Geological deformation monitoring method and device based on remote sensing data, equipment and medium

Abstract

The invention discloses a geological deformation monitoring method and device based on remote sensing data, equipment and a medium. The method comprises the following steps: acquiring real-time multi-source remote sensing data corresponding to a target area of a transformer substation; carrying out fusion analysis based on the real-time multi-source remote sensing data, and extracting texture features corresponding to the external damage hidden danger target; based on the dynamic change model and the texture features, obtaining an external damage hidden danger probability distribution diagram and a man-made interference area corresponding to the external damage hidden danger target; acquiring surface deformation monitoring data of the target area based on the man-made interference area; optimizing the space reference of the SAR satellite image based on the Beidou positioning data to obtain a three-dimensional deformation field; constructing a live-action three-dimensional base map corresponding to the transformer substation, and generating an interactive visual report based on the live-action three-dimensional base map; according to the invention, wide-area intelligent identification of external damage hidden dangers of the transformer substation and high-precision three-dimensional monitoring of geological deformation are realized, and the real-time performance and reliability of collaborative analysis of multi-source data are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a geological deformation monitoring method, device and equipment based on remote sensing data, and belongs to the field of geological monitoring. BACKGROUND

[0002] The safe and stable operation of a substation, as a core hub of a power system, is of great importance to social production and life. However, the substation is often threatened by both external damage (such as construction damage and illegal loading) and geological deformation (such as ground subsidence and karst collapse).

[0003] Traditional monitoring methods mainly rely on manual inspection and single-point measurement technology (such as GPS and level), but have many defects. On the one hand, manual inspection is costly and time-consuming, and it is difficult to achieve wide-area real-time monitoring, and the efficiency and coverage are obviously insufficient. On the other hand, the spatial resolution of single-point GPS measurement is low, and it is difficult to capture the overall trend of regional deformation. In addition, optical remote sensing technology is easily affected by cloudy and rainy weather, and SAR satellites have all-weather capability, but in urban areas, the signal is incoherent due to building obstruction, thereby reducing the deformation monitoring accuracy. Moreover, existing monitoring methods are mostly independent in dealing with external damage and geological deformation, lack of collaborative analysis of multi-source data (such as optical, SAR and Beidou), and it is difficult to achieve comprehensive risk assessment, and there is a significant lack of data fusion.

[0004] In addition, although the existing InSAR technology can achieve millimeter-level deformation monitoring, it is not combined with high-precision positioning of Beidou, and the vertical deformation error is large; and the identification of external damage relies on visual interpretation, with low automation and easy to miss key risk points.

[0005] Therefore, how to break through the limitations of traditional monitoring methods, achieve wide-area intelligent identification of external damage of substations and high-precision three-dimensional monitoring of geological deformation, and improve the real-time performance and reliability of collaborative analysis of multi-source data, has become a technical problem to be solved. SUMMARY

[0006] In order to break through the limitations of traditional monitoring methods, achieve wide-area intelligent identification of external damage of substations and high-precision three-dimensional monitoring of geological deformation, and improve the real-time performance and reliability of collaborative analysis of multi-source data, the present application provides a geological deformation monitoring method, device, equipment and medium based on remote sensing data.

[0007] In a first aspect, the present application provides a geological deformation monitoring method based on remote sensing data, comprising:

[0008] obtaining real-time multi-source remote sensing data corresponding to a target area of a substation, the real-time multi-source remote sensing data comprising optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data and LiDAR point cloud data;

[0009] Perform fusion analysis based on the real-time multi-source remote sensing data to extract texture features corresponding to the external damage hidden danger target, the external damage hidden danger target including construction machinery, illegal stacking, temporary buildings and vegetation intrusion areas in the target area;

[0010] Based on the dynamic change model and the texture features, obtain an external damage hidden danger probability distribution map corresponding to the external damage hidden danger target, and based on the external damage hidden danger probability distribution map, obtain a human disturbance area;

[0011] Based on the human disturbance area and the time series InSAR technology, process the SAR satellite image to obtain ground surface deformation monitoring data of the target area, the ground surface deformation monitoring data including a deformation rate map and a cumulative deformation amount;

[0012] Based on the Beidou corner reflector, obtain Beidou positioning data; and based on the Beidou positioning data, optimize the spatial reference of the SAR satellite image to obtain a three-dimensional deformation field;

[0013] Based on the LiDAR point cloud data and the unmanned aerial vehicle oblique photogrammetry data, construct a real scene three-dimensional base map corresponding to the substation, and based on the real scene three-dimensional base map, the external damage hidden danger probability distribution map, the geological deformation monitoring data and the three-dimensional deformation field, generate an interactive visualization report.

