Earthquake prediction multi-source remote sensing data set-oriented pre-earthquake abnormal information extraction method

Abstract

The invention discloses a pre-earthquake abnormal information extraction method for an earthquake prediction multi-source remote sensing data set. The method specifically comprises the following steps: S1, obtaining and preprocessing multi-source remote sensing data; S2, establishing an earthquake abnormal feature extraction model based on an IBPT framework; S3, extracting and analyzing earthquake abnormal information; according to the invention, by introducing the multi-source remote sensing data and combining the IBPT algorithm, the recognition precision of the seismic abnormal signal is improved, and the problem of prediction instability caused by single parameter analysis in the traditional method is solved; and by utilizing model optimization and evaluation, the accuracy and generalization ability of earthquake prediction are enhanced, and the pre-earthquake abnormal information can be extracted more efficiently and accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite pre-earthquake abnormal signal recognition, and in particular to a method for extracting pre-earthquake abnormal information from a multi-source remote sensing data set for earthquake prediction. Background Art

[0002] In recent years, earthquake prediction has received increasing attention as an important means of disaster prevention and mitigation. However, the complexity and randomness of earthquakes make their prediction still a global technical challenge. Traditional earthquake monitoring methods mainly rely on ground observation equipment such as seismographs, but these methods have certain limitations in coverage and timeliness. With the development of remote sensing technology, satellite observations provide a new perspective for earthquake prediction. Satellites can continuously observe physical changes in the earth's surface, atmosphere, ionosphere and other layers from space over a large range, and effectively capture abnormal changes in multiple parameters before and after earthquakes. As early as 2006, NASA's "Aqua" satellite and "MODIS" sensor began to provide a large amount of infrared and hyperspectral data for global seismic activities. These data are widely used in the study of earthquake precursors, such as surface temperature anomalies, cloud changes and ionospheric disturbances. However, early studies focused on the analysis of a single parameter, and it was difficult to verify the universality and reliability of these abnormal signals on a global scale. With the increase in the amount of satellite data and the advancement of artificial intelligence technology, researchers have begun to explore the integrated analysis of multi-source remote sensing data, and extract pre-earthquake abnormal information from complex multi-dimensional data through machine learning methods to improve the accuracy and efficiency of earthquake prediction. Therefore, how to use multi-source remote sensing data, especially hyperspectral and infrared data, to develop a set of technical methods to accurately identify pre-earthquake abnormal information has become an important research direction in the current field of earthquake prediction. Summary of the invention

[0003] The purpose of the present invention is to provide a method for extracting abnormal information from multi-source remote sensing data sets for earthquake prediction based on the Inverse Boosted Pruned Tree (IBPT) framework of AdaBoost, so as to solve the problems of complex multi-source data processing and insufficient prediction accuracy in existing earthquake prediction methods.

[0004] To achieve the above objectives, the present invention provides the following technical solutions: including S1: multi-source remote sensing data acquisition and preprocessing, S2: establishment of earthquake anomaly feature extraction model based on IBPT framework, S3: earthquake anomaly information extraction and analysis, S4: earthquake prediction performance evaluation and model optimization.

[0005] Step S1: Multi-source remote sensing data acquisition and preprocessing Multi-source remote sensing data including infrared, hyperspectral and gas monitoring were selected, covering 1,371 earthquake events with a magnitude of 6 or above from 2006 to 2013 worldwide. The preprocessing steps include: first, the raw data is processed by sliding time window technology, with the time window length set to 5 days and the sliding interval set to 1 day, and discontinuous abnormal data is removed. At the same time, data standardization is performed to ensure the homogeneity of different types of remote sensing data, making it suitable for subsequent feature extraction.

