Geological disaster early warning method and accurate early warning system based on multi-source data fusion

Abstract

The invention discloses a geological disaster early warning method and a precise early warning system based on multi-source data fusion, and relates to the technical field of geological disaster early warning. According to the method, multi-source heterogeneous data such as remote sensing, meteorological and geological monitoring are fused, a standardized protocol is utilized to unify a data format and temporal-spatial resolution, a standardized data set is formed, key features are extracted by adopting principal component analysis and a recursive feature elimination algorithm, and a long-short-term memory network and a convolutional neural network model are combined, so that the real-time performance of the system is improved. According to the method, the disaster risk is accurately predicted, the space risk distribution diagram is generated, in addition, through application of the real-time stream processing framework and the self-adaptive learning algorithm, rapid distribution of early warning signals and dynamic optimization of model parameters are achieved, the accuracy and timeliness of an early warning system are remarkably improved, and the geological disaster risk is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster early warning technology, and in particular to a geological disaster early warning method and a precise early warning system based on multi-source data fusion. Background Art

[0002] Research on geological disaster early warning is a key area in the field of disaster prevention and mitigation, directly related to the safety of people's lives and property and social stability. With global climate change and intensified human activities, the frequency and destructive power of geological disasters such as landslides, collapses, and mudslides are increasing. Efficient and accurate early warning technologies are urgently needed to reduce disaster risks. Traditional geological disaster early warning methods often rely on a single data source or simple model analysis, such as predictions based solely on meteorological forecasts or geological exploration data. When faced with complex terrain and changing environments, such methods often suffer from insufficient data dimensions or insufficient analysis depth, resulting in low prediction accuracy and poor timeliness, making it difficult to meet the needs of real-time and regional adaptability. The core challenge lies in how to effectively integrate multi-source heterogeneous data and achieve efficient analysis. First, remote sensing, meteorological, geological exploration, and real-time monitoring data vary significantly in format and temporal and spatial resolution, making data fusion difficult. Second, existing algorithms cannot simultaneously ensure computational efficiency and pattern recognition accuracy when processing large-scale multidimensional data. Finally, the complex mechanism of disaster occurrence makes it difficult for a single model to capture the dynamic characteristics of multiple coupled factors, resulting in untimely and unreliable early warning signals. Therefore, how to efficiently integrate multi-source data and optimize the analysis process through intelligent algorithms to achieve accurate and real-time geological disaster risk prediction has become a key issue that needs to be urgently addressed in this study. Summary of the Invention

[0003] The purpose of the present invention is to provide a geological disaster early warning method and a precise early warning system based on multi-source data fusion, and to provide an efficient and precise solution for geological disaster early warning by utilizing multi-source data fusion and intelligent algorithm optimization.

[0004] The purpose of the present invention can be achieved through the following technical solutions:

[0005] This application provides a geological disaster early warning method based on multi-source data fusion, which includes the following steps:

[0006] Acquire multi-source heterogeneous data from remote sensing, meteorological, and geological monitoring equipment, use standardized protocol conversion formats, unify spatiotemporal resolutions, and obtain standardized data sets;

[0007] Principal component analysis is used to extract remote sensing image features, meteorological dynamic parameters, and geological static properties of the normalized dataset to obtain a multidimensional feature vector set. When the feature dimension in the multidimensional feature vector set exceeds a preset threshold, a recursive feature elimination algorithm is used to filter key features to obtain a streamlined feature set.

[0008] Based on the simplified feature set, a long short-term memory network model is used to capture the time series dependency between meteorological dynamic parameters and geological static attributes, and to obtain the trend of disaster risk changes.

[0009] Based on the changing trend of disaster risk, a convolutional neural network is used to extract the spatial correlation characteristics of complex terrain to obtain a spatial risk distribution map. When the risk value of a certain area in the spatial risk distribution map exceeds the preset threshold, the gradient boosting decision tree algorithm is used to calculate the probability of disaster occurrence in the area and obtain the regional risk probability value;

[0010] According to the regional risk probability value, a real-time stream processing framework is used to generate early warning signals and distribute them to the monitoring terminal to obtain real-time early warning results. Then, an adaptive learning algorithm is used to update the long short-term memory network model parameters according to the newly collected data to obtain an optimized prediction model.

[0011] Furthermore, a normalized data set is obtained, specifically including:

[0012] Acquire remote sensing, meteorological, and geological data from multiple monitoring devices, using pre-set acquisition interface protocols to obtain raw data sets;

[0013] The original dataset is converted into XML or JSON format through a standardized protocol to obtain a unified format dataset. When the spatiotemporal resolution of the unified format dataset is inconsistent, the spatial and temporal resolution is adjusted through an interpolation algorithm to obtain a dataset with consistent resolution.

[0014] Based on the dataset with consistent resolution, data fusion technology is used to integrate remote sensing, meteorological and geological features to obtain a fused dataset;

[0015] Based on the fused dataset, a normalized dataset is generated and stored in a distributed database to obtain the final dataset.

[0016] Furthermore, a multi-dimensional feature vector set is obtained, specifically including:

[0017] Principal component analysis is used to extract remote sensing image features, meteorological dynamic parameters, and geological static properties from the normalized data set to obtain an initial feature set. When the dimension of the initial feature set is higher than the preset threshold, principal component analysis is used to reduce the dimension to obtain a reduced-dimensional feature set.

[0018] According to the dimensionality reduction feature set, the principal component contribution rate of each feature is calculated to determine the main feature vector, and then the cluster analysis method is used to group the main feature vectors to obtain a multidimensional feature vector set;

[0019] Through the vector set generation algorithm, the multi-dimensional feature vector set is integrated to generate the final feature vector set. When the resolution of the final feature vector set is lower than the preset threshold, the remote sensing image features are spatially enhanced to obtain the optimized feature vector set.

