Sky-ground cooperation-based exploration disturbance identification method

Abstract

The invention discloses a sky-ground cooperation-based exploration disturbance identification method, which comprises the following steps of: evaluating exploration activity distribution characteristics and disturbance probability prediction by adopting a similar mode matching result, and dynamically planning a sky platform path and residence time to obtain an optimized movement track scheme; identifying potential faults and weather interruption factors, and adjusting a path through a multi-platform cooperation mechanism to obtain a position covering observation coverage sequence; calculating the dispersion degree and the distribution density of disturbance points, and quantifying a looseness index to judge an aggregation mode to obtain clustering analysis output; comparing the disturbance dynamics feature description with a preset dangerous case feature library, and judging the occurrence of a precursor signal combination to obtain an early warning trigger signal; and scheduling encrypted observation and real-time fusion analysis by adopting an early warning trigger signal, determining a dangerous case type and a development trend, and correspondingly outputting an emergency response decision. According to the invention, the precision of disturbance monitoring and the timeliness of early warning are obviously improved, and powerful technical support is provided for risk prevention and control in a complex environment.

Description

Technical Field

[0001] This invention belongs to the field of geological exploration, and in particular relates to a method for identifying exploration disturbances based on sky-ground coordination. Background Technology

[0002] In the field of natural resource exploration and environmental monitoring, identifying disturbances caused by exploration activities is crucial for ensuring ecological security and the sustainability of resource development. Research in this area is not only fundamental to maintaining geological environmental stability but also an important means of addressing potential risks in resource development. However, current identification methods often suffer from insufficient cross-platform information integration and limited dynamic response capabilities. Especially when facing complex terrain and variable environments, it is difficult to achieve comprehensive capture and timely analysis of disturbance signals, often resulting in delayed or inaccurate monitoring results, which affects decision-making efficiency.

[0003] A deeper analysis of the challenges in this field reveals that the core technical difficulties lie in the collaborative processing of multi-source data and the time constraints of dynamic monitoring. Collaborative processing of multi-source data requires the effective fusion of data from different observation platforms, both in the sky and on the ground. For example, satellite imagery offers wide coverage but has limited resolution, while ground-based equipment data is precise but lacks sufficient coverage. This difference in spatial scale and accuracy makes it difficult to form a unified description of disturbance characteristics during data integration. Furthermore, this difference directly leads to the challenge of time constraints in dynamic monitoring, as different platforms have inconsistent data acquisition cycles and response speeds. For instance, satellites may only revisit every few days, while ground disturbances may change significantly within hours, making it difficult to guarantee real-time monitoring.

[0004] Specifically, in actual operations, exploration activities often occur in remote or complex areas, and disturbance signals may be missing due to cloud cover or equipment malfunction, preventing monitoring personnel from obtaining complete information in a timely manner. For example, during a mining exploration, ground equipment detected abnormal vibrations, but because the aerial platform failed to cover the area in time and lacked supporting image data, it was impossible to determine whether the disturbance would cause a wider geological hazard, thus missing the best opportunity for intervention.

[0005] Therefore, how to overcome the limitations of inconsistent spatial scale and temporal response based on the collaborative processing of multi-source data, and achieve rapid and accurate identification of exploration disturbance signals, has become a key problem that urgently needs to be solved. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for identifying exploration disturbances based on a sky-ground collaborative approach, comprising:

[0007] Acquire multi-source data collected by satellites and UAVs, extract spatial resolution parameters and temporal frequency parameters from the multi-source data, and process the parameters using a hierarchical segmentation method to obtain a registered sub-region fusion dataset;

[0008] Based on the registered sub-region fusion dataset, analyze the topological relationships and attribute association rules, construct a multi-dimensional index structure, and retrieve historical perturbation patterns to obtain similar pattern matching results;

[0009] Based on the similarity pattern matching results, the distribution characteristics of exploration activities and the probability of disturbance are evaluated, and the optimized movement trajectory scheme is obtained by dynamically planning the path and dwell time of the sky platform.

[0010] Based on the optimized movement trajectory scheme, potential faults and weather-related disruptions are identified, and the path is adjusted through a multi-platform collaborative mechanism to obtain a supplementary observation coverage sequence.

[0011] Based on the supplementary observation coverage sequence, spatiotemporal gaps caused by cloud cover and sensor failure are detected, and spatiotemporal interpolation is applied to estimate the missing information to obtain a compensated disturbance signal dataset.

[0012] Based on the compensated disturbance signal dataset, the dispersion and distribution density of disturbance points are calculated, and the looseness index is quantified to determine the clustering pattern and obtain the clustering analysis output.

[0013] If the clustering analysis output shows high looseness, then a vector field model is constructed by fusing satellite differential interferometry and ground tiltmeter data, and the propagation direction is analyzed to obtain a description of the disturbance dynamics characteristics;

[0014] By comparing the disturbance dynamics feature description with a preset danger feature database, the occurrence of precursor signal combinations is determined to obtain an early warning trigger signal;

[0015] Based on the early warning trigger signal, the system schedules encrypted observation and real-time fusion analysis to determine the type and development trend of the hazard and output corresponding emergency response decisions.

[0016] Preferably, the process of processing parameters using a hierarchical segmentation method to obtain the registered sub-region fusion dataset includes:

[0017] Acquire multi-source data composed of satellite imagery and UAV grids;

[0018] Spatial resolution parameters and temporal frequency parameters are extracted based on the multi-source data;

[0019] The hierarchical segmentation levels are determined based on the spatial resolution parameters;

[0020] A hierarchical segmentation framework is used to perform hierarchical segmentation on the multi-source data, generating multiple sub-regions;

[0021] Extract the corresponding spatial resolution parameters and temporal frequency parameters for the sub-region;

[0022] If the difference in spatial resolution parameters of a sub-region exceeds a preset threshold, the sub-region boundaries will be adjusted and the sub-regions will be re-segmented.

[0023] The spatial resolution parameters of the registered sub-regions are aligned using a registration transformation to obtain the registered sub-regions.

[0024] The registration sub-regions are sorted according to the time frequency parameters to determine the fusion order;

[0025] The registration sub-region is processed using the nearest neighbor interpolation algorithm to generate a fused dataset.

[0026] Preferably, the process of constructing a multidimensional index structure and retrieving historical perturbation patterns to obtain similar pattern matching results includes:

[0027] By performing preliminary processing on the sub-region fusion dataset, topological relationship features are extracted, and a pre-established classification model is used to perform hierarchical labeling of the topological relationships to obtain a structured representation of the topological relationships.

[0028] Based on the structured representation of the topological relationship, association analysis is performed in conjunction with attribute rules to obtain the corresponding mapping between attribute rules and topological relationships, and to determine the distribution of attribute rules in different topological levels.

[0029] Based on the distribution of the attribute rules, a multidimensional index structure is constructed, and the topological relationships and attribute rules are embedded in the index structure to obtain a multidimensional index framework.

[0030] The multidimensional indexing framework is used to retrieve historical disturbance data, and the support vector machine algorithm is used to classify the historical disturbance data to determine the disturbance category most relevant to the current dataset.

[0031] If the topological relationship matching degree between the retrieved historical perturbation category and the current dataset is higher than the preset threshold, it is marked as a candidate pattern, and a preliminary pattern matching set is obtained.

[0032] Based on the preliminary pattern matching set and the filtering conditions of similar results, the candidate patterns are compared a second time to obtain the final pattern matching results and determine the historical perturbation pattern that is closest to the current dataset.

[0033] By structuring and storing the final pattern matching results and associating them with the regional analysis results of the fused dataset, a complete data record is obtained that can be queried subsequently.

[0034] Preferably, the process of dynamically planning the optimized movement trajectory scheme for the sky platform path and dwell time includes:

[0035] Historical data of exploration activities are obtained, and similarity pattern matching analysis is performed on the historical data to determine the preliminary patterns of distribution characteristics and obtain the classification results of distribution characteristics;

[0036] Based on the classification results of the distribution characteristics, a probability assessment method is used to calculate the disturbance probability, and a preset threshold is used for screening to determine the distribution location of high-risk areas.

[0037] If the distribution of high-risk areas exceeds the preset threshold range, the priority ranking of disturbance probabilities is generated based on the prediction results to divide the key monitoring areas.

[0038] Based on the real-time location data of the sky platform obtained from the key monitoring areas, a preliminary plan for the movement path is calculated using dynamic programming methods to determine the initial framework for path planning.