[0014] The application has the advantages of breaking through the limitations of traditional monitoring methods, realizing wide-area intelligent identification of external damage hidden dangers of substations and high-precision three-dimensional monitoring of geological deformation, and improving the real-time performance and reliability of multi-source data collaborative analysis.

[0015] Further, after the spatial reference of the SAR satellite image is optimized based on the Beidou positioning data to obtain a three-dimensional deformation field, the method further comprises:

[0016] Obtain geological survey data, the geological survey data including geological drilling data and construction period foundation settlement monitoring reports, the geological drilling data including soil compression coefficients; and perform spatial matching between the deformation rate map and the soil compression coefficients to calculate a foundation stability index;

[0017] Based on a risk assessment model, obtain the foundation stability index, the external damage hidden danger probability distribution map and the three-dimensional deformation field, and real-time risk levels of each of a plurality of sub-areas in the target area corresponding to the three;

[0018] For each of the real-time risk levels, determine whether the real-time risk level is higher than a target risk threshold; if the real-time risk level is higher than the target risk threshold, the sub-area corresponding to the real-time risk level is automatically pushed to an operation and maintenance platform, and an adaptive protection strategy is started for the sub-area.

[0019] The beneficial effect of the above further scheme is that the potential risk area can be more accurately identified, and the accuracy of risk assessment is improved; the real-time monitoring and early warning mechanism can timely inform the operation and maintenance personnel to take measures, effectively reducing the probability of risk events; and the adaptive protection strategy is automatically started according to the risk level, which can dynamically adjust the protection measures according to the actual situation, not only improving the timeliness of risk response, but also enhancing the effectiveness of protection measures, and better guaranteeing the safe operation of the substation.

[0020] Further, before judging whether the real-time risk level is higher than the target risk threshold for each of the real-time risk levels, further comprising:

[0021] The real-time multi-source remote sensing data is normalized, and based on the processed real-time multi-source remote sensing data, a current state space is extracted, the current state space including a ground subsidence rate, a distance between the external damage hidden danger target and a key facility, a moving speed of the external damage hidden danger target, or an amplitude of geological deformation;

[0022] Based on a deep Q network, Q values of each preset action corresponding to the current state space are calculated, the preset action corresponding to an adjustment amplitude of a risk threshold;

[0023] The action with the maximum Q value is selected as a current decision action, the current risk threshold is adjusted according to the current decision action, and the adjusted risk threshold is taken as the target risk threshold.

[0024] The beneficial effect of the above further scheme is that the dynamic adjustment mechanism can more flexibly adapt to the actual situation of the monitoring area, and in a complex geological and human interference environment, false negatives and false positives can be effectively reduced, and the accuracy of early warning can be improved.

[0025] Further, after the adjusted risk threshold is taken as the target risk threshold, further comprising:

[0026] Based on a deep generation model, historical multi-source remote sensing data corresponding to the target area and the real-time multi-source remote sensing data are input, and a prediction evolution scene corresponding to each of the sub-regions is generated, the prediction evolution scene including a development trend of geological deformation and a motion trajectory of the external damage hidden danger target;

[0027] Based on a pre-trained discriminant model, each of the generated prediction evolution scenes is respectively risk assessed, and a prediction risk level corresponding to each of the prediction evolution scenes is calculated;

[0028] Based on the target risk threshold, a high-risk scene in each of the sub-regions is screened out;

[0029] For each screened high-risk scene, based on the specific content of the high-risk scene, obtain early warning protection information.

