 20, no. 2 (2023): 240-254.
results also indicate to the necessity of continued research and
improvement, notably in lowering prediction errors and
enhancing model resilience.
IJISRT24JUN2025 www.ijisrt.com 2612
Volume 9, Issue 6, June – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24JUN2025
[7]. Lu, Shengyu, Chuyang Cai, Zhi Zhong, Zhongxian
Cai, Xu Guo, Heng Zhang, and Jie Li. "Ultradeep
carbonate reservoir lithofacies classification based
on a deep convolutional neural network—A case
study in the Tarim Basin, China." Interpretation 11,
no. 3 (2023): T551- T566.
[8]. Choi, Guebin, and Hee-Seok Oh. "Elastic-band
transform for visualization and detection." Pattern
Recognition Letters 166 (2023): 119-125.
[9]. Wang, Hongzhou, Jun Lin, Dan Shao, Xintong Dong,
and Yue Li. "Multi-scale interactive network in the
application of DAS seismic data processing."
Frontiers in Earth Science 10 (2023): 991860.
[10]. Zhu, Guang, Xiaohong Chen, Jingye Li, and
Kangkang Guo. "Data- Driven Seismic Impedance
Inversion Based on Multi-Scale Strategy." Remote
Sensing 14, no. 23 (2022): 6056.
[11]. Zhang, Fanchang, Lei Zhu, and Xunyong Xu. "A
Knowledgeembedded Close-looped Deep Learning
Framework for Intelligent Inversion of Multi-
solution Problems." Geophysics 89, no. 2 (2023): 1-
19.
[12]. Gouda, Mohammed Fathy, Abdul Halim Abdul
Latiff, and Seyed Yasser Moussavi Alashloo.
"Estimation of Litho-Fluid Facies Distribution from
Zero-Offset Acoustic and Shear Impedances."
Applied Sciences 12, no. 15 (2022): 7754.
[13]. Park, Min Jun, Julio Frigerio, Bob Clapp, and Biondo
Biondi. "DeepNRMS: Unsupervised Deep Learning
for Noise-Robust CO2 Monitoring in Time-Lapse
Seismic Images." arXiv preprint arXiv:2310.
07897 (2023).
[14]. Sun, Youzhuang, Junhua Zhang, and Yongan Zhang.
"Gate recurrent unit network combines with
Adaboost algorithm to classify fluid types by well
logging parameters." Physics of Fluids 36, no. 1
(2024).
[15]. Wang, Chong Ming, Xing Jian Wang, Yang Chen,
Xue Mei Wen, Yong Heng Zhang, and Qing Wu Li.
"Deep learning based on self- supervised pre-
training: Application on sandstone content
prediction." Frontiers in Earth Science 10 (2023):
1081998.