[0039] Based on the initial framework of the path planning, the allocation requirements for dwell time are analyzed, and adjustments are made using time window constraints to obtain an optimized configuration of dwell time.

[0040] By optimizing the dwell time configuration and combining iterative calculations with the preliminary movement path plan, a final optimized trajectory plan is generated, and the coverage integrity of the trajectory plan is determined.

[0041] Based on the coverage integrity of the trajectory scheme, the operating parameters of the sky platform are obtained, the execution strategy of dynamically adjusting the movement path and dwell time is adjusted, and the final operation plan is determined.

[0042] Preferably, the process of adjusting the path through a multi-platform collaborative mechanism to obtain the supplementary observation coverage sequence includes:

[0043] Acquire optimized trajectory data and real-time weather data, and overlay potential fault records to form a comprehensive risk dataset;

[0044] Identify weather disruption factors and potential fault locations from the comprehensive risk dataset, and mark high-risk trajectory segments to obtain risk-marked trajectories;

[0045] Based on the risk-marked trajectory, query the available resources for multi-platform collaboration, determine the location and status of schedulable platforms, and obtain a list of collaborative resources.

[0046] The risk-marked trajectory is adjusted by using the collaborative resource list to bypass high-risk trajectory segments and obtain the adjusted trajectory set.

[0047] The adjusted trajectory set is assigned to the collaborative resource list using the nearest neighbor matching algorithm to generate a preliminary supplementary observation allocation scheme.

[0048] If the initial supplementary observation allocation scheme has observation coverage gaps, a secondary path adjustment is triggered to fill the gap areas and obtain a complete supplementary trajectory.

[0049] The complete complement trajectory serialization process generates a complement observation coverage sequence.

[0050] Preferably, the process of obtaining the compensated perturbation signal dataset by applying spatiotemporal interpolation to estimate missing information includes:

[0051] Obtain the supplementary observation coverage sequence and the main disturbance signal dataset;

[0052] The signal values ​​at corresponding positions in the supplementary observation coverage sequence are compared point by point to determine whether there is a deviation. If the deviation exceeds a preset threshold, the corresponding position is marked as a potential spatiotemporal gap.

[0053] Based on the location of the potential spatiotemporal vulnerability, extract the effective signal values ​​of the surrounding neighboring points to form a local spatiotemporal neighborhood window;

[0054] Using the effective signal values ​​within the local spatiotemporal neighborhood window, the Kriging interpolation algorithm is used to calculate the estimated location of the vulnerability, thus obtaining a preliminary filling signal.

[0055] Spatial gradient and temporal continuity indices are obtained from the initial filling signal and the neighborhood window signal. It is determined whether the filling value is consistent with the neighborhood. If they are inconsistent, the value is adjusted to the neighborhood weighted average to determine the corrected filling signal.

[0056] The corrected filling signal is used to replace the spatiotemporal gap position in the original dataset to obtain the compensated perturbation signal dataset.

[0057] By performing point-by-point difference calculations between the compensated disturbance signal dataset and the original coverage sequence, it is determined whether the remaining deviations are all within a preset threshold, and the final compensated disturbance signal dataset is determined.

[0058] Preferably, the process of quantifying the looseness index to determine the clustering pattern and obtain the clustering analysis output includes:

[0059] The original data points are extracted from the compensated disturbance signal dataset to obtain the spatial location information of the disturbance points and determine the preliminary distribution range.

[0060] Based on the preliminary distribution range, a grid division method is used to map the disturbance points to the corresponding grid cells, and the number of points in each cell is counted.

[0061] Based on the number of points in each grid cell, the local density value is calculated. If the local density value exceeds the preset threshold, it is marked as a high-density area to identify potential clustering areas.

[0062] The spatial coordinates of the disturbance points are extracted from the high-density region, the distance relationship between the points is analyzed, and the specific value of the degree of dispersion is obtained.

[0063] Based on the specific numerical value of the degree of dispersion, combined with the local density value, the looseness index is calculated to determine the overall distribution characteristics of the disturbance points.

[0064] The clustering patterns are divided by comparing the looseness index with the preset pattern classification criteria and the K-means clustering algorithm is used to obtain the final clustering analysis results.

[0065] Based on the clustering analysis results, corresponding pattern classification labels are generated, and a description of the clustering pattern of the perturbation signal is output.

[0066] Preferably, the process of analyzing the propagation direction to obtain a description of the disturbance dynamics includes:

[0067] Acquire satellite differential interferometry data and ground tiltmeter data, and perform spatiotemporal alignment processing to obtain a synchronized dataset in a unified coordinate system;

[0068] The K-means clustering algorithm was used to cluster the synchronous dataset, resulting in multiple cluster centers and intra-cluster point distributions.

[0069] Calculate the variance of the distance from each point within a cluster to the cluster center to determine the looseness;

[0070] If the looseness is higher than a preset threshold, the vector field construction process is initiated;

[0071] By fitting the displacement gradient in the synchronous dataset using the least squares method, a continuous vector field model is constructed to obtain the vector value of each grid point.

[0072] By analyzing the consistency of vector directions between adjacent grid points in the vector field model, the direction of disturbance propagation is determined, and the main propagation path is obtained.

[0073] The disturbance dynamics features are extracted based on the vector amplitude changes along the main propagation path to obtain a feature description sequence.

[0074] Preferably, the process of obtaining the early warning trigger signal includes:

[0075] By collecting disturbance dynamics data in the environment and using sensor networks to obtain real-time dynamic change information, a preliminary set of dynamic characteristics is determined.

[0076] Based on the preliminary set of dynamic features, feature extraction techniques are used to reduce the dimensionality of the data, resulting in a simplified set of feature vectors.

[0077] The simplified feature vector group is compared one by one with the preset danger feature library. If the comparison result exceeds the preset threshold range, it is determined to be a potential precursor signal.

[0078] The potential precursor signals are combined and analyzed to obtain the correlation patterns between the signals and to determine whether there are signal combinations that meet the conditions for a dangerous situation.

[0079] If a combination of signals that meets the conditions for a dangerous situation is detected, the early warning triggering mechanism is activated, and a corresponding early warning signal is output.

[0080] Based on the output of the warning signal, relevant systems are linked to transmit information and determine the final warning response level;

[0081] The corresponding emergency response procedures are obtained by classifying and processing the aforementioned early warning response levels.

[0082] Preferably, the process of determining the type and development trend of the hazard and outputting corresponding emergency response decisions includes:

[0083] Upon receiving an early warning trigger signal, the system initiates the collection of encrypted observation data.

[0084] Acquire encrypted observation data, decrypt it to form a real-time observation dataset;

[0085] The random forest algorithm was used to perform multi-source data fusion on the real-time observation dataset to obtain the fusion analysis results;

[0086] Based on the fusion analysis results, multidimensional indicators are extracted to determine the type of danger.

[0087] If the type of danger is sudden, then the trend curve is obtained by using time series forecasting method to analyze the development trend;

[0088] The rate of change is calculated using the trend curve to determine the level of the development trend;

[0089] Based on the type of hazard and the level of development trend, pre-established response rules are matched to output emergency response decision support.

[0090] Compared with the prior art, the present invention has the following advantages and technical effects:

[0091] This invention discloses a disturbance monitoring and early warning technology based on multi-source data fusion. Addressing the challenge of accurately capturing and predicting the spatiotemporal distribution characteristics of disturbance signals in complex environments, it proposes a systematic solution. The solution acquires satellite imagery and UAV grid data, extracts spatial resolution and temporal frequency parameters, processes and fuses them into a registered sub-region dataset using a hierarchical segmentation framework, analyzes topological relationships and attribute rules, constructs a multi-dimensional index structure to retrieve historical disturbance patterns, assesses the distribution characteristics and probability of exploration activities, and dynamically plans the path of the aerial platform to optimize its movement trajectory. Simultaneously, this invention identifies potential faults and weather disruptions, adjusts paths using a multi-platform collaborative mechanism, applies spatiotemporal interpolation algorithms to compensate for missing information, quantifies the dispersion and distribution density of disturbance points, determines clustering patterns, and constructs a vector field model to analyze propagation direction if the dispersion is high. Finally, it generates early warning trigger signals by comparing with a hazard feature database and provides support for emergency response decision-making. This invention significantly improves the accuracy of disturbance monitoring and the timeliness of early warning, providing strong technical support for risk prevention and control in complex environments. Attached Figure Description

[0092] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0093] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0094] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0095] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0096] like Figure 1 As shown, this embodiment provides a method for identifying exploration disturbances based on sky-ground collaboration, including:

[0097] Acquire multi-source data from satellites and drones, extract spatial resolution parameters and temporal frequency parameters from the multi-source data, and process the parameters using a hierarchical segmentation method to obtain a registered sub-region fusion dataset;

[0098] Based on the registered sub-region fusion dataset, the topological relationships and attribute association rules are analyzed, a multi-dimensional index structure is constructed, and historical perturbation patterns are retrieved to obtain similar pattern matching results.