[0030] The beneficial effect of the further scheme is that the deep generative model generates a predicted evolution scene, which can simulate the future development of geological deformation and external hidden dangers, and provide forward-looking risk prediction; by generating a predicted evolution scene in advance and conducting risk assessment, timely warnings can be issued before the risk occurs, providing sufficient time for operation and maintenance personnel to take preventive measures and reducing the actual occurrence probability of risk events.

[0031] Further, before the real-time multi-source remote sensing data corresponding to the target area of the substation is obtained, the method further comprises:

[0032] Based on the federated learning framework, update information of a local model corresponding to each data source node is uploaded to the center server for updating of a global model, the update information including gradients of model parameters or updated model parameters; the local model is used for preliminary processing and feature extraction of the initial multi-source remote sensing data collected locally;

[0033] The model parameters of the updated global model are distributed back to each data source node, so that each data source node corresponds to the local model and adjusts and optimizes according to the update of the global model;

[0034] Based on homomorphic encryption technology, the update information of the local model is encrypted; and on each data source node, the initial multi-source remote sensing data collected locally is processed using the updated local model to generate local monitoring results, and the local monitoring results are aggregated to the center server to form global real-time multi-source remote sensing data.

[0035] The beneficial effect of the further scheme is that the data privacy of the data source nodes is effectively protected, the risk of data leakage is reduced, and the data privacy protection capability of the entire system is further enhanced; each data source node can indirectly share the data features and patterns of other nodes through the update of the global model, so that the model can learn more extensive data features, and the generalization ability and accuracy of the model are enhanced.

[0036] Further, before the update information of the local model corresponding to each data source node is uploaded to the center server for updating of the global model based on the federated learning framework, the method further comprises:

[0037] For each data source node, the initial multi-source remote sensing data collected by the data source node is subjected to quality evaluation to obtain an evaluation result corresponding to the data source node, and the evaluation indexes of the quality evaluation include data integrity, noise level and data consistency;

[0038] Based on each of the evaluation results, the allocation weight corresponding to each of the data source nodes is dynamically adjusted, so that the global model is updated based on each of the allocation weights; the data source node with a higher weight contributes more to the global model.

[0039] The beneficial effect of the further scheme is that low-quality data can be found and processed in time, ensuring that the data input into the model has high reliability and availability; dynamic adjustment of the weight makes the global model rely more on the update information of high-quality data source nodes, thereby improving the reliability and accuracy of the model.

[0040] Further, before the quality of the initial multi-source remote sensing data collected by each data source node is evaluated to obtain the evaluation result corresponding to the data source node, the method further comprises:

[0041] Based on a preset time window, it is determined whether there is a data source node that has not collected initial multi-source remote sensing data within the time window, the time window being a parameter representing the time range of data collection;

[0042] For each data source node that has not collected initial multi-source remote sensing data within the time window, the historical multi-source remote sensing data of the data source node is obtained, and if there is historical multi-source remote sensing data related to the current monitoring task, first substitute data is generated based on the historical multi-source remote sensing data; if there is no historical multi-source remote sensing data related to the current monitoring task, it is determined whether there is another data source node associated with the data source node, and if there is another data source node, second substitute data is generated based on the relevant data collected by the other data source nodes through data interpolation or machine learning method; if there is no other data source node, the allocation weight of the data source node is set to zero and does not participate in the update of the global model; the first substitute data or the second substitute data is used as the initial multi-source remote sensing data corresponding to the data source node, and the step of evaluating the quality of the initial multi-source remote sensing data collected by the data source node to obtain the evaluation result corresponding to the data source node is executed;

[0043] For each data source node that has collected initial multi-source remote sensing data within the time window, the step of evaluating the quality of the initial multi-source remote sensing data collected by the data source node to obtain the evaluation result corresponding to the data source node is executed.

[0044] This step is the processing process before "for each data source node, quality evaluation is performed on the initial multi-source remote sensing data collected by the data source node to obtain the evaluation result corresponding to the data source node", that is, "how to determine the initial multi-source remote sensing data in different specific cases". After determining the initial multi-source remote sensing data, "for each data source node, quality evaluation is performed on the initial multi-source remote sensing data collected by the data source node to obtain the evaluation result corresponding to the data source node" can be executed.