[0006] Step S2: Establishment of earthquake anomaly feature extraction model based on IBPT framework Based on the preprocessed data, the inverse boosted pruning tree (IBPT) algorithm is used to construct a seismic anomaly feature extraction model. The IBPT framework uses multiple pruned decision trees as base learners, and continuously adjusts the sample weights to enhance the model's prediction ability for difficult-to-classify data. First, 80% of the training set data is used for model training. The IBPT algorithm is used to continuously optimize the model structure through iterative training, and pruning reduces the possibility of overfitting. The final feature classification model is generated through the complexity penalty term and sample weighting mechanism of the decision tree.

[0007] Step S3: Extraction and analysis of earthquake anomaly information Based on the IBPT model established in step S2, the abnormal features of multi-source remote sensing data before the earthquake are extracted, focusing on parameters closely related to earthquake precursors such as hyperspectral infrared radiation, total ozone and CO content. The remote sensing data of each earthquake event is compared with the simulated data of non-seismic events, and the IBPT model is used to identify potential pre-earthquake abnormal information and analyze the spatiotemporal distribution characteristics of abnormal changes before the earthquake. The model is updated through multiple iterations to gradually optimize the accuracy of feature extraction.

[0008] Step S4: Earthquake prediction performance evaluation and model optimization After the abnormal feature extraction is completed, the remaining 20% ​​of the test set data is used to verify the model. A variety of evaluation indicators, such as ROC curve, MCC (Matthews correlation coefficient), AUC (area under the curve), etc., are used to evaluate the classification accuracy and prediction performance of the model. Combined with the prediction results, the model hyperparameters are tuned, such as adjusting the number of decision trees, pruning depth, and learning rate, to improve the generalization ability of the model. In practical applications, the stability and reliability of the IBPT model are optimized by testing earthquake events of different magnitudes and in different regions.

[0009] Compared with the prior art, the invention has the following beneficial effects: by introducing multi-source remote sensing data and combining the IBPT algorithm, the recognition accuracy of earthquake abnormal signals is improved, and the problem of unstable prediction caused by single parameter analysis in traditional methods is solved. This method is not only suitable for large-scale data set processing, but also can extract pre-earthquake abnormal information more efficiently and accurately by continuously optimizing model parameters, thereby improving earthquake prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 Workflow diagram of the Inverse Enhanced Pruned Tree (IBPT) framework. DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example

[0012] (1) Multi-source remote sensing data acquisition and preprocessing: First, the project selected infrared and hyperspectral remote sensing data obtained from 1,371 earthquake events with a magnitude of 6 or above worldwide from 2006 to 2013. The data included surface temperature, total ozone, CO content, etc. All data were obtained through NASA's Aqua satellite and NOAA's OLR data platform, with a temporal and spatial resolution of 1.0°. To ensure data homogeneity, the project used sliding time window technology for preprocessing, setting the time window length to 5 days and the sliding interval to 1 day. After the data was standardized, it was denoised to remove abnormal data caused by satellite payload interference to ensure data accuracy.

[0013] To enhance the robustness of the model, the project also generated 1,371 non-seismic events (simulated earthquakes) and matched these simulated earthquake events with real earthquake events based on the temporal and spatial extent of the actual earthquake events.

[0014] (2) Establishment of earthquake anomaly feature extraction model based on IBPT framework: After completing data preprocessing, the project built an earthquake anomaly feature extraction model based on the Inverse Enhanced Pruned Tree (IBPT) algorithm. The IBPT framework uses pruned decision trees as base learners and uses a set of feature parameters (such as total ozone, CO content, infrared radiation, etc.) for feature extraction and classification. The project first uses 80% of the data for model training and 20% of the data for model testing. In each iteration, by calculating the Gini impurity, the project prunes the decision tree to reduce the risk of overfitting, and by increasing the weight of correctly classified samples and reducing the weight of incorrectly classified samples, the generalization ability of the model is improved.

[0015] After the model was built, a variety of performance evaluation indicators (such as ROC curve, AUC, MCC, etc.) were used to verify the classification performance of the model. The AUC of the final model reached 0.87, showing a high classification accuracy.