[0099] Based on the similarity pattern matching results, the distribution characteristics of exploration activities and the probability of disturbance are evaluated, and the optimized movement trajectory scheme of the sky platform path and dwell time is obtained by dynamic planning.

[0100] Based on the optimized trajectory scheme, potential faults and weather-related disruptions are identified, and the path is adjusted through a multi-platform collaborative mechanism to obtain a supplementary observation coverage sequence.

[0101] Based on the supplementary observation coverage sequence, spatiotemporal gaps caused by cloud cover and sensor failure are detected, and spatiotemporal interpolation is applied to estimate the missing information to obtain a compensated disturbance signal dataset.

[0102] Based on the compensated disturbance signal dataset, the dispersion and distribution density of disturbance points are calculated, and the looseness index is quantified to determine the clustering pattern and obtain the clustering analysis output.

[0103] If the cluster analysis output shows high looseness, then a vector field model is constructed by fusing satellite differential interferometry and ground tiltmeter data, and the propagation direction is analyzed to obtain a description of the disturbance dynamics.

[0104] By comparing the disturbance dynamics characteristics with a pre-set hazard feature database, the occurrence of precursor signal combinations is determined to obtain early warning trigger signals;

[0105] Based on the early warning trigger signal, the system schedules encrypted observation and real-time fusion analysis to determine the type and development trend of the hazard and output corresponding emergency response decisions.

[0106] Furthermore, the process of processing parameters using a hierarchical segmentation method to obtain the registered sub-region fusion dataset includes:

[0107] Acquire multi-source data composed of satellite imagery and UAV grids;

[0108] Spatial resolution parameters and temporal frequency parameters are extracted from multi-source data;

[0109] Determine the hierarchical segmentation levels based on spatial resolution parameters;

[0110] A hierarchical segmentation framework is used to perform hierarchical segmentation of multi-source data, generating multiple sub-regions;

[0111] Extract the corresponding spatial resolution parameters and temporal frequency parameters from the sub-regions;

[0112] If the difference in spatial resolution parameters of a sub-region exceeds a preset threshold, the sub-region boundaries will be adjusted and the sub-regions will be re-segmented.

[0113] The spatial resolution parameters of the registered sub-regions are aligned using a registration transformation to obtain the registered sub-regions.

[0114] The registration sub-regions are sorted according to the time frequency parameters to determine the fusion order;

[0115] The nearest neighbor interpolation algorithm is used to process the registration sub-regions and generate a fused dataset.

[0116] Furthermore, in this embodiment, when acquiring multi-source data composed of satellite imagery and UAV grid data, the satellite imagery comes from high-resolution remote sensing satellites, providing a spatial resolution of 0.5 meters per square meter, with a temporal frequency of once a week; while the UAV grid data has a higher resolution, such as 0.1 meters per square meter, but a lower temporal frequency, once a month. After extracting the spatial resolution parameters and temporal frequency parameters of these data, it was found that the spatial resolution of the two is significantly different, requiring further processing.

[0117] In one possible implementation, when determining the hierarchical segmentation levels based on spatial resolution parameters, satellite imagery is used as the base level for large-scale coverage, while UAV data serves as the high-precision level for supplementing local details. After hierarchically segmenting the multi-source data using the hierarchical segmentation framework, multiple sub-regions are generated; for example, a region might be divided into 100 sub-regions, each 1 square kilometer in size. After extracting the corresponding spatial resolution and temporal frequency parameters for each sub-region, if it is found that the resolution of satellite imagery within a sub-region is 0.5 meters, while that of UAV data is 0.1 meters, exceeding a preset threshold such as 0.3 meters, the sub-region boundaries are adjusted, and the region is re-segmented into smaller areas, such as 0.5 square kilometers, to reduce the resolution difference.

[0118] For example, when aligning the spatial resolution parameters of the registration transformation sub-regions, this embodiment uses a geometric correction method to downsample the high-resolution UAV imagery to a resolution of 0.5 meters, consistent with the satellite imagery, to obtain the registration sub-regions. This step ensures spatial consistency of the data during subsequent fusion, avoiding the loss of details due to resolution differences. The registration sub-regions are sorted according to time-frequency parameters. When determining the fusion order, satellite imagery data with higher time frequencies is processed first, followed by the UAV data, ensuring information integrity in the time dimension.

[0119] In one possible implementation, this embodiment uses a nearest neighbor interpolation algorithm to process the registration sub-region, finding the nearest known value for each pixel to fill in the gaps and generate a fused dataset. This method is computationally simple and effectively preserves the features of the original data, making it particularly suitable for scenarios with clear boundaries in green space monitoring. The fused dataset ultimately exhibits the advantages of both the wide coverage of satellite imagery and the high precision of UAV data.

[0120] For example, the technical benefits of the above methods include improved data consistency, ensuring the coordination of multi-source data in spatial and temporal dimensions, thereby enhancing the accuracy and efficiency of green space monitoring. Through hierarchical segmentation and registration transformation, the impact of data differences on analysis results can be effectively reduced; through time-frequency sorting and interpolation fusion, more comprehensive dynamic monitoring data can be generated, providing a reliable basis for land planning and ecological protection.

[0121] Furthermore, the process of constructing a multidimensional index structure and retrieving historical perturbation patterns to obtain similar pattern matching results includes:

[0122] By performing preliminary processing on the sub-region fusion dataset, topological relationship features are extracted, and a pre-established classification model is used to perform hierarchical labeling of the topological relationships to obtain a structured representation of the topological relationships.

[0123] For the structured representation of topological relationships, association analysis is performed in conjunction with attribute rules to obtain the corresponding mapping between attribute rules and topological relationships, and to determine the distribution of attribute rules in different topological levels;

[0124] Based on the distribution of attribute rules, a multidimensional index structure is constructed, and topological relationships and attribute rules are embedded into the index structure to obtain a multidimensional index framework.

[0125] Historical perturbation data is retrieved using a multidimensional indexing framework, and the support vector machine algorithm is used to classify the historical perturbation data to determine the perturbation category most relevant to the current dataset.

[0126] If the topological relationship matching degree between the retrieved historical perturbation category and the current dataset is higher than the preset threshold, it is marked as a candidate pattern, and a preliminary pattern matching set is obtained.

[0127] Based on the preliminary pattern matching set and the screening criteria of similar results, the candidate patterns are compared a second time to obtain the final pattern matching results and determine the historical perturbation pattern that is closest to the current dataset.

[0128] By storing the final pattern matching results in a structured manner and associating them with the regional analysis results of the fused dataset, a complete data record is obtained that can be queried subsequently.

[0129] Furthermore, in the processing of sub-region datasets after the fusion of remote sensing multi-source data, this embodiment first needs to extract topological relationship features.

[0130] Specifically, a topological relationship feature map is formed by analyzing the boundary connections, inclusions, or separations between registered sub-regions. This feature reflects the spatial connectivity of land features, such as the adjacency between farmland plots and roads.

[0131] One possible implementation involves using a pre-established classification model to hierarchically label topological relationships. This model is trained on a graph neural network; after inputting topological relationship features, it outputs multi-level labels, such as dividing the topology into first-level global connections, second-level local clusters, and third-level fine-grained nodes, thus obtaining a structured topological representation. This hierarchical labeling facilitates efficient subsequent association analysis.

[0132] For example, association analysis can be performed on structured topological representations by incorporating attribute rules. Attribute rules include land cover types such as vegetation indices or soil moisture thresholds. By statistically analyzing the distribution of attributes at different topological levels, it was found that areas with high vegetation indices are concentrated in the secondary cluster layer, forming a corresponding mapping and revealing the hierarchical distribution of attribute rules. This analysis can highlight the attribute characteristics of disturbance-sensitive areas.