[0045] The beneficial effects of the above further scheme are: reducing the data missing problem caused by the fact that part of the data source nodes do not collect data, improving the integrity and efficiency of data collection, through the generation and use of substitute data, the global model can be updated according to the plan, avoiding the delay caused by incomplete data, avoiding repeated collection and storage of data, and optimizing resource allocation.

[0046] In a second aspect, the present application provides a geological deformation monitoring device based on remote sensing data, which is applied to the above method and comprises:

[0047] The first acquisition module is configured to acquire real-time multi-source remote sensing data corresponding to a target area of a substation, wherein the real-time multi-source remote sensing data comprises optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data, and LiDAR point cloud data.

[0048] The feature extraction module is configured to perform fusion analysis based on the real-time multi-source remote sensing data, and extract texture features corresponding to external damage hidden danger targets, wherein the external damage hidden danger targets comprise construction machinery, illegal stacking objects, temporary buildings, and vegetation intrusion areas in the target area.

[0049] The second acquisition module is configured to acquire an external damage hidden danger probability distribution map corresponding to the external damage hidden danger targets based on a dynamic change model and the texture features, and acquire a human interference area based on the external damage hidden danger probability distribution map.

[0050] The third acquisition module is configured to process the SAR satellite images based on the human interference area and time series InSAR technology, and acquire ground surface deformation monitoring data of the target area, wherein the ground surface deformation monitoring data comprises a deformation rate map and a cumulative deformation amount.

[0051] The reference optimization module is configured to acquire Beidou positioning data based on a Beidou corner reflector, and optimize a spatial reference of the SAR satellite images based on the Beidou positioning data to obtain a three-dimensional deformation field.

[0052] The report generation module is used to construct a real-scene 3D base map of the substation based on the LiDAR point cloud data and the UAV oblique photogrammetry data, and to generate an interactive visualization report based on the real-scene 3D base map, the probability distribution map of external damage hazards, the geological deformation monitoring data, and the 3D deformation field.

[0053] Thirdly, this application provides an electronic device, including a processor and a memory, wherein the processor is coupled to the memory;

[0054] The processor is configured to execute a computer program stored in the memory, causing the electronic device to perform the method as described in any of the first aspects.

[0055] Fourthly, this application provides a computer-readable storage medium including a computer program or instructions that, when executed on a computer, cause the computer to perform the method as described in any of the first aspects. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the geological deformation monitoring method based on remote sensing data in Embodiment 1 of this application.

[0057] Figure 2 This is a schematic diagram of the geological deformation monitoring device based on remote sensing data in Embodiment 2 of this application.

[0058] Figure 3 This is a schematic diagram of the structure of the electronic device in Embodiment 3 of this application. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0060] Example 1:

[0061] Embodiment 1 of this application provides a method for monitoring geological deformation based on remote sensing data. This method can be executed by a device, which can be a server or a terminal device. The server can be an independent physical server, a server cluster composed of multiple physical servers, a distributed system, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet computer, desktop computer, etc., but is not limited to these.

[0062] like Figure 1 As shown, a geological deformation monitoring method based on remote sensing data, using electronic equipment as the execution subject, is described in its main process flow as follows (steps S101 to S106):

[0063] Step S101: Obtain real-time multi-source remote sensing data corresponding to the target area of the substation, wherein the real-time multi-source remote sensing data includes optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data, and LiDAR point cloud data.

[0064] In this embodiment, the real-time multi-source remote sensing data of the target area can be obtained by various remote sensing technologies. According to the requirements of the current monitoring task, appropriate optical satellites and SAR satellites are selected, a data collection plan is planned, and an unmanned aerial vehicle is dispatched to perform oblique photogrammetry according to a preset flight route, while a LiDAR device is started to collect data. Optical satellite images and SAR satellite images are obtained through a satellite data receiving station.

[0065] The optical satellite images are obtained by using high-resolution optical satellites (such as Sentinel-2, Landsat 8, etc.) to obtain optical images of the target area. These images can provide rich surface texture and color information, which is helpful for identifying external damage hidden targets (such as construction machinery, temporary buildings, etc.).