[0016] (3) Extraction and analysis of earthquake anomaly information: Using the trained IBPT model, the project extracts pre-earthquake anomaly information from pre-processed multi-source remote sensing data. This method uses the model to identify pre-earthquake anomaly features, especially focusing on the spatiotemporal changes of infrared radiation, total ozone, and CO content. Through time series analysis technology, the project can accurately locate significant anomaly features within 15-30 days before an earthquake, and then predict the time and epicenter of a potential earthquake.

[0017] (4) Earthquake prediction performance evaluation and model optimization: After the model was initially completed, the project used the remaining 20% ​​test set to evaluate its prediction performance. By comparing with traditional classification models such as random forests and deep neural networks, the IBPT model showed higher prediction accuracy in multiple indicators such as MCC and AUC. Based on the test results, the project further adjusted the pruning depth and learning rate of the model to optimize the final classification effect. Through iterative optimization, the IBPT model showed good generalization ability in applications of different earthquake sizes and geographical regions.

[0018] (5) Earthquake monitoring application and model verification: Based on the optimized IBPT model, the project applied the model to real-time earthquake monitoring. Through actual earthquake case verification, the model can effectively capture abnormal signals before earthquakes and show high accuracy in practical applications. In order to further optimize the model, the project also introduced the Bayesian hyperparameter optimization method to adjust the model parameters so that the model can achieve the best performance on the actual data set.

[0019] Although embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for extracting pre-earthquake abnormal information from multi-source remote sensing data sets for earthquake prediction, characterized in that: The steps include: Step S1: Multi-source remote sensing data acquisition and preprocessing: Multi-source remote sensing data including infrared, hyperspectral and gas monitoring are selected, covering 1371 earthquake events with magnitude 6 and above from 2006 to 2013 worldwide. The preprocessing steps include: first, the raw data is processed by sliding time window technology, the time window length is set to 5 days, the sliding interval is 1 day, discontinuous abnormal data is removed, and data standardization is performed to ensure the homogeneity of different types of remote sensing data, so that it is suitable for subsequent feature extraction; Step S2: Establishment of earthquake anomaly feature extraction model based on IBPT framework: Based on the preprocessed data, the inverse enhanced pruned tree IBPT algorithm is used to construct a seismic anomaly feature extraction model. The IBPT framework uses multiple pruned decision trees as base learners, and continuously adjusts the sample weights to enhance the model's prediction ability for difficult-to-classify data. First, 80% of the training set data is used for model training. The IBPT algorithm is used to continuously optimize the model structure through iterative training, and pruning reduces the possibility of overfitting. The final feature classification model is generated through the complexity penalty term of the decision tree and the sample weighting mechanism. Step S3: Extraction and analysis of earthquake anomaly information: Based on the IBPT model established in step S2, the abnormal features of multi-source remote sensing data before the earthquake are extracted, focusing on parameters closely related to earthquake precursors such as hyperspectral infrared radiation, total ozone and CO content. The remote sensing data of each earthquake event is compared with the simulated data of non-earthquake events. The IBPT model is used to identify potential pre-earthquake abnormal information, analyze the spatiotemporal distribution characteristics of abnormal changes before the earthquake, and gradually optimize the accuracy of feature extraction through multiple iterations to update the model. Step S4: Earthquake prediction performance evaluation and model optimization: After the abnormal feature extraction is completed, the remaining 20% ​​of the test set data is used to verify the model. A variety of evaluation indicators, such as ROC curve, Matthews correlation coefficient, area under the AUC curve, etc., are used to evaluate the classification accuracy and prediction performance of the model. Combined with the prediction results, the model hyperparameters are tuned, such as adjusting the number of decision trees, pruning depth and learning rate, to improve the generalization ability of the model. In practical applications, the stability and reliability of the IBPT model are optimized by testing earthquake events of different magnitudes and in different regions.

2. The method for extracting pre-earthquake abnormal information from a multi-source remote sensing data set for earthquake prediction according to claim 1, characterized in that: The IBPT framework in step S2 combines a decision tree model and an adaptive boosting algorithm, in which each basic classifier prioritizes high-error samples through weighted allocation. This method reduces the interference of noise samples on model accuracy by pruning decision trees, and uses Gini impurity and complexity penalty factors to ensure optimal pruning to generate the final earthquake anomaly detection model.