[0133] One possible implementation involves constructing a multidimensional index structure based on the distribution, such as a hybrid index combining R-trees and KD-trees, embedding topological hierarchy and attribute rules to form a complete multidimensional index framework. This framework supports high-dimensional, fast queries, significantly improving retrieval efficiency.

[0134] Specifically, this embodiment uses a multidimensional indexing framework to quickly retrieve historical disturbance data.

[0135] For example, given the topological and attribute features of the current dataset, the index can return hundreds of historical records within seconds. These records are then classified using a support vector machine algorithm to determine the most relevant perturbation category, such as the type of vegetation degradation caused by drought. This classification utilizes a hyperplane to separate historical samples, ensuring accurate correlation assessment.

[0136] For example, if the matching degree between the retrieved historical disturbance category and the current topological relationship is higher than the 0.85 threshold, it is marked as a candidate pattern, forming a preliminary pattern matching set. This step filters out highly similar historical cases and avoids interference from irrelevant data.

[0137] In one possible implementation, a second comparison is performed based on the initial set, combined with similarity filtering conditions such as cosine similarity greater than 0.9, to obtain the final pattern matching result.

[0138] For example, the historical flood disturbance pattern that is closest to the current fused dataset was finally identified, which shows a 20% decrease in vegetation cover under similar topology.

[0139] Specifically, by structuring and storing the final pattern matching results and correlating them with regional analysis results from the fused dataset, a complete data record is obtained. This record facilitates subsequent querying and retrieval, supports disturbance prediction and change monitoring, significantly improves the application value of remote sensing data in environmental disturbance analysis, and enhances the timeliness and accuracy of decision-making.

[0140] Furthermore, the process of dynamically planning the optimized movement trajectory scheme for the sky platform path and dwell time includes:

[0141] Obtain historical data of exploration activities, perform similarity pattern matching analysis on the historical data, determine the preliminary patterns of distribution characteristics, and obtain the classification results of distribution characteristics;

[0142] Based on the classification results of distribution characteristics, the probability assessment method is used to calculate the disturbance probability, and combined with the preset threshold for screening, the distribution location of high-risk areas is determined.

[0143] If the distribution of high-risk areas exceeds the preset threshold range, the priority ranking of disturbance probabilities is generated based on the prediction results to divide the key monitoring areas.

[0144] Based on the real-time location data of the Sky Platform obtained from the key monitoring areas, a preliminary plan for the movement path is calculated using dynamic programming methods to determine the initial framework for path planning.

[0145] Based on the initial framework of path planning, the allocation requirements for dwell time are analyzed, and adjustments are made using time window constraints to obtain an optimized configuration of dwell time.

[0146] By optimizing the configuration of dwell time and combining iterative calculations with the preliminary plan of the movement path, the final optimized trajectory plan is generated, and the coverage integrity of the trajectory plan is judged.

[0147] Based on the coverage integrity of the trajectory scheme, obtain the operating parameters of the sky platform, dynamically adjust the execution strategy of movement path and dwell time, and determine the final operation plan.

[0148] Furthermore, in this embodiment, after obtaining historical data of exploration activities, a similar pattern matching analysis is first performed on these data.

[0149] Understandably, by comparing the distribution patterns of disturbances observed in historical surveys with current data characteristics, similar spatial clustering or diffusion patterns can be identified, thereby determining the preliminary patterns of distribution characteristics and obtaining classification results. This classification helps to quickly categorize disturbance types, such as concentrated or striped distributions.

[0150] Specifically, based on the classification results of distribution characteristics, this embodiment uses a probability assessment method to calculate the disturbance probability. For example, the frequency of disturbances in historical similar patterns is converted into probability values. If the disturbance occurrence rate in a certain area reaches 70% in similar patterns, the current probability assessment is 0.7. This is then combined with a pre-established threshold, such as 0.5, to filter and determine the distribution location of high-risk areas. This step effectively highlights potential threat areas and improves risk identification efficiency.

[0151] In one embodiment, if the distribution of high-risk areas exceeds a preset threshold range, such as a probability exceeding 0.8, a priority ranking of disturbance probabilities is generated based on the prediction results. For example, the probabilities are arranged from high to low as 0.9, 0.85, 0.7, etc., to obtain the division of key monitoring areas. This ranking is beneficial for centralized resource allocation and avoids the inefficiency caused by comprehensive coverage.

[0152] For example, in this embodiment, based on the division of key monitoring areas, real-time location data of the Sky Platform is obtained, and a preliminary plan for the movement path is calculated using dynamic programming methods to determine the initial framework of the path planning. Here, dynamic programming can consider distance and coverage requirements to generate sequential paths from the starting point to each high-risk point, ensuring that the preliminary plan covers the main areas.

[0153] Specifically, this embodiment analyzes the allocation requirements of dwell time based on the initial framework of path planning and adjusts it using time window constraints. For example, the initial allocation is 30 minutes of dwell time per point, but if the time window only allows a total duration of 4 hours, it is preferable to adjust the allocation to 45 minutes for high-risk points and 20 minutes for low-risk points, thus obtaining an optimized configuration of dwell time. This optimization can improve the quality of data collection.

[0154] In one possible implementation, this embodiment iteratively calculates the path by optimizing the dwell time configuration and combining it with a preliminary path plan. For example, it repeatedly adjusts the path order and dwell time until the total coverage reaches 95%, generating a final optimized trajectory plan and determining the coverage integrity of the trajectory plan. This is beneficial for achieving efficient inspection and reducing omissions.

[0155] For example, based on the coverage integrity of the trajectory plan, the operational parameters of the sky platform, such as speed and battery life, are obtained. The execution strategy for the movement path and dwell time is dynamically adjusted; for instance, if battery life is insufficient, the dwell time at secondary points is shortened, thus determining the final operational plan. This dynamic adjustment ensures the feasibility of the plan and improves the overall exploration response speed and accuracy.

[0156] Furthermore, the process of obtaining the supplementary observation coverage sequence by adjusting the path through a multi-platform collaborative mechanism includes:

[0157] Acquire optimized trajectory data and real-time weather data, and overlay potential fault records to form a comprehensive risk dataset;

[0158] Identify weather disruption factors and potential fault locations from the comprehensive risk dataset, and mark high-risk trajectory segments to obtain risk-marked trajectories;

[0159] Based on the risk-marked trajectory, query the available resources for multi-platform collaboration, determine the location and status of schedulable platforms, and obtain a list of collaborative resources.

[0160] By adjusting the path of the risk-marked trajectory through the collaborative resource list, the set of adjusted trajectories is obtained by bypassing high-risk trajectory segments.

[0161] The nearest neighbor matching algorithm is used to allocate the adjusted trajectory set to the collaborative resource list, generating a preliminary supplementary observation allocation scheme;

[0162] If there are gaps in the initial supplementary observation allocation plan, a secondary path adjustment will be triggered to fill the gaps and obtain a complete supplementary trajectory.

[0163] The supplementary observation coverage sequence is generated by complete supplementary trajectory serialization.

[0164] Furthermore, in the field of long-term exploration missions performed by the sky platform, this embodiment acquires optimized trajectory data and real-time weather data, and then overlays potential fault records to form a comprehensive risk dataset. This process helps to comprehensively capture multi-source risk factors.

[0165] Specifically, this embodiment uses the optimized trajectory obtained from the preceding dynamic programming as a basis, and incorporates real-time meteorological information such as wind speed and rainfall intensity, as well as historical fault records such as sensor failure locations, to form a multi-dimensional dataset, thereby improving the accuracy of risk identification.

[0166] In one possible implementation, this embodiment identifies weather disruption factors and potential fault locations from a comprehensive risk dataset and marks high-risk trajectory segments by threshold comparison.

[0167] For example, when the wind speed exceeds 15 m / s or the density of fault records is higher than 3 per 10 km, the corresponding trajectory segment is marked as high risk. This not only avoids the bias of judgment based on a single factor, but also highlights complex risk areas and significantly reduces the probability of platform outage.

[0168] For example, this embodiment queries the available resources for multi-platform collaboration based on the risk-marked trajectory, determines the location and status of schedulable platforms, and obtains a list of collaborative resources. In actual operation, if there is a high-risk segment on the main platform's trajectory, the location and energy status of nearby idle or low-load platforms can be quickly queried to form an available list. This collaborative mechanism helps to achieve efficient resource allocation and improve the overall task continuity.