[0066] The SAR satellite images are obtained by using synthetic aperture radar (SAR) satellites (such as Sentinel-1, COSMO-SkyMed, etc.) to obtain SAR images of the target area. SAR images have all-weather and all-time imaging capabilities, and can detect small changes in the ground surface, making them suitable for geological deformation monitoring.

[0067] The unmanned aerial vehicle oblique photogrammetry data is obtained by using an unmanned aerial vehicle equipped with an oblique photogrammetry camera to take multiple-angle photographs of the target area, thereby obtaining high-resolution oblique photogrammetry data. The unmanned aerial vehicle oblique photogrammetry data can be used to construct a real scene three-dimensional model, providing a more intuitive three-dimensional background for identifying and monitoring external damage hidden targets.

[0068] The LiDAR point cloud data is obtained by using airborne or ground LiDAR equipment to obtain high-precision point cloud data of the target area. LiDAR point cloud data can provide elevation information of the ground surface, which is helpful for analyzing terrain changes and geological deformation.

[0069] Step S102: Perform fusion analysis based on the real-time multi-source remote sensing data to extract texture features corresponding to external damage hidden targets, wherein the external damage hidden targets include construction machinery, illegal stacking, temporary buildings, and vegetation intrusion areas in the target area.

[0070] In this embodiment, by preprocessing the obtained multi-source remote sensing data, errors and noise introduced during data collection can be eliminated, ensuring the consistency and usability of the data. Preprocessing can include geometric correction, radiation correction, cloud removal processing, etc. Remote sensing data of different resolutions and formats are unified to the same geographic coordinate system and resolution, which facilitates subsequent fusion analysis.

[0071] In this embodiment, image processing and computer vision technologies can be used to extract texture features of potential external damage targets from optical satellite imagery and UAV oblique photogrammetry data. For example, edge detection and texture analysis can be used to identify the outlines and texture patterns of targets such as construction machinery, illegally loaded objects, and temporary buildings; coherence analysis and phase unpacking processing of SAR satellite imagery can be performed to extract surface deformation information as auxiliary features for identifying potential external damage targets; and by combining elevation information from LiDAR point cloud data, the terrain changes in the target area can be analyzed to further enhance the ability to identify potential external damage targets.

[0072] Multi-source data fusion algorithms, such as wavelet transform, principal component analysis (PCA), or deep learning-based fusion methods, can be used to fuse the extracted texture features with deformation information to generate a comprehensive feature map. Then, machine learning algorithms (such as support vector machines, random forests, etc.) or deep learning models (such as convolutional neural networks) can be used to classify and identify the fused feature map to accurately extract the location and feature information of targets with external damage risks.

[0073] In the fields of geological deformation monitoring and power facility safety monitoring, external damage hazard targets refer to various targets or factors that may pose external damage risks to the safe operation of power facilities such as substations and transmission lines. These targets are usually related to human activities or changes in the natural environment and may trigger geological deformation or directly damage power facilities.

[0074] Construction machinery refers to the mechanical equipment used in various construction activities around substations or transmission lines, such as excavators, cranes, and bulldozers. Illegally stockpiled materials refer to heavy objects illegally piled up around substations or transmission lines, such as building materials, earthwork, and large equipment. Temporary structures refer to buildings temporarily erected around substations or transmission lines, such as sheds and temporary warehouses. Vegetation encroachment refers to vegetation growing around substations or transmission lines, such as trees and shrubs, which may affect the safe operation of power facilities. In addition to the above common targets, other factors that may damage power facilities include illegal excavation, illegal construction, and geological disasters (such as landslides and mudslides).

[0075] Step S103: Based on the dynamic change model and the texture features, obtain the probability distribution map of external damage hazards corresponding to the target of external damage hazard, and obtain the human interference area based on the probability distribution map of external damage hazard.

[0076] The dynamic change model can be constructed in advance. First, historical monitoring data of the target area is collected, including past external damage hidden danger event records, geological deformation data, etc., which can be used as input of the model, and a time series analysis method (such as ARIMA model) or a machine learning algorithm (such as long short-term memory network LSTM) is used to construct a dynamic change model for predicting the spatio-temporal change trend of the external damage hidden danger target.