[0020] Furthermore, a simplified feature set is obtained, specifically including:

[0021] Determine whether the dimension of the feature vector set exceeds the preset threshold. If it exceeds, calculate the importance score of each feature through the recursive feature elimination algorithm to obtain the feature importance ranking;

[0022] Using feature importance sorting, the features with the lowest importance are removed one by one from the feature vector set to obtain a temporary feature subset;

[0023] Evaluate the model performance of the temporary feature subset through cross-validation to determine whether the performance meets the preset standards and obtain a performance evaluation result. If the performance evaluation result meets the preset standards, the temporary feature subset is determined as a streamlined feature set; if it does not meet the standards, repeatedly remove the lowest importance features and evaluate to obtain a new temporary feature subset;

[0024] Based on the streamlined feature set, the correlation matrix between the features is calculated, and the feature combinations with correlations lower than the preset threshold are obtained to obtain the low-correlation feature set.

[0025] Furthermore, the disaster risk change trend is obtained, including:

[0026] The time series dependency is obtained by using the long short-term memory network to model the time series of the simplified feature set.

[0027] Based on the time series dependencies, the interaction weights between dynamic meteorological parameters and static geological attributes are calculated to identify the driving factors of disaster risk changes;

[0028] Using a long-short-term memory network prediction model, we generate disaster risk trends based on driving factors and time series dependencies. When the confidence level of the predicted trend exceeds a preset threshold, we perform spatial interpolation on the trend to generate a regional disaster risk distribution map.

[0029] According to the regional disaster risk distribution map, cluster analysis is used to divide high-risk areas and generate disaster risk zoning results.

[0030] Furthermore, a spatial risk distribution map is obtained, including:

[0031] Preprocessing tools are used to grid-segment complex terrain data to obtain a terrain grid dataset. Convolutional neural networks are then used to extract features from the terrain grid dataset to obtain a spatial correlation feature set.

[0032] When the dimension of the spatial correlation feature set exceeds the preset threshold, the spatial correlation feature set is processed by the feature selection algorithm based on information gain to obtain the filtered feature subset;

[0033] Through the classification layer of the convolutional neural network, a preliminary disaster risk distribution map is generated based on the filtered key feature set, and then a smoothing filter is used for denoising to obtain a smoothed risk distribution map;

[0034] Based on the smoothed risk distribution map, the risk change trend is judged by superimposing time series data, and a dynamic risk distribution map is generated. Spatial features are extracted from the dynamic risk distribution map, and visualization tools are used to generate the final disaster risk spatial distribution map.

[0035] Furthermore, the regional risk probability value is obtained, including:

[0036] Determine whether the risk value of a certain area in the spatial risk distribution map exceeds the preset threshold, and then standardize the risk distribution data of the area to obtain a standardized risk data set;

[0037] The standardized risk data set is trained and predicted using a gradient boosting decision tree algorithm to obtain the probability of regional disaster occurrence. When the probability of regional disaster occurrence exceeds the probability threshold, a local weighted regression analysis is performed on the risk distribution map of the region using the spatial analysis module to obtain the risk distribution trend.

[0038] According to the risk distribution trend, the kernel density estimation method is used to locate the high-risk points in the region and determine the coordinate set of the high-risk points;

[0039] The spatial clustering algorithm is used to group the coordinate sets of high-risk points to obtain a set of high-risk areas. When the area of a certain area in the set of high-risk areas exceeds the area threshold, the environmental factor data of the area is integrated through the data fusion module to obtain a comprehensive environmental risk feature.

[0040] According to the comprehensive environmental risk characteristics, the logistic regression algorithm was used to conduct a secondary risk assessment of the area to obtain the final risk level.

[0041] Furthermore, a simplified feature set is obtained, specifically including:

[0042] Obtain regional risk probability values, use stream processing technology to parse data streams in real time, obtain structured risk data, and then compare the structured risk data with preset thresholds. If the risk probability value exceeds the threshold, an early warning signal is generated.

[0043] The warning signal is encoded using a signal generation algorithm to obtain a standardized signal format, which is then transmitted to the monitoring terminal through a distribution mechanism to determine the signal distribution status;

[0044] Obtain distribution status feedback from the monitoring terminal to determine whether the signal is successfully delivered. If the signal is not successfully delivered, retransmit the warning signal through the backup channel to obtain the final distribution result;

[0045] According to the distribution results and risk probability values, the regional risk status is updated to obtain real-time warning results.

[0046] Furthermore, an optimized prediction model is obtained, which specifically includes:

[0047] Adaptive algorithms are used to calculate long-term memory weights and short-term memory weights to determine network model parameters. When the network model parameter changes exceed a preset threshold, the gradient descent method is used to update the long-term and short-term memory networks to obtain an optimized model.

[0048] The newly collected data is inferred by the optimization model to obtain the prediction results. Then, according to the deviation between the prediction results and the actual data, the adaptive algorithm parameters are updated using a dynamic adjustment method to obtain the adjusted algorithm.

[0049] When the prediction error of the adjusted algorithm is lower than the preset threshold, a real-time warning result is generated through the optimization model. Based on the real-time warning result, feedback data is obtained, the newly collected data is updated, and the next round of data processing begins.