[0169] It should be noted that in this embodiment, the risk-marked trajectory is adjusted by using a collaborative resource list to bypass high-risk trajectory segments and obtain the adjusted trajectory set.

[0170] Specifically, alternative routes with a deviation angle of less than 30 degrees are prioritized to bypass areas with severe weather, while avoiding fault-prone areas, generating multiple sets of alternative trajectories. This adjustment helps maintain exploration coverage while reducing exposure risk.

[0171] For example, in this embodiment, the nearest neighbor matching algorithm is used to allocate the adjusted trajectory set to the collaborative resource list, generating a preliminary supplementary observation allocation scheme.

[0172] In one embodiment, nearest neighbor matching is performed based on the Euclidean distance between the platform's current location and the risk segment, and the short-distance high-risk segment is assigned to a nearby backup platform to fill the gap, thereby forming a preliminary plan. This allocation method can quickly respond to sudden risks and enhance the system's robustness.

[0173] In one possible implementation, if the initial supplementary observation allocation scheme has observation coverage gaps, such as missing some key monitoring areas, a secondary path adjustment is triggered to fill the gap areas and obtain a complete supplementary trajectory.

[0174] For example, by expanding the search radius and replanning local paths to fill gaps, ensuring coverage of over 95%, this iterative supplement helps to achieve blind-spot-free observation.

[0175] For example, a supplementary observation coverage sequence is generated by serializing the complete supplementary trajectory, and multiple platform trajectories are concatenated in chronological order to form a continuous coverage sequence.

[0176] Specifically, the sequence is first ordered by start time, and then overlapping segments are merged. This sequential approach helps coordinate the action order of multiple platforms, improving the prediction of disturbances and the overall response efficiency of exploration activities. Based on the historical high-risk area delineation, this supplementary mechanism further strengthens the dynamic risk response capability, ensuring that the sky platform cluster maintains a stable exploration distribution in complex environments.

[0177] Furthermore, the process of applying spatiotemporal interpolation to estimate the missing information and obtain the compensated perturbation signal dataset includes:

[0178] Obtain the supplementary observation coverage sequence and the main disturbance signal dataset;

[0179] The signal values ​​at corresponding locations in the supplementary observation coverage sequence are compared point by point to determine whether there is a deviation. If the deviation exceeds a preset threshold, the corresponding location is marked as a potential spatiotemporal gap.

[0180] Based on the location of potential spatiotemporal vulnerabilities, extract valid signal values ​​from surrounding neighboring points to form a local spatiotemporal neighborhood window;

[0181] By using the effective signal values ​​within a local spatiotemporal neighborhood window, the Kriging interpolation algorithm is used to calculate the estimated location of the vulnerability, thus obtaining a preliminary filling signal.

[0182] Spatial gradient and temporal continuity indices are obtained from the initial filling signal and the neighborhood window signal to determine whether the filling value is consistent with the neighborhood. If they are inconsistent, the value is adjusted to the neighborhood weighted average to determine the corrected filling signal.

[0183] The corrected and filled signal will replace the spatiotemporal gap locations in the original dataset to obtain a compensated perturbation signal dataset.

[0184] By performing point-by-point difference calculations between the compensated disturbance signal dataset and the original coverage sequence, it is determined whether the remaining deviations are all within the preset threshold, and the final compensated disturbance signal dataset is determined.

[0185] Furthermore, in this embodiment, when comparing the supplementary observation coverage sequence with the main disturbance signal dataset, the significance of this process can be understood conceptually first. The supplementary observation coverage sequence is used to fill gaps or interruptions in the observations, while the main disturbance signal dataset records the original signal fluctuations. By comparing the signal values ​​of the two point by point, potential spatiotemporal gaps can be discovered, i.e., locations with large signal deviations. Such deviations may be caused by observation interruptions or missing data, and identifying these locations helps in subsequent signal compensation work.

[0186] Specifically, in this embodiment, when determining significant deviations, a preset threshold of 5 units is assumed. If the signal value at a certain location is 10 in the coverage sequence but 16 in the main disturbance signal dataset, the deviation is 6, exceeding the threshold, and therefore it is marked as a potential spatiotemporal vulnerability. This marking method can help accurately locate the problem area, providing a basis for subsequent patching. After marking, the signal values ​​of neighboring points around the vulnerability are extracted to form a local spatiotemporal neighborhood window. For example, the signal values ​​of 3 points before and after the vulnerability point are taken to form a small window containing 7 points for further analysis.

[0187] For example, in this embodiment, when using the Kriging interpolation algorithm to calculate the estimated location of the vulnerability, it can be understood as a prediction method based on spatial correlation. Assuming the signal values ​​within the neighborhood window are distributed as 8, 9, 11, missing, 12, 10, 9, by analyzing the spatial relationships of these points, the value at the missing location is estimated to be close to 11, forming a preliminary filling signal. The advantage of this method is that it fully utilizes the regularity of the surrounding data, improving the rationality of the filling.

[0188] Specifically, when determining the spatial gradient and temporal continuity between the initial fill signal and the neighboring window signal, if the estimated value 11 is found to be inconsistent with the neighboring trend—for example, if the neighboring signal shows a downward trend while 11 is too high—it is adjusted to the neighborhood weighted average, such as 10.5, as the corrected fill signal. This adjustment ensures the smoothness of the signal and avoids subsequent analysis errors caused by abrupt changes.

[0189] For example, in this embodiment, after replacing the spatiotemporal vulnerability location and obtaining the compensated disturbance signal dataset, the effect can be verified by calculating the point-by-point difference. Assuming the value at a certain point in the original coverage sequence is 10, and the value in the compensated dataset is 10.2, the deviation is 0.2, far below the threshold of 5, indicating a good compensation effect. This verification method ensures the reliability of the final signal dataset, providing more complete data support for subsequent business analysis.

[0190] Specifically, the generation of the final compensation disturbance signal dataset not only fills in the data gaps but also ensures the continuity and consistency of the signals. This complete dataset can provide more accurate references for path adjustment and resource scheduling in the fields of trajectory optimization and risk identification, reducing decision-making errors caused by data gaps and improving efficiency.

[0191] Furthermore, the process of quantifying the looseness index to determine the clustering pattern and obtain the clustering analysis output includes:

[0192] The original data points are extracted from the compensated disturbance signal dataset to obtain the spatial location information of the disturbance points and determine the preliminary distribution range.

[0193] Based on the initial distribution range, a grid division method is used to map the disturbance points to the corresponding grid cells, and the number of points in each cell is obtained.

[0194] Based on the number of points in each grid cell, the local density value is calculated. If the local density value exceeds the preset threshold, it is marked as a high-density area to identify potential clustering areas.

[0195] Extract the spatial coordinates of disturbance points from high-density areas, analyze the distance relationship between points, and obtain specific values ​​of the degree of dispersion.

[0196] Based on the specific numerical value of the degree of dispersion and combined with the local density value, the looseness index is calculated to determine the overall distribution characteristics of the disturbance points.

[0197] By comparing the looseness index with the preset pattern classification criteria, the K-means clustering algorithm is used to divide the clustering patterns and obtain the final clustering analysis results.

[0198] Based on the clustering analysis results, corresponding pattern classification labels are generated, and a description of the clustering pattern of the perturbation signal is output.

[0199] Furthermore, in this embodiment, when processing the compensated disturbance signal dataset, the spatial location information of the original data points is extracted to initially determine the distribution range. Assuming that within a monitoring area, the location data of the disturbance points are recorded in two-dimensional coordinates, covering an area of ​​1000 meters by 1000 meters, analyzing these coordinates can preliminarily define the main distribution range of the disturbance points, providing a basis for subsequent grid division.

[0200] For example, in this embodiment, the grid division method divides the entire monitoring area into 10-meter by 10-meter grid cells, and the number of disturbance points within each cell is counted. Assuming a cell contains 15 disturbance points while adjacent cells contain only 3, this difference in quantity provides data support for subsequent density analysis. The fineness of the grid division can be adjusted according to actual monitoring needs; the finer the division, the higher the analysis accuracy.

[0201] For example, in calculating local density values ​​and determining thresholds, this embodiment can set a density threshold of 10 points per square meter. If the density value of a certain grid cell is 12, it is marked as a high-density area. This marking method helps to quickly locate potential clustering areas, providing a focus for subsequent analysis.