[0077] The extracted texture features and the prediction results of the dynamic change model are combined to calculate the external damage hidden danger probability of each pixel or region in the target area. The external damage hidden danger probability distribution map can be generated by interpolation and spatial analysis method, which can intuitively show the risk degree of external damage hidden danger in different positions in the target area.

[0078] According to the external damage hidden danger probability distribution map, the area with high probability value can be identified as the human interference area. Combined with geographic information system (GIS) technology, spatial analysis can be performed on the human interference area to extract its geographic range, area, shape and other information, which provides basis for subsequent monitoring and early warning.

[0079] Step S104: Based on the human interference area and the time series InSAR technology, the SAR satellite image is processed to obtain the ground surface deformation monitoring data of the target area, and the ground surface deformation monitoring data includes deformation rate map and cumulative deformation amount.

[0080] In this embodiment, the SAR satellite image can be preprocessed first, including image registration, flat earth effect removal and other operations to ensure the geometric consistency between images. Time series InSAR technology (such as PS-InSAR, SBAS-InSAR, etc.) is used to perform interference processing and phase unwrapping on the processed SAR satellite image sequence to extract the ground surface deformation information of the target area. Through time series analysis, the deformation rate and cumulative deformation amount of each pixel are calculated to generate the deformation rate map and the cumulative deformation amount map.

[0081] The spatial range of the human interference area, the deformation rate map and the cumulative deformation amount map are superimposed and analyzed to extract the ground surface deformation data in the human interference area, and the deformation characteristics of the human interference area are analyzed, such as the change trend of the deformation rate, the size of the cumulative deformation amount, etc. The influence degree of human activities on geological deformation can be evaluated.

[0082] Step S105: Based on the Beidou corner reflector, Beidou positioning data is obtained, and the spatial reference of the SAR satellite image is optimized based on the Beidou positioning data to obtain a three-dimensional deformation field.

[0083] In this embodiment, by reasonably arranging the Beidou corner reflector in the target area, it can ensure that it can receive Beidou satellite signals and have good geometric distribution. The Beidou positioning device receives the Beidou satellite signal to obtain high-precision Beidou positioning data of the Beidou corner reflector. The Beidou positioning data can include longitude, latitude and elevation information.

[0084] The Beidou positioning data of the Beidou corner reflector is used as a control point to jointly solve the SAR satellite image, which can optimize the spatial reference of the SAR satellite image. Through geometric correction and error compensation, the positioning accuracy of the SAR satellite image is improved, and the consistency with the Beidou positioning data is ensured.

[0085] Combined with the optimized SAR satellite image and the Beidou positioning data, a three-dimensional deformation inversion algorithm is used to calculate the three-dimensional deformation field of the target area and generate three-dimensional deformation field data. The three-dimensional deformation field data can include horizontal and vertical deformation components, providing support for comprehensive analysis of geological deformation.

[0086] Step S106: Based on the LiDAR point cloud data and the unmanned aerial oblique photogrammetry data, a real three-dimensional base map corresponding to the substation is constructed, and based on the real three-dimensional base map, the external damage hidden danger probability distribution map, the geological deformation monitoring data and the three-dimensional deformation field, an interactive visualization report is generated.

[0087] In this embodiment, by processing the LiDAR point cloud data, a digital elevation model of the target area can be generated. By fusing the unmanned aerial oblique photogrammetry data with the digital elevation model, a real three-dimensional model of the substation can be constructed. By texture mapping and detail enhancement of the real three-dimensional model, a high-resolution real three-dimensional base map can be generated, providing an intuitive three-dimensional background for subsequent visualization analysis.

[0088] Superimposing the external damage hidden danger probability distribution map, the geological deformation monitoring data and the three-dimensional deformation field data onto the real three-dimensional base map can realize the visualization fusion of multi-source monitoring data. By using geographic information system (GIS) and three-dimensional visualization technology, an interactive visualization report system is developed to support users to view and analyze monitoring data intuitively through map browsing, data query, chart display and other functions. In the visualization report, early warning information display and protection suggestion prompt can be provided to help operation and maintenance personnel to understand the risk situation in time and take corresponding measures.