[0050] This application provides a geological disaster early warning system based on multi-source data fusion, which is used to implement a geological disaster early warning method based on multi-source data fusion, including:

[0051] The data acquisition and preprocessing module is used to obtain raw data from multiple heterogeneous data sources such as remote sensing, meteorological, and geological monitoring equipment, and convert the format using standardized protocols to unify the spatiotemporal resolution and generate a standardized data set;

[0052] The feature extraction and dimensionality reduction module uses principal component analysis to extract remote sensing image features, meteorological dynamic parameters, and geological static properties from the normalized data set to form a multidimensional feature vector set. When the feature dimension exceeds the preset threshold, a recursive feature elimination algorithm is used to filter key features to obtain a streamlined feature set.

[0053] The time series analysis module uses a long short-term memory network model to capture the time series dependencies between meteorological dynamic parameters and geological static attributes based on a streamlined feature set, predict disaster risk trends, and generate regional disaster risk distribution maps;

[0054] The spatial correlation analysis module uses a convolutional neural network to extract spatial correlation features in complex terrain and generate a spatial risk distribution map. For areas exceeding a preset risk threshold, the gradient boosting decision tree algorithm is used to calculate the probability of disaster occurrence in the area and obtain the regional risk probability value.

[0055] The real-time warning and signal distribution module uses a real-time stream processing framework to generate warning signals based on regional risk probability values and distributes them to monitoring terminals to obtain real-time warning results.

[0056] The model optimization and adaptive learning module uses an adaptive learning algorithm to update the LSTM network model parameters according to the newly collected data, optimize the prediction model, and update the adaptive algorithm parameters through dynamic adjustment methods.

[0057] The beneficial effects of the present invention are:

[0058] This invention integrates multi-source heterogeneous data such as remote sensing, meteorology, geological exploration, and real-time monitoring, and adopts a standardized protocol conversion format to unify spatiotemporal resolution to obtain a normalized data set. This effectively solves the problem of difficult data fusion, improves data availability and consistency, and provides a high-quality data foundation for subsequent analysis and early warning.

[0059] By using principal component analysis and recursive feature elimination algorithms to extract features and reduce dimensionality of normalized datasets, a streamlined feature set is obtained. This effectively addresses the computational efficiency and pattern recognition accuracy issues of existing algorithms when processing large-scale multidimensional data. Long-short-term memory network models and convolutional neural networks are used to capture time series dependencies and spatial correlation characteristics, respectively, generating disaster risk trends and spatial risk distribution maps, thereby achieving accurate and real-time geological disaster risk prediction. This not only improves the accuracy and timeliness of early warnings, but also enhances the system's regional adaptability.

[0060] Early warning signals are generated through a real-time stream processing framework and distributed to monitoring terminals, achieving rapid propagation of early warning signals. At the same time, an adaptive learning algorithm is used to update model parameters based on newly collected data and optimize the prediction model, further improving the intelligence level of the early warning system and the reliability of the prediction. This not only solves the problem of early warning signal generation being not timely and reliable in traditional methods, but also effectively reduces the risk of geological disasters through accurate early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] For better understanding and implementation, the technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0062] Figure 1 A flow chart of a geological disaster early warning method based on multi-source data fusion provided in Example 1 of the present application;

[0063] Figure 2 A schematic diagram of a process for obtaining a disaster risk change trend using a geological disaster early warning method based on multi-source data fusion provided in Example 1 of the present application;

[0064] Figure 3A schematic diagram of a process for obtaining a spatial risk distribution map using a geological disaster early warning method based on multi-source data fusion provided in Example 1 of the present application;

[0065] Figure 4 This is a structural diagram of the geological disaster early warning system based on multi-source data fusion provided in Example 2 of the present application. DETAILED DESCRIPTION

[0066] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present application. Rather, they are merely examples of methods and systems consistent with certain aspects of the present application, as detailed in the appended claims.

[0067] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0068] The following describes in detail the specific implementation methods, features and effects of the present invention in conjunction with the accompanying drawings and preferred embodiments.

[0069] Example 1

[0070] See also Figure 1-Figure 3 This embodiment provides a geological disaster early warning method based on multi-source data fusion, including the following steps:

[0071] S1. Obtain multi-source heterogeneous data from remote sensing, meteorological, and geological monitoring equipment, use standardized protocol conversion formats, unify spatiotemporal resolution, and obtain a standardized data set;

[0072] Furthermore, in step S1, a normalized data set is obtained, which specifically includes:

[0073] Acquire remote sensing, meteorological, and geological data from multiple monitoring devices, using pre-set acquisition interface protocols to obtain raw data sets;

[0074] The original dataset is converted into XML or JSON format through a standardized protocol to obtain a unified format dataset. When the spatiotemporal resolution of the unified format dataset is inconsistent, the spatial and temporal resolution is adjusted through an interpolation algorithm to obtain a dataset with consistent resolution.

[0075] Based on the dataset with consistent resolution, data fusion technology is used to integrate remote sensing, meteorological and geological features to obtain a fused dataset;

[0076] Based on the fused dataset, a normalized dataset is generated and stored in a distributed database to obtain the final dataset.

[0077] S2. Principal component analysis is used to extract remote sensing image features, meteorological dynamic parameters, and geological static properties of the normalized dataset to obtain a multidimensional feature vector set. When the feature dimension in the multidimensional feature vector set exceeds a preset threshold, a recursive feature elimination algorithm is used to screen key features to obtain a streamlined feature set.

[0078] Furthermore, in step S2, a multidimensional feature vector set is obtained, specifically including:

[0079] Principal component analysis is used to extract remote sensing image features, meteorological dynamic parameters, and geological static properties from the normalized data set to obtain an initial feature set. When the dimension of the initial feature set is higher than the preset threshold, principal component analysis is used to reduce the dimension to obtain a reduced-dimensional feature set.