[0202] It should be noted that the threshold setting can be dynamically adjusted based on historical data or experience values ​​to adapt to the needs of different scenarios.

[0203] For example, when analyzing the spatial coordinates of perturbation points within a high-density region, the degree of dispersion can be assessed by calculating the average distance between points. Suppose that in one high-density region the average distance between points is 2 meters, while in another region it is 5 meters; this difference reflects the density of the distribution. Such analysis helps to further refine the understanding of the distribution characteristics of perturbation points.

[0204] For example, when calculating the looseness index, local density values ​​and dispersion values ​​can be combined for a comprehensive evaluation. If the density value is high but the distance between points is large, the looseness index may indicate that the distribution in that area is relatively loose. The introduction of this index provides a quantitative basis for subsequent pattern classification, helping to more accurately describe distribution characteristics.

[0205] For example, when using the K-means clustering algorithm to classify clustering patterns, perturbation points can be divided into several classes based on their spatial location and looseness index. Suppose that the final classification results in three classes, representing tight clustering, loose clustering, and random distribution, respectively. This classification method can intuitively reflect the spatial clustering patterns of perturbation points, providing a clear pattern reference for subsequent analysis.

[0206] For example, when generating pattern classification labels and outputting clustering pattern descriptions, each clustering pattern can be labeled with a specific tag, such as "high-density compact" or "low-density dispersed." If a certain type of disturbance point is labeled "high-density compact," it indicates that this area may require focused attention. This labeled output method facilitates rapid understanding and application of analysis results, contributing to improved monitoring efficiency and decision support capabilities.

[0207] Furthermore, the process of analyzing the propagation direction to obtain a description of the disturbance dynamics includes:

[0208] Acquire satellite differential interferometry data and ground tiltmeter data, and perform spatiotemporal alignment processing to obtain a synchronized dataset in a unified coordinate system;

[0209] The K-means clustering algorithm was used to cluster the synchronous dataset, resulting in multiple cluster centers and the distribution of points within each cluster;

[0210] Calculate the variance of the distance from each point within a cluster to the cluster center to determine the looseness;

[0211] If the looseness is higher than the preset threshold, the vector field construction process will be initiated.

[0212] By fitting the displacement gradient in the synchronous dataset using the least squares method, a continuous vector field model is constructed to obtain the vector value of each grid point.

[0213] By analyzing the consistency of vector directions between adjacent grid points in the vector field model, the direction of disturbance propagation is determined, and the main propagation path is obtained.

[0214] The perturbation dynamics features are extracted based on the vector amplitude changes along the main propagation path, resulting in a feature description sequence.

[0215] Furthermore, in this embodiment, during the acquisition of satellite differential interferometry data and ground tiltmeter data, surface deformation data is obtained through a satellite remote sensing platform, while ground tiltmeter data on local tilt angle changes are acquired. Assuming the satellite data covers an area of ​​100 square kilometers with a resolution of 5 meters, and the ground tiltmeter data is distributed at key monitoring points, approximately 200 meters apart, this data acquisition method can capture surface change characteristics at both macroscopic and microscopic levels.

[0216] For example, during spatiotemporal alignment, this embodiment matches the timestamps of satellite data with those of ground tiltmeter data to ensure that the two types of data are within the same time window. Simultaneously, geographic coordinate transformation unifies the data to the same coordinate system. Assuming the time alignment error is controlled within 1 second and the spatial alignment error within 0.5 meters, the accuracy of subsequent analysis can be guaranteed.

[0217] For example, when using the K-means clustering algorithm to cluster a synchronous dataset, this embodiment groups data points according to the characteristics of deformation amplitude and tilt angle. Assuming that five clusters are ultimately obtained, each cluster center represents a typical land surface change pattern, while the distribution of points within the cluster reflects the uniformity of change within the region. This method helps identify potential anomalous areas.

[0218] For example, when calculating the variance of the distance from each point within a cluster to the cluster center to determine the looseness, this embodiment sets a threshold; if the variance value is greater than 10, the looseness is considered high. Suppose the variance value of a certain cluster is 12, indicating that the data points in that area are relatively scattered and may have complex deformation characteristics, requiring further analysis.

[0219] For example, if the looseness exceeds a preset threshold, when initiating the vector field construction process, this embodiment constructs a 10-meter resolution grid model by meshing the displacement data in the synchronous dataset. The data for each grid point comes from the interpolation results of surrounding data points, thus forming a continuous vector field. This method can intuitively reflect the overall trend of surface deformation.

[0220] For example, when constructing a continuous vector field model by fitting displacement gradients using the least squares method, this embodiment assumes that the displacement gradient within a certain grid region exhibits a linear trend, and obtains the vector value for each grid point through fitting. This method can smooth data noise and improve the reliability of the model.

[0221] For example, when analyzing the consistency of vector directions between adjacent grid points in a vector field model, the angle difference between adjacent grid points is checked. If the angle difference is less than 15 degrees, the directions are considered consistent, thus determining the direction of disturbance propagation. Assuming that the vector directions of 10 consecutive grid points along a certain path are consistent, this path can be marked as the main propagation path. This analysis helps reveal the dynamic evolution patterns of surface changes.

[0222] For example, when extracting disturbance dynamics features based on vector amplitude changes along the main propagation path, by recording the vector amplitude at each grid point along the path, assuming the amplitude gradually increases from 2.5 to 5.0, it indicates that the disturbance intensity gradually increases along the path. This feature description sequence can provide important basis for subsequent geological risk assessment.

[0223] Furthermore, the process of obtaining the early warning trigger signal includes:

[0224] By collecting disturbance dynamics data in the environment and using sensor networks to obtain real-time dynamic change information, a preliminary set of dynamic characteristics is determined.

[0225] Based on the preliminary set of dynamic features, feature extraction techniques are used to reduce the dimensionality of the data, resulting in a simplified set of feature vectors.

[0226] The simplified feature vector group is compared one by one with the preset danger feature library. If the comparison result exceeds the preset threshold range, it is determined to be a potential precursor signal.

[0227] Combine and analyze potential precursor signals to obtain correlation patterns between signals and determine whether there are signal combinations that meet the conditions for a dangerous situation.

[0228] If a combination of signals that meets the conditions for a dangerous situation is detected, the early warning triggering mechanism is activated, and a corresponding early warning signal is output.

[0229] Based on the output of the early warning signal, relevant systems are coordinated to transmit information and determine the final early warning response level;

[0230] Classify and process the early warning response levels to obtain the corresponding emergency response procedures.

[0231] For example, this embodiment collects disturbance dynamics data in the environment, uses a sensor network to obtain real-time dynamic change information, and determines a preliminary set of dynamic features.

[0232] For example, deploying tiltmeters and GNSS sensor networks in crustal deformation monitoring areas allows these sensors to continuously record surface tilt angles and displacement rates, forming an initial set of features such as strain accumulation and deformation velocity. This approach helps capture minute crustal disturbances, avoiding the omission of early anomalous signals and thus improving the sensitivity of precursor identification. Based on the acquired dynamic feature set, feature extraction techniques are used to reduce the dimensionality of the data, resulting in a simplified set of feature vectors.

[0233] Specifically, in one embodiment, principal component analysis (PCA) is used to extract the main components governing the changes in deformation direction and amplitude from multidimensional data, for example, reducing the original displacement, tilt, and strain rate data to 5-10 key vectors. This dimensionality reduction significantly reduces the computational burden while retaining more than 90% of the information variation, effectively improving the efficiency and accuracy of subsequent comparisons. The simplified feature vector set is then compared one by one with a preset hazard feature database. If the comparison result exceeds a preset threshold range, it is determined to be a potential precursor signal.

[0234] For example, the hazard feature database contains typical vector patterns from historical earthquakes or landslide events, such as accelerated deformation characteristics in specific areas. When the similarity between the current vector and a pattern in the database is less than 0.8, it is identified as a potential precursor. This comparison mechanism can promptly screen for anomalies, reduce false alarms, and enhance the targeting of early warnings. By combining and analyzing potential precursor signals, correlation patterns between signals are obtained to determine whether there are signal combinations that meet the hazard conditions.

[0235] In one possible implementation, the deformation synchronization and propagation delay among multiple sensor points are analyzed. For example, if the tilt changes of adjacent points simultaneously exceed a threshold and exhibit directional correlation, a hazard combination pattern is confirmed. This combination analysis strengthens the mutual verification effect of multi-source data, greatly improves the reliability of the judgment, and avoids isolated misjudgments based on a single signal. If a signal combination that meets the hazard conditions is detected, the early warning triggering mechanism is activated, and a corresponding early warning signal is output.