[0089] This scheme realizes wide-area monitoring of substations and surrounding areas by fusing optical satellite images, SAR satellite images, unmanned aerial oblique photogrammetry data and LiDAR point cloud data, breaking through the limitations of traditional monitoring methods, and being able to fully cover the target area and obtain multi-source data in real time, providing more comprehensive and timely information for external damage hidden danger and geological deformation monitoring.

[0090] The scheme fuses optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data, and LiDAR point cloud data to extract texture features of external damage hidden danger targets, and realizes wide-area intelligent identification of external damage hidden danger targets such as construction machinery, illegal stacking, temporary buildings, and vegetation intrusion areas.

[0091] The scheme uses Beidou corner reflectors to obtain high-precision positioning data, optimizes the spatial reference of SAR satellite images, further improves the precision of deformation monitoring, obtains a three-dimensional deformation field, combines LiDAR point cloud data and unmanned aerial vehicle oblique photogrammetry data to construct a real three-dimensional base map of the substation, fuses geological deformation monitoring data with the three-dimensional base map, and realizes high-precision three-dimensional visualization and analysis of geological deformation.

[0092] The scheme generates an interactive visualization report, which can support users to view and analyze monitoring data in real time, discover potential risks in time and take measures, and improve the practicality and reliability of the monitoring system.

[0093] Further, after optimizing the spatial reference of the SAR satellite image based on the Beidou positioning data and obtaining the three-dimensional deformation field, the method further comprises:

[0094] Obtaining geological survey data, the geological survey data including geological drilling data and construction period foundation settlement monitoring reports, the geological drilling data including soil layer compression coefficients, and spatially matching the deformation rate map with the soil layer compression coefficients to calculate a foundation stability index;

[0095] Based on a risk assessment model, obtaining the foundation stability index, the external damage hidden danger probability distribution map, and the three-dimensional deformation field, and corresponding real-time risk levels of each of a plurality of sub-regions in the target region;

[0096] For each real-time risk level, if the real-time risk level is higher than a target risk threshold, the sub-region corresponding to the real-time risk level is automatically pushed to an operation and maintenance platform, and an adaptive protection strategy is started for the sub-region.

[0097] Collecting geological survey data of the target region, the geological survey data including geological drilling data and construction period foundation settlement monitoring reports; the geological drilling data provides detailed information of the stratum, such as soil layer type, thickness, compression coefficient, etc.; the construction period foundation settlement monitoring report records the settlement of the foundation during the construction process.

[0098] The deformation rate map is spatially matched with the soil layer compression coefficient in the geological drilling data. Through geographic information system (GIS) technology, the deformation rate map and the geological drilling data can be superimposed in the same geographic coordinate system, ensuring that the position of each geological drilling point is consistent with the corresponding position in the deformation rate map. Based on the matched data, the foundation stability index is calculated using a foundation stability analysis model. The model considers factors such as the compression coefficient of the soil layer and the deformation rate to evaluate the stability of the foundation.

[0099] In this embodiment, the foundation stability index is calculated using the following formula:

[0100] ;

[0101] The risk assessment model can be a machine learning-based model (such as random forest, support vector machine) or a deep learning model (such as convolutional neural network), and is pre-trained with the foundation stability index, external damage probability distribution map, and three-dimensional deformation field data as input. Through historical data training, it can identify the contribution of different factors to risk.

[0102] The risk assessment model calculates the real-time risk level of each sub-region in the target area according to the input data. The risk level can be divided into three levels: low risk, medium risk, and high risk. The specific classification criteria can be determined based on historical data and expert experience.

[0103] For each sub-region, if the real-time risk level is higher than the target risk threshold, the information of the sub-region is automatically pushed to the operation and maintenance platform. The pushed information can include the location of the sub-region, the risk level, the main risk factors, etc. The operation and maintenance platform automatically starts the corresponding adaptive protection strategy according to the received risk information. The adaptive protection strategy can include increasing the inspection frequency, temporarily reinforcing the foundation, suspending nearby construction activities, etc., to reduce risk and ensure the safe operation of the substation.