[0080] According to the dimensionality reduction feature set, the principal component contribution rate of each feature is calculated to determine the main feature vector, and then the cluster analysis method is used to group the main feature vectors to obtain a multidimensional feature vector set;

[0081] Through the vector set generation algorithm, the multi-dimensional feature vector set is integrated to generate the final feature vector set. When the resolution of the final feature vector set is lower than the preset threshold, the remote sensing image features are spatially enhanced to obtain the optimized feature vector set.

[0082] Furthermore, in step S2, a simplified feature set is obtained, specifically including:

[0083] Determine whether the dimension of the feature vector set exceeds the preset threshold. If it exceeds, calculate the importance score of each feature through the recursive feature elimination algorithm to obtain the feature importance ranking;

[0084] Using feature importance sorting, the features with the lowest importance are removed one by one from the feature vector set to obtain a temporary feature subset;

[0085] Evaluate the model performance of the temporary feature subset through cross-validation to determine whether the performance meets the preset standards and obtain a performance evaluation result. If the performance evaluation result meets the preset standards, the temporary feature subset is determined as a streamlined feature set; if it does not meet the standards, repeatedly remove the lowest importance features and evaluate to obtain a new temporary feature subset;

[0086] Based on the streamlined feature set, the correlation matrix between the features is calculated, and the feature combinations with correlations lower than the preset threshold are obtained to obtain the low-correlation feature set.

[0087] S3. Based on the simplified feature set, a long short-term memory network model is used to capture the time series dependency between meteorological dynamic parameters and geological static attributes to obtain the trend of disaster risk changes;

[0088] Furthermore, in step S3, the disaster risk change trend is obtained, including:

[0089] S31. Use the long short-term memory network to perform time series modeling on the simplified feature set to obtain time series dependencies;

[0090] S32. Calculate the interaction weights between dynamic meteorological parameters and static geological attributes based on time series dependencies to determine the driving factors of disaster risk changes;

[0091] S33. Generate disaster risk change trends based on driving factors and time series dependencies using a long short-term memory network prediction model. When the confidence level of the predicted trend exceeds a preset threshold, perform spatial interpolation on the change trend to obtain a regional disaster risk distribution map.

[0092] S34. Based on the regional disaster risk distribution map, cluster analysis is used to divide high-risk areas and generate disaster risk zoning results.

[0093] Specifically, in one possible implementation, when the long short-term memory network is used for time series modeling, the core is to capture the temporal dependencies in a streamlined feature set. For example, for geological disaster risk prediction, the streamlined feature set may include features such as rainfall, surface settlement rate, and soil moisture. The long short-term memory network analyzes the dynamic changes of these features in the time dimension through memory units and gating mechanisms. For example, assuming that there is rainfall in a certain area for 30 consecutive days, the network can identify the trend of rainfall gradually increasing and then decreasing. Combined with the surface settlement rate, it can extract the dependency of accelerated settlement rate after a surge in rainfall. This modeling method can effectively capture long-term and short-term time patterns.

[0094] Specifically, when calculating the interaction weights between meteorological dynamic parameters and geological static attributes, this can be achieved through the attention mechanism. Meteorological dynamic parameters such as rainfall intensity and geological static attributes such as rock stability, the interaction weights reflect the mutual influence between the two. For example, the rainfall intensity in a certain area is 50 mm / hour and the rock stability rating is level 3. The attention mechanism may calculate that the weight of rainfall intensity on disaster risk is 0.7, while the rock stability is 0.3, which indicates that rainfall is the main driving factor.

[0095] Preferably, the weights are optimized through multiple iterations to ensure the accuracy of the driving factors. In one embodiment, the long short-term memory network prediction model generates a disaster risk change trend based on the driving factors and time series dependencies. For example, the model predicts that the disaster risk in a certain area will change from low to high in the next 7 days with a confidence level of 0.85, which is higher than the preset threshold of 0.8. It should be noted that the confidence level is calculated through the probability distribution of the model output to ensure the reliability of the prediction. At this time, the trend can be further used for spatial interpolation processing.

[0096] For example, the inverse distance weighted interpolation method, combined with monitoring point data, maps risk trends onto a two-dimensional grid, generating a regional disaster risk distribution map that visually displays the distribution of high-risk areas. It is understandable that when cluster analysis is used to delineate high-risk areas, the K-means algorithm can be used. For example, based on the risk distribution map, setting K = 3 divides the region into high, medium, and low risk zones. The risk value of the central grid in a certain region is 0.9, while the values of the surrounding grids are 0.7 and 0.6, respectively. After clustering, the central grid is classified as a high-risk zone. Specifically, the clustering results can be used to generate disaster risk zoning maps to guide resource allocation. This data-driven zoning approach enhances the targeted nature of regional management. For example, in the implementation of a combination of spatial interpolation and clustering, monitoring point data in a mountainous area is sparse. After interpolation, a continuous risk distribution map is generated. Cluster analysis further identifies areas with high landslide risk. This multi-step collaborative analysis, from time series modeling to spatial distribution generation, ensures the comprehensiveness and accuracy of disaster risk assessment.

[0097] S4. Based on the changing trend of disaster risk, a convolutional neural network is used to extract the spatial correlation characteristics of complex terrain to obtain a spatial risk distribution map. When the risk value of a certain area in the spatial risk distribution map exceeds the preset threshold, the probability of disaster occurrence in the area is calculated using the gradient boosting decision tree algorithm to obtain the regional risk probability value;

[0098] Furthermore, in step S4, a spatial risk distribution map is obtained, which specifically includes:

[0099] S41. Using a preprocessing tool to perform grid segmentation on the complex terrain data to obtain a terrain grid dataset, and then performing feature extraction on the terrain grid dataset using a convolutional neural network to obtain a spatial correlation feature set;

[0100] S42: When the dimension of the spatial correlation feature set exceeds a preset threshold, the spatial correlation feature set is processed by a feature selection algorithm based on information gain to obtain a filtered feature subset.