3. The method for extracting pre-earthquake abnormal information from a multi-source remote sensing dataset for earthquake prediction according to claim 2, characterized in that: The pruning process of the IBPT framework uses the weighted cost complexity method. First, a complete decision tree is constructed on the training data. Then, the pruned candidate tree is calculated based on the Gini coefficient of the node. By evaluating the training error of each candidate pruned tree, the pruned tree with the smallest error is selected as the final model, ensuring that the most informative nodes are retained when processing pre-earthquake abnormal signals, thereby optimizing the generalization ability of the overall model.

4. The method for extracting pre-earthquake abnormal information from a multi-source remote sensing dataset for earthquake prediction according to claim 3, characterized in that: The earthquake anomaly information extraction in step S3 is carried out through a detailed analysis of the spatiotemporal dynamic changes of multi-source remote sensing parameters. The remote sensing parameters involved include: atmospheric infrared radiation, surface temperature, total ozone, total carbon dioxide, methane concentration and water vapor mixing ratio. The sliding time window technology is used to segment the time series data of each remote sensing parameter to ensure that no pre-earthquake abnormal signals are missed.

5. The method for extracting pre-earthquake abnormal information from a multi-source remote sensing data set for earthquake prediction according to claim 4, characterized in that: The data preprocessing process includes using Z-Score standardization to eliminate scale differences in remote sensing data. At the same time, interpolation algorithms are used to complete data for occasional data missing due to satellite equipment failures or space environment influences. Data processing also combines the dynamic time warping DTW algorithm to align the time series changes in remote sensing data to ensure a clear distinction between pre-earthquake abnormal signals and environmental noise.

6. The method for extracting pre-earthquake abnormal information from a multi-source remote sensing data set for earthquake prediction according to claim 5, characterized in that: The earthquake abnormal signal extraction in step S3 is achieved by fusing multiple remote sensing data sets of different types, using dynamic clustering technology to automatically classify them in different time windows, separating the pre-earthquake abnormal signals from normal meteorological and geophysical phenomena, and continuously adjusting the model parameters through the back propagation algorithm to improve the sensitivity to pre-earthquake abnormal signals.

7. The method for extracting pre-earthquake abnormal information from a multi-source remote sensing data set for earthquake prediction according to claim 6, characterized in that: In step S4, the earthquake prediction performance of the model is evaluated using a variety of evaluation indicators, such as the area under the ROC curve (AUC), the Mathews correlation coefficient (MCC), and the F1 score, to ensure that the model can effectively distinguish between pre-earthquake anomalies and background noise in real applications. During the evaluation process, the model is also applied to multiple earthquake datasets around the world to verify its cross-regional robustness and applicability, and the model parameters are further optimized through cross-validation.

8. The method for extracting pre-earthquake abnormal information from multi-source remote sensing data sets for earthquake prediction according to claim 7, characterized in that: The multi-source remote sensing data in step S1 include Atmospheric Infrared Sounder data from NASA's Aqua satellite and Outgoing Longwave Radiation data from NOAA. These data have a spatial resolution of 1°×1° and are collected twice a day, ensuring the breadth and accuracy of pre-earthquake data coverage. In addition, the remote sensing data are processed by the Kalman filter algorithm before being input into the model to eliminate sudden outliers and enhance the reliability of the data.

9. The method for extracting pre-earthquake abnormal information from multi-source remote sensing data sets for earthquake prediction according to claim 8, characterized in that: During the model optimization process, the grid search method is used to tune the hyperparameters to determine the optimal pruning tree depth, number of leaf nodes, and number of iterations to ensure that the model performs optimally on different earthquake data sets. Five-fold cross validation is used to reduce overfitting, and feature importance analysis is combined to determine the most critical remote sensing parameters for earthquake prediction.

Images