[0236] Specifically, this embodiment can also automatically generate yellow or orange warning level signals and push them to the monitoring center via the network. This triggering mechanism ensures rapid response, buys valuable time for emergency preparedness, and significantly reduces disaster losses. Based on the warning signal output, relevant systems are linked to transmit information and determine the final warning response level.

[0237] For example, early warning signals can be linked to meteorological and emergency response platforms, and adjusted to a red level based on real-time rainfall data. This coordinated transmission improves the timeliness of information sharing and facilitates collaborative responses among multiple departments. By classifying and processing early warning response levels, corresponding emergency response procedures can be obtained, completing the closed-loop operation of the entire early warning process.

[0238] In one embodiment, the red level corresponds to immediate evacuation and engineering reinforcement procedures, while the yellow level indicates enhanced monitoring. This classification process forms a complete closed loop, ensuring efficient operation of the entire chain from monitoring to response, ultimately achieving effective utilization of disturbance dynamics characteristics and precise prevention and control of disaster risks.

[0239] Furthermore, the process of determining the type and development trend of a hazard and correspondingly outputting emergency response decisions includes:

[0240] Upon receiving an early warning trigger signal, the system initiates the collection of encrypted observation data.

[0241] Acquire encrypted observation data, decrypt it to form a real-time observation dataset;

[0242] The random forest algorithm was used to fuse multi-source data in the real-time observation dataset to obtain the fusion analysis results;

[0243] Based on the results of the fusion analysis, multidimensional indicators are extracted to determine the type of danger.

[0244] If the type of emergency is sudden, then the trend curve is obtained by using time series forecasting methods to analyze the development trend.

[0245] The rate of change is calculated using the trend curve to determine the level of the development trend;

[0246] Based on the type of hazard and the level of its development trend, pre-established response rules are matched to output emergency response decision support.

[0247] In one possible implementation, after receiving the warning trigger signal from the previous warning process, the system immediately starts the scheduling module to trigger the acquisition of encrypted observation data.

[0248] Specifically, the scheduling module sends acquisition instructions to the distributed sensor nodes, requiring the nodes to intensively collect multi-source environmental disturbance data, such as vibration, pressure, and displacement, within a short period of time, and to perform hardware-level encryption on the raw data during transmission to ensure data security during transmission.

[0249] For example, assuming the early warning trigger signal corresponds to a medium-risk level, the scheduling module preferably activates a high-frequency acquisition mode, where nodes collect encrypted data packets every 5 seconds for 10 minutes to form an encrypted observation dataset. This intensive acquisition method helps capture instantaneous changes in sudden disturbances and avoids missing key precursor details.

[0250] It should be noted that after acquiring the encrypted observation data, the system first performs batch decryption through a pre-configured key management center to obtain the complete real-time observation dataset. This dataset contains time-series data from multiple sensor channels, and the decryption process employs a symmetric encryption algorithm to ensure high efficiency and low latency.

[0251] In one embodiment, a random forest algorithm is applied to the real-time observation dataset for multi-source data fusion. The random forest constructs multiple decision trees to perform parallel voting fusion of data from different sensors, outputting a comprehensive fusion analysis result.

[0252] For example, in a disturbance event, the vibration sensor showed a sudden increase of 20% in amplitude, the pressure sensor showed a synchronous increase of 15%, and the displacement sensor showed a slight shift. After fusion using a random forest, the system yielded an anomaly confidence score of 0.85, significantly higher than the accuracy of single-sensor judgment. This fusion helps reduce noise interference and improves the ability to identify complex disturbance patterns. Based on the fusion analysis results, the system extracts multi-dimensional indicators, such as abnormal amplitude, spectral energy, and correlation coefficients, to determine the type of hazard.

[0253] For example, if the fusion results show that high-frequency components dominate and their amplitude increases sharply, it is classified as a sudden emergency, while a slow-accumulating one is classified as a gradual one. This classification provides a precise basis for subsequent processing.

[0254] Specifically, when the type of emergency is sudden, the system uses time series forecasting methods to analyze the development trend.

[0255] For example, using the ARIMA model to make short-term predictions on fused time-series data generates a trend curve for the next 30 minutes. In a simulated event, the trend curve shows that the disturbance magnitude is expected to rise from the current 0.8 units to 1.5 units, exhibiting an exponential growth pattern. Further calculation of the rate of change using the trend curve, for example, an increase of 0.12 units per minute, determines the trend level to be high. This rate assessment helps quantify the urgency of risk and supports rapid decision-making.

[0256] Preferably, based on the identified risk type and development trend level, the system matches a pre-established response rule base and outputs emergency response decision support.

[0257] For example, an emergency situation is matched with a high-level development trend to the rule of "immediate evacuation and activation of emergency equipment," generating a decision report containing a specific action list. This matching mechanism ensures that the response is highly consistent with the actual risk, significantly improving emergency response efficiency and targeting, and avoiding resource waste or insufficient response. The entire process forms a closed-loop support, emphasizing real-time performance and reliability from data collection to decision output.

[0258] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying exploration disturbances based on sky-ground collaboration, characterized in that, include: Acquire multi-source data collected by satellites and UAVs, extract spatial resolution parameters and temporal frequency parameters from the multi-source data, and process the parameters using a hierarchical segmentation method to obtain a registered sub-region fusion dataset; Based on the registered sub-region fusion dataset, analyze the topological relationships and attribute association rules, construct a multi-dimensional index structure, and retrieve historical perturbation patterns to obtain similar pattern matching results; Based on the similarity pattern matching results, the distribution characteristics of exploration activities and the probability of disturbance are evaluated, and the optimized movement trajectory scheme is obtained by dynamically planning the path and dwell time of the sky platform. Based on the optimized movement trajectory scheme, potential faults and weather-related disruptions are identified, and the path is adjusted through a multi-platform collaborative mechanism to obtain a supplementary observation coverage sequence. Based on the supplementary observation coverage sequence, spatiotemporal gaps caused by cloud cover and sensor failure are detected, and spatiotemporal interpolation is applied to estimate the missing information to obtain a compensated disturbance signal dataset. Based on the compensated disturbance signal dataset, the dispersion and distribution density of disturbance points are calculated, and the looseness index is quantified to determine the clustering pattern and obtain the clustering analysis output. If the clustering analysis output shows high looseness, then a vector field model is constructed by fusing satellite differential interferometry and ground tiltmeter data, and the propagation direction is analyzed to obtain a description of the disturbance dynamics characteristics; By comparing the disturbance dynamics feature description with a preset danger feature database, the occurrence of precursor signal combinations is determined to obtain an early warning trigger signal; Based on the early warning trigger signal, the system schedules encrypted observation and real-time fusion analysis to determine the type and development trend of the hazard and output corresponding emergency response decisions.

2. The method according to claim 1, characterized in that, The process of obtaining the registered sub-region fusion dataset by processing parameters using a hierarchical segmentation method includes: Acquire multi-source data composed of satellite imagery and UAV grids; Spatial resolution parameters and temporal frequency parameters are extracted based on the multi-source data; The hierarchical segmentation levels are determined based on the spatial resolution parameters; A hierarchical segmentation framework is used to perform hierarchical segmentation on the multi-source data, generating multiple sub-regions; Extract the corresponding spatial resolution parameters and temporal frequency parameters for the sub-region; If the difference in spatial resolution parameters of a sub-region exceeds a preset threshold, the sub-region boundaries will be adjusted and the sub-regions will be re-segmented. The spatial resolution parameters of the registered sub-regions are aligned using a registration transformation to obtain the registered sub-regions. The registration sub-regions are sorted according to the time frequency parameters to determine the fusion order; The registration sub-region is processed using the nearest neighbor interpolation algorithm to generate a fused dataset.