[0101] The feature selection algorithm based on information gain calculates the information gain of each feature for geological hazard classification. The higher the information gain, the greater the influence of the feature on the classification result. By setting an information gain threshold, features with information gain greater than the threshold are selected as key features, and those with less contribution to geological hazard classification are filtered out.

[0102] S43, generating a preliminary disaster risk distribution map based on the filtered key feature set through the classification layer of the convolutional neural network, and then using a smoothing filter to perform denoising to obtain a smoothed risk distribution map;

[0103] Among them, the classification layer of the convolutional neural network will classify different areas according to different key features, determine the disaster risk level of each area, and thus draw a preliminary disaster risk distribution. The smoothing filter will average or weighted average the pixel values in the image to reduce the noise points in the image, make the boundaries of the risk distribution smoother, and make the risk distribution clearer.

[0104] S44. Based on the smoothed risk distribution map, the risk change trend is judged by superimposing time series data, and a dynamic risk distribution map is generated. The smoothed risk distribution maps at different time points are superimposed to analyze the changes in risk areas, the rise and fall of risk levels, etc., in order to predict the future risk change trend; spatial features are extracted from the dynamic risk distribution map, and visualization tools are used to generate the final disaster risk spatial distribution map.

[0105] Specifically, grid segmentation of complex terrain data is an important preprocessing step in disaster risk assessment. For example, for geological disaster risk analysis in a mountainous area, the preprocessing tool can segment the terrain data into a fixed grid size, such as each grid covering an area of 100 meters × 100 meters, to generate a terrain grid data set containing slope, elevation, and slope direction. In one embodiment, the tool extracts terrain features through a digital elevation model with a slope range of 0 to 45 degrees and an elevation range of 500 to 2000 meters to ensure that the grid data reflects the complexity of the terrain; when extracting spatial correlation features through a convolutional neural network, the network uses convolution kernels to capture spatial patterns between grids.

[0106] Specifically, the convolution kernel size is set to 3×3, and the terrain grid dataset is scanned to extract the spatial correlation between slope changes and elevation distribution. For example, a grid with a sharp slope change in a certain area may overlap with an area with a sudden elevation change. The convolutional neural network recognizes this spatial pattern and forms a feature set containing spatial correlation features. When the dimension of the feature set is too high, a feature selection algorithm based on information gain is used for dimensionality reduction. Preferably, the algorithm calculates the contribution of each feature to the disaster risk and eliminates features whose contribution is lower than a threshold of 0.1.

[0107] For example, the information gain of the slope change feature is 0.8, while that of secondary features such as aspect is only 0.05. The algorithm retains features related to slope and elevation and generates a filtered feature subset. This approach effectively reduces computational complexity. The classification layer of the convolutional neural network generates a preliminary disaster risk distribution map based on this filtered feature subset. It can be understood that the classification layer maps the feature subset to risk levels, such as low, medium, and high risk. For example, if a grid has a slope of 30 degrees and a sudden change in elevation, the classification layer predicts its risk level as high, generating a preliminary distribution map that visually displays high-risk areas. A smoothing filter is used to remove noise and optimize the distribution map quality. In one possible implementation, the preliminary distribution map is processed using a Gaussian filter with a filter kernel size of 5×5 to eliminate isolated noise points. For example, if the risk value of a grid is abnormally high, but the surrounding grids are all low-risk, the grid value is smoothed after filtering to generate a smoothed risk distribution map, improving the reliability of the visualization.

[0108] Time series data overlay is used to analyze the trend of risk changes. For example, the rainfall data and surface displacement data of the past 30 days are combined and superimposed on a smoothed risk distribution map to determine the rise and fall of risk levels. For example, the rainfall in a certain area increases from 20 mm / day to 50 mm / day, and the surface displacement rate accelerates. The dynamic risk distribution map shows that the risk rises from low to high, and it is predicted that the risk will continue to rise in the next 7 days. When extracting spatial features from the dynamic risk distribution map and generating the final disaster risk spatial distribution map, visualization tools such as heat maps can be used. It should be noted that heat maps use color depth to indicate risk levels, with red indicating high risk. For example, the risk value of the central grid in a mountainous area is 0.9, and the value of the surrounding grids is 0.6. The heat map clearly shows that high-risk areas are concentrated in the steep slopes of the mountain. This visualization method intuitively presents spatial distribution characteristics and facilitates the allocation of disaster prevention and control resources.

[0109] Furthermore, in step S4, the regional risk probability value is obtained, which specifically includes:

[0110] Determine whether the risk value of a certain area in the spatial risk distribution map exceeds the preset threshold, and then standardize the risk distribution data of the area to obtain a standardized risk data set;

[0111] The standardized risk data set is trained and predicted using a gradient boosting decision tree algorithm to obtain the probability of regional disaster occurrence. When the probability of regional disaster occurrence exceeds the probability threshold, a local weighted regression analysis is performed on the risk distribution map of the region using the spatial analysis module to obtain the risk distribution trend.

[0112] According to the risk distribution trend, the kernel density estimation method is used to locate the high-risk points in the region and determine the coordinate set of the high-risk points;

[0113] The spatial clustering algorithm is used to group the coordinate sets of high-risk points to obtain a set of high-risk areas. When the area of a certain area in the set of high-risk areas exceeds the area threshold, the environmental factor data of the area is integrated through the data fusion module to obtain a comprehensive environmental risk feature.

[0114] According to the comprehensive environmental risk characteristics, the logistic regression algorithm was used to conduct a secondary risk assessment of the area to obtain the final risk level.

[0115] Specifically, for example, determining whether the risk value of a certain area in the spatial risk distribution map exceeds a preset threshold is a key step in disaster risk assessment. The risk value is usually calculated based on factors such as terrain and rainfall, reflecting the possibility of a disaster. In one possible implementation, suppose a risk distribution map of a mountainous area is represented in grid form, each grid covers 100 meters × 100 meters, the risk value range is 0 to 1, and the preset threshold is 0.7. Specifically, the risk value of a certain grid is 0.8, which exceeds the threshold and is marked as a high-risk area. This judgment method is intuitive and efficient, and facilitates rapid screening of key areas.

[0116] It should be noted that standardization converts risk distribution data into a unified scale to improve the accuracy of subsequent algorithms. In one embodiment, the minimum-maximum normalization method is used to map the risk value to the range of 0 to 1. For example, the risk value range of a certain area is 0.3 to 0.9, and the standardized value is 0 to 1, eliminating dimensional differences. This method retains the data distribution characteristics and is suitable for heterogeneous data analysis.

[0117] Preferably, the gradient boosting decision tree algorithm predicts the probability of disaster occurrence by integrating multiple weak classifiers. In one possible implementation, the algorithm uses a standardized risk data set as input and is trained in combination with features such as slope and rainfall. For example, if the slope of an area is 25 degrees and the rainfall is 40 mm / day, the algorithm predicts a disaster probability of 0.85, which exceeds the probability threshold of 0.8 and triggers subsequent analysis. This method enhances prediction robustness through iterative optimization.

[0118] It can be understood that local weighted regression analysis is used to capture the trend of risk distribution and emphasize the spatial correlation of local data. In one embodiment, for high-probability areas, the regression analysis is centered on the grid, assigning higher weights to adjacent grids to generate a smooth trend graph. For example, the risk value of the central grid in a certain area is higher, and the values of the surrounding grids decrease. The regression analysis reveals that the risk increases along the slope direction. This method highlights local characteristics and is suitable for complex terrain analysis. Specifically, kernel density estimation locates high-risk points and identifies high-risk concentrated areas through probability density functions. In one possible implementation, the kernel density is based on a Gaussian kernel and the bandwidth is set to 50 meters.

[0119] For example, kernel density analysis of a certain mountainous area shows that high-risk points are concentrated in areas with slopes greater than 30 degrees, with coordinate sets such as (x1, y1) and (x2, y2). This positioning method is intuitive and facilitates precise prevention and control. For example, the spatial clustering algorithm groups high-risk points to form a set of high-risk areas. In one embodiment, the DBSCAN algorithm is used, and the neighborhood radius is set to 100 meters and the minimum number of points is 5. For example, after clustering a certain area, three high-incidence areas are obtained, with areas of 0.5, 0.8, and 1.2 square kilometers, respectively. Areas exceeding the area threshold of 1 square kilometer enter the next step of analysis. This method effectively aggregates spatial data and reduces misjudgments.

[0120] In one possible implementation, the data fusion module integrates environmental factor data, such as soil moisture and vegetation coverage, to generate comprehensive environmental risk characteristics. For example, the soil moisture in a high-incidence area is 70% and the vegetation coverage is only 20%. The fused characteristics indicate high environmental vulnerability. This fusion method comprehensively characterizes the risk drivers and improves the reliability of the assessment. Preferably, the logistic regression algorithm performs a secondary risk assessment based on the comprehensive environmental risk characteristics and outputs the final risk level. In one embodiment, the algorithm takes the feature value as input and predicts the risk level as low, medium, or high. For example, the characteristics of a certain area show high soil moisture and low vegetation coverage, and logistic regression predicts a high risk level. This method enhances classification accuracy through probabilistic modeling, which facilitates disaster warning and resource allocation.

[0121] S5. Based on the regional risk probability value, a real-time stream processing framework is used to generate warning signals and distribute them to the monitoring terminal to obtain real-time warning results. Then, an adaptive learning algorithm is used to update the long short-term memory network model parameters according to the newly collected data to obtain an optimized prediction model.

[0122] Furthermore, in step S5, a simplified feature set is obtained, specifically including:

[0123] Obtain regional risk probability values, use stream processing technology to parse data streams in real time, obtain structured risk data, and then compare the structured risk data with preset thresholds. If the risk probability value exceeds the threshold, an early warning signal is generated.

[0124] The warning signal is encoded using a signal generation algorithm to obtain a standardized signal format, which is then transmitted to the monitoring terminal through a distribution mechanism to determine the signal distribution status;

[0125] Obtain distribution status feedback from the monitoring terminal to determine whether the signal is successfully delivered. If the signal is not successfully delivered, retransmit the warning signal through the backup channel to obtain the final distribution result;

[0126] According to the distribution results and risk probability values, the regional risk status is updated to obtain real-time warning results.

[0127] Furthermore, in step S5, an optimized prediction model is obtained, which specifically includes:

[0128] Adaptive algorithms are used to calculate long-term memory weights and short-term memory weights to determine network model parameters. When the network model parameter changes exceed a preset threshold, the gradient descent method is used to update the long-term and short-term memory networks to obtain an optimized model.

[0129] The newly collected data is inferred by the optimization model to obtain the prediction results. Then, according to the deviation between the prediction results and the actual data, the adaptive algorithm parameters are updated using a dynamic adjustment method to obtain the adjusted algorithm.

[0130] When the prediction error of the adjusted algorithm is lower than the preset threshold, a real-time warning result is generated through the optimization model. Based on the real-time warning result, feedback data is obtained, the newly collected data is updated, and the next round of data processing begins.