3. The method according to claim 1, characterized in that, The process of constructing a multidimensional index structure and retrieving historical perturbation patterns to obtain similar pattern matching results includes: By performing preliminary processing on the sub-region fusion dataset, topological relationship features are extracted, and a pre-established classification model is used to perform hierarchical labeling of the topological relationships to obtain a structured representation of the topological relationships. Based on the structured representation of the topological relationship, association analysis is performed in conjunction with attribute rules to obtain the corresponding mapping between attribute rules and topological relationships, and to determine the distribution of attribute rules in different topological levels. Based on the distribution of the attribute rules, a multidimensional index structure is constructed, and the topological relationships and attribute rules are embedded in the index structure to obtain a multidimensional index framework. The multidimensional indexing framework is used to retrieve historical disturbance data, and the support vector machine algorithm is used to classify the historical disturbance data to determine the disturbance category most relevant to the current dataset. If the topological relationship matching degree between the retrieved historical perturbation category and the current dataset is higher than the preset threshold, it is marked as a candidate pattern, and a preliminary pattern matching set is obtained. Based on the preliminary pattern matching set and the filtering conditions of similar results, the candidate patterns are compared a second time to obtain the final pattern matching results and determine the historical perturbation pattern that is closest to the current dataset. By storing the final pattern matching results in a structured manner and associating them with the regional analysis results of the fused dataset, a complete data record is obtained that can be queried subsequently.

4. The method according to claim 1, characterized in that, The process of dynamically planning the optimized movement trajectory scheme for the sky platform's path and dwell time includes: Historical data of exploration activities are obtained, and similarity pattern matching analysis is performed on the historical data to determine the preliminary patterns of distribution characteristics and obtain the classification results of distribution characteristics; Based on the classification results of the distribution characteristics, a probability assessment method is used to calculate the disturbance probability, and a preset threshold is used for screening to determine the distribution location of high-risk areas. If the distribution of high-risk areas exceeds the preset threshold range, the priority ranking of disturbance probabilities is generated based on the prediction results to divide the key monitoring areas. Based on the real-time location data of the sky platform obtained from the key monitoring areas, a preliminary plan for the movement path is calculated using dynamic programming methods to determine the initial framework for path planning. Based on the initial framework of the path planning, the allocation requirements for dwell time are analyzed, and adjustments are made using time window constraints to obtain an optimized configuration of dwell time. By optimizing the dwell time configuration and combining iterative calculations with the preliminary movement path plan, a final optimized trajectory plan is generated, and the coverage integrity of the trajectory plan is determined. Based on the coverage integrity of the trajectory scheme, the operating parameters of the sky platform are obtained, the execution strategy of dynamically adjusting the movement path and dwell time is adjusted, and the final operation plan is determined.

5. The method according to claim 1, characterized in that, The process of obtaining supplementary observation coverage sequences by adjusting the path through a multi-platform collaborative mechanism includes: Acquire optimized trajectory data and real-time weather data, and overlay potential fault records to form a comprehensive risk dataset; Identify weather disruption factors and potential fault locations from the comprehensive risk dataset, and mark high-risk trajectory segments to obtain risk-marked trajectories; Based on the risk-marked trajectory, query the available resources for multi-platform collaboration, determine the location and status of schedulable platforms, and obtain a list of collaborative resources. The risk-marked trajectory is adjusted by using the collaborative resource list to bypass high-risk trajectory segments and obtain the adjusted trajectory set. The adjusted trajectory set is assigned to the collaborative resource list using the nearest neighbor matching algorithm to generate a preliminary supplementary observation allocation scheme. If the initial supplementary observation allocation scheme has observation coverage gaps, a secondary path adjustment is triggered to fill the gap areas and obtain a complete supplementary trajectory. The complete complement trajectory serialization process generates a complement observation coverage sequence.

6. The method according to claim 1, characterized in that, The process of using spatiotemporal interpolation to estimate missing information and obtain a compensated perturbation signal dataset includes: Obtain the supplementary observation coverage sequence and the main disturbance signal dataset; The signal values ​​at corresponding positions in the supplementary observation coverage sequence are compared point by point to determine whether there is a deviation. If the deviation exceeds a preset threshold, the corresponding position is marked as a potential spatiotemporal gap. Based on the location of the potential spatiotemporal vulnerability, extract the effective signal values ​​of the surrounding neighboring points to form a local spatiotemporal neighborhood window; Using the effective signal values ​​within the local spatiotemporal neighborhood window, the Kriging interpolation algorithm is used to calculate the estimated location of the vulnerability, thus obtaining a preliminary filling signal. Spatial gradient and temporal continuity indices are obtained from the initial filling signal and the neighborhood window signal. It is determined whether the filling value is consistent with the neighborhood. If they are inconsistent, the value is adjusted to the neighborhood weighted average to determine the corrected filling signal. The corrected filling signal is used to replace the spatiotemporal gap position in the original dataset to obtain the compensated perturbation signal dataset. By performing point-by-point difference calculations between the compensated disturbance signal dataset and the original coverage sequence, it is determined whether the remaining deviations are all within a preset threshold, and the final compensated disturbance signal dataset is determined.

7. The method according to claim 1, characterized in that, The process of quantifying the looseness index to determine the clustering pattern and obtain the clustering analysis output includes: The original data points are extracted from the compensated disturbance signal dataset to obtain the spatial location information of the disturbance points and determine the preliminary distribution range. Based on the preliminary distribution range, a grid division method is used to map the disturbance points to the corresponding grid cells, and the number of points in each cell is counted. Based on the number of points in each grid cell, the local density value is calculated. If the local density value exceeds the preset threshold, it is marked as a high-density area to identify potential clustering areas. The spatial coordinates of the disturbance points are extracted from the high-density region, the distance relationship between the points is analyzed, and the specific value of the degree of dispersion is obtained. Based on the specific numerical value of the degree of dispersion, combined with the local density value, the looseness index is calculated to determine the overall distribution characteristics of the disturbance points. The clustering patterns are divided by comparing the looseness index with the preset pattern classification criteria and the K-means clustering algorithm is used to obtain the final clustering analysis results. Based on the clustering analysis results, corresponding pattern classification labels are generated, and a description of the clustering pattern of the perturbation signal is output.

8. The method according to claim 1, characterized in that, The process of analyzing the propagation direction to obtain a description of the disturbance dynamics includes: Acquire satellite differential interferometry data and ground tiltmeter data, and perform spatiotemporal alignment processing to obtain a synchronized dataset in a unified coordinate system; The K-means clustering algorithm was used to cluster the synchronous dataset, resulting in multiple cluster centers and intra-cluster point distributions. Calculate the variance of the distance from each point within a cluster to the cluster center to determine the looseness; If the looseness is higher than a preset threshold, the vector field construction process is initiated; By fitting the displacement gradient in the synchronous dataset using the least squares method, a continuous vector field model is constructed to obtain the vector value of each grid point. By analyzing the consistency of vector directions between adjacent grid points in the vector field model, the direction of disturbance propagation is determined, and the main propagation path is obtained. The disturbance dynamics features are extracted based on the vector amplitude changes along the main propagation path to obtain a feature description sequence.

9. The method according to claim 1, characterized in that, The process of receiving a warning trigger signal includes: By collecting disturbance dynamics data in the environment and using sensor networks to obtain real-time dynamic change information, a preliminary set of dynamic characteristics is determined. Based on the preliminary set of dynamic features, feature extraction techniques are used to reduce the dimensionality of the data, resulting in a simplified set of feature vectors. The simplified feature vector group is compared one by one with the preset danger feature database. If the comparison result exceeds the preset threshold range, it is determined to be a potential precursor signal. The potential precursor signals are combined and analyzed to obtain the correlation patterns between the signals and to determine whether there are signal combinations that meet the conditions for a dangerous situation. If a combination of signals that meets the conditions for a dangerous situation is detected, the early warning triggering mechanism is activated, and a corresponding early warning signal is output. Based on the output of the warning signal, relevant systems are linked to transmit information and determine the final warning response level; The corresponding emergency response procedures are obtained by classifying and processing the aforementioned early warning response levels.

10. The method according to claim 1, characterized in that, The process of determining the type and trend of a hazard and making corresponding emergency response decisions includes: Upon receiving an early warning trigger signal, the system initiates the collection of encrypted observation data. Acquire encrypted observation data, decrypt it to form a real-time observation dataset; The random forest algorithm was used to perform multi-source data fusion on the real-time observation dataset to obtain the fusion analysis results; Based on the fusion analysis results, multidimensional indicators are extracted to determine the type of danger. If the type of danger is sudden, then the trend curve is obtained by using time series forecasting method to analyze the development trend; The rate of change is calculated using the trend curve to determine the level of the development trend; Based on the type of hazard and the level of development trend, pre-established response rules are matched to output emergency response decision support.

Images