[0131] Example 2

[0132] See also Figure 4 This embodiment provides a geological disaster early warning system based on multi-source data fusion, which is used to implement a geological disaster early warning method based on multi-source data fusion, including:

[0133] The data acquisition and preprocessing module is used to obtain raw data from multiple heterogeneous data sources such as remote sensing, meteorological, and geological monitoring equipment, and convert the format using standardized protocols to unify the spatiotemporal resolution and generate a standardized data set. It also includes data fusion technology to integrate feature information from different sources and store the fused data in a distributed database.

[0134] The feature extraction and dimensionality reduction module uses principal component analysis to extract remote sensing image features, meteorological dynamic parameters, and geological static properties from the normalized data set to form a multidimensional feature vector set. When the feature dimension exceeds the preset threshold, a recursive feature elimination algorithm is used to filter key features to obtain a streamlined feature set.

[0135] The time series analysis module uses a long short-term memory network model to capture the time series dependencies between meteorological dynamic parameters and geological static attributes based on a streamlined feature set, predict disaster risk trends, and generate regional disaster risk distribution maps;

[0136] The spatial correlation analysis module uses a convolutional neural network to extract spatial correlation features in complex terrain and generate a spatial risk distribution map. For areas exceeding a preset risk threshold, the gradient boosting decision tree algorithm is used to calculate the probability of disaster occurrence in the area and obtain the regional risk probability value.

[0137] The real-time warning and signal distribution module uses a real-time stream processing framework to generate warning signals based on regional risk probability values and distributes them to monitoring terminals to obtain real-time warning results. It is also used to obtain distribution status feedback from the monitoring terminals to ensure the successful delivery of warning signals.

[0138] The model optimization and adaptive learning module adopts an adaptive learning algorithm to update the LSTM network model parameters according to the newly collected data, optimize the prediction model, and update the adaptive algorithm parameters through dynamic adjustment methods to reduce the prediction error and improve the accuracy and reliability of the early warning system.

[0139] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A geological disaster early warning method based on multi-source data fusion, characterized by: The steps include: Acquire multi-source heterogeneous data from remote sensing, meteorological, and geological monitoring equipment, use standardized protocol conversion formats, unify spatiotemporal resolutions, and obtain standardized data sets; Principal component analysis is used to extract remote sensing image features, meteorological dynamic parameters, and geological static properties of the normalized dataset to obtain a multidimensional feature vector set. When the feature dimension in the multidimensional feature vector set exceeds a preset threshold, a recursive feature elimination algorithm is used to filter key features to obtain a streamlined feature set. Based on the simplified feature set, a long short-term memory network model is used to capture the time series dependency between meteorological dynamic parameters and geological static attributes, and to obtain the trend of disaster risk changes. Based on the changing trend of disaster risk, a convolutional neural network is used to extract the spatial correlation characteristics of complex terrain to obtain a spatial risk distribution map. When the risk value of a certain area in the spatial risk distribution map exceeds the preset threshold, the gradient boosting decision tree algorithm is used to calculate the probability of disaster occurrence in the area and obtain the regional risk probability value; According to the regional risk probability value, a real-time stream processing framework is used to generate early warning signals and distribute them to the monitoring terminal to obtain real-time early warning results. Then, an adaptive learning algorithm is used to update the long short-term memory network model parameters according to the newly collected data to obtain an optimized prediction model.

2. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized by: Get the normalized data set, including: Acquire remote sensing, meteorological, and geological data from multiple monitoring devices, using pre-set acquisition interface protocols to obtain raw data sets; The original dataset is converted into XML or JSON format through a standardized protocol to obtain a unified format dataset. When the spatiotemporal resolution of the unified format dataset is inconsistent, the spatial and temporal resolution is adjusted through an interpolation algorithm to obtain a dataset with consistent resolution. Based on the dataset with consistent resolution, data fusion technology is used to integrate remote sensing, meteorological and geological features to obtain a fused dataset; Based on the fused dataset, a normalized dataset is generated and stored in a distributed database to obtain the final dataset.

3. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized in that: Get a multidimensional feature vector set, including: Principal component analysis is used to extract remote sensing image features, meteorological dynamic parameters, and geological static properties from the normalized data set to obtain an initial feature set. When the dimension of the initial feature set is higher than the preset threshold, principal component analysis is used to reduce the dimension to obtain a reduced-dimensional feature set. According to the dimensionality reduction feature set, the principal component contribution rate of each feature is calculated to determine the main feature vector, and then the cluster analysis method is used to group the main feature vectors to obtain a multidimensional feature vector set; Through the vector set generation algorithm, the multi-dimensional feature vector set is integrated to generate the final feature vector set. When the resolution of the final feature vector set is lower than the preset threshold, the remote sensing image features are spatially enhanced to obtain the optimized feature vector set.

4. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized in that: The reduced feature set is obtained, including: Determine whether the dimension of the feature vector set exceeds the preset threshold. If it exceeds, calculate the importance score of each feature through the recursive feature elimination algorithm to obtain the feature importance ranking; Using feature importance sorting, the features with the lowest importance are removed one by one from the feature vector set to obtain a temporary feature subset; Evaluate the model performance of the temporary feature subset through cross-validation to determine whether the performance meets the preset standards and obtain a performance evaluation result. If the performance evaluation result meets the preset standards, the temporary feature subset is determined as a streamlined feature set; if it does not meet the standards, repeatedly remove the lowest importance features and evaluate to obtain a new temporary feature subset; Based on the streamlined feature set, the correlation matrix between the features is calculated, and the feature combinations with correlations lower than the preset threshold are obtained to obtain the low-correlation feature set.

5. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized in that: Obtain disaster risk change trends, including: The time series dependency is obtained by using the long short-term memory network to model the time series of the simplified feature set. Based on the time series dependencies, the interaction weights between dynamic meteorological parameters and static geological attributes are calculated to identify the driving factors of disaster risk changes; Using a long-short-term memory network prediction model, we generate disaster risk trends based on driving factors and time series dependencies. When the confidence level of the predicted trend exceeds a preset threshold, we perform spatial interpolation on the trend to generate a regional disaster risk distribution map. According to the regional disaster risk distribution map, cluster analysis is used to divide high-risk areas and generate disaster risk zoning results.

6. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized in that: Obtain a spatial risk distribution map, including: Preprocessing tools are used to grid-segment complex terrain data to obtain a terrain grid dataset. Convolutional neural networks are then used to extract features from the terrain grid dataset to obtain a spatial correlation feature set. When the dimension of the spatial correlation feature set exceeds the preset threshold, the spatial correlation feature set is processed by the feature selection algorithm based on information gain to obtain the filtered feature subset; Through the classification layer of the convolutional neural network, a preliminary disaster risk distribution map is generated based on the filtered key feature set, and then a smoothing filter is used for denoising to obtain a smoothed risk distribution map; Based on the smoothed risk distribution map, the risk change trend is judged by superimposing time series data, and a dynamic risk distribution map is generated. Spatial features are extracted from the dynamic risk distribution map, and visualization tools are used to generate the final disaster risk spatial distribution map.

7. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized by: Get the regional risk probability value, including: Determine whether the risk value of a certain area in the spatial risk distribution map exceeds the preset threshold, and then standardize the risk distribution data of the area to obtain a standardized risk data set; The standardized risk data set is trained and predicted using a gradient boosting decision tree algorithm to obtain the probability of regional disaster occurrence. When the probability of regional disaster occurrence exceeds the probability threshold, a local weighted regression analysis is performed on the risk distribution map of the region using the spatial analysis module to obtain the risk distribution trend. According to the risk distribution trend, the kernel density estimation method is used to locate the high-risk points in the region and determine the coordinate set of the high-risk points; The spatial clustering algorithm is used to group the coordinate sets of high-risk points to obtain a set of high-risk areas. When the area of a certain area in the set of high-risk areas exceeds the area threshold, the environmental factor data of the area is integrated through the data fusion module to obtain a comprehensive environmental risk feature. According to the comprehensive environmental risk characteristics, the logistic regression algorithm was used to conduct a secondary risk assessment of the area to obtain the final risk level.

8. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized by: The reduced feature set is obtained, including: Obtain regional risk probability values, use stream processing technology to parse data streams in real time, obtain structured risk data, and then compare the structured risk data with preset thresholds. If the risk probability value exceeds the threshold, an early warning signal is generated. The warning signal is encoded using a signal generation algorithm to obtain a standardized signal format, which is then transmitted to the monitoring terminal through a distribution mechanism to determine the signal distribution status; Obtain distribution status feedback from the monitoring terminal to determine whether the signal is successfully delivered. If the signal is not successfully delivered, retransmit the warning signal through the backup channel to obtain the final distribution result; According to the distribution results and risk probability values, the regional risk status is updated to obtain real-time warning results.

9. The geological disaster early warning method based on multi-source data fusion according to claim 1 is characterized in that: The optimized prediction model is obtained, including: Adaptive algorithms are used to calculate long-term memory weights and short-term memory weights to determine network model parameters. When the network model parameter changes exceed a preset threshold, the gradient descent method is used to update the long-term and short-term memory networks to obtain an optimized model. The newly collected data is inferred by the optimization model to obtain the prediction results. Then, according to the deviation between the prediction results and the actual data, the adaptive algorithm parameters are updated using a dynamic adjustment method to obtain the adjusted algorithm. When the prediction error of the adjusted algorithm is lower than the preset threshold, a real-time warning result is generated through the optimization model. Based on the real-time warning result, feedback data is obtained, the newly collected data is updated, and the next round of data processing begins.

10. A geological disaster early warning system based on multi-source data fusion, used to implement the geological disaster early warning method based on multi-source data fusion according to any one of claims 1 to 9, characterized in that: include: The data acquisition and preprocessing module is used to obtain raw data from multiple heterogeneous data sources such as remote sensing, meteorological, and geological monitoring equipment, convert the data into different formats using standardized protocols, unify the temporal and spatial resolutions, and generate standardized data sets. The feature extraction and dimensionality reduction module uses principal component analysis to extract remote sensing image features, meteorological dynamic parameters, and geological static properties from the normalized data set to form a multidimensional feature vector set. When the feature dimension exceeds the preset threshold, a recursive feature elimination algorithm is used to filter key features to obtain a streamlined feature set. The time series analysis module uses a long short-term memory network model to capture the time series dependencies between meteorological dynamic parameters and geological static attributes based on a streamlined feature set, predict disaster risk trends, and generate regional disaster risk distribution maps; The spatial correlation analysis module uses a convolutional neural network to extract spatial correlation features in complex terrain and generate a spatial risk distribution map. For areas exceeding a preset risk threshold, the gradient boosting decision tree algorithm is used to calculate the probability of disaster occurrence in the area and obtain the regional risk probability value. The real-time warning and signal distribution module uses a real-time stream processing framework to generate warning signals based on regional risk probability values and distributes them to monitoring terminals to obtain real-time warning results. The model optimization and adaptive learning module uses an adaptive learning algorithm to update the LSTM network model parameters according to the newly collected data, optimize the prediction model, and update the adaptive algorithm parameters through dynamic adjustment methods.

Images