[0104] By calculating the foundation stability index and combining geological conditions with deformation monitoring data, the stability of the foundation is quantified, which can more accurately identify potential risk areas and improve the accuracy of risk assessment. By calculating and updating the risk level of each sub-region in real time and issuing an early warning immediately when the risk level exceeds the threshold, the real-time monitoring and early warning mechanism can timely notify the operation and maintenance personnel to take measures, effectively reducing the probability of risk events. According to the risk level, the adaptive protection strategy is automatically started, which can dynamically adjust the protection measures according to the actual situation, not only improving the timeliness of risk response, but also enhancing the effectiveness of protection measures, better ensuring the safe operation of the substation.

[0105] In this embodiment, before determining whether the real-time risk level is higher than the target risk threshold for each real-time risk level, the method further includes:

[0106] normalize the real-time multi-source remote sensing data, and based on the processed real-time multi-source remote sensing data, extract a current state space, the current state space including a ground subsidence rate, a distance between the external damage hidden danger target and a key facility, a moving speed of the external damage hidden danger target, or an amplitude of geological deformation;

[0107] based on a deep Q network, calculate Q values of each preset action corresponding to the current state space, the preset action corresponding to an adjustment amplitude of a risk threshold value;

[0108] select an action with the largest Q value as a current decision action, adjust the current risk threshold value according to the current decision action, and take the adjusted risk threshold value as the target risk threshold value.

[0109] By normalizing the collected real-time multi-source remote sensing data, it can be ensured that the data of different data sources has comparability and consistency. The normalization processing method includes data standardization (such as Z-score standardization) and data normalization (such as Min-Max normalization). For example, the real-time multi-source remote sensing data can be normalized to the range of 0 to 1.

[0110] From the normalized real-time multi-source remote sensing data, the features of the current state space can be extracted. The current state space includes the ground subsidence rate, the distance between the external damage hidden danger target and the key facility, the moving speed of the external damage hidden danger target, and the amplitude of geological deformation.

[0111] In this embodiment, key features related to risk assessment can be selected to ensure that the extracted features can fully reflect the state of the current monitoring area; for example, the ground subsidence rate can be extracted from SAR satellite images through time series InSAR technology; the distance between the external damage hidden danger target and the key facility can be calculated through optical satellite images and unmanned aerial vehicle oblique photogrammetry data.

[0112] The deep Q network is a deep Q network (DQN) model. A pre-trained deep Q network model is used, which is trained through historical data and expert experience and can learn the mapping relationship between different state space features and optimal actions. When applied, the extracted current state space features are input into the deep Q network to calculate the Q value of each preset action. The Q value represents the expected cumulative reward of executing a certain action in the current state. The preset action includes the adjustment amplitude of the risk threshold value, such as reducing 5%, reducing 10%, increasing 5%, or increasing 10%, etc.

[0113] The action corresponding to the maximum Q value is selected as the current decision action from the calculated Q values. For example, if the action corresponding to a 5% reduction in the risk threshold has the maximum Q value, the action is selected as the current decision action, and the current risk threshold is adjusted according to the current decision action. The adjusted risk threshold will be used for subsequent risk assessment and early warning, and the adjusted risk threshold is updated as the target risk threshold for subsequent real-time risk assessment. At the same time, the updated target risk threshold is fed back to the monitoring system to ensure that the system uses the latest threshold for risk assessment and early warning.

[0114] For example, when the ground subsidence rate of a certain sub-region is found to suddenly accelerate, the target risk threshold of the region can be dynamically reduced based on past experience (similar situations learned from historical data) to issue an early warning.

[0115] Compared with the traditional fixed early warning threshold, this dynamic adjustment mechanism can more flexibly adapt to the actual situation of the monitoring area, and can effectively reduce false negatives and false positives in complex geological and human interference environments, improving the accuracy of early warning. For example, in a substation surrounded by complex geological conditions and frequent construction activities, through dynamic adjustment of the early warning threshold, potential foundation stability problems can be found at the beginning of construction, and timely measures can be taken to avoid accidents.

[0116] Through the deep Q network, the risk threshold can be automatically adjusted based on real-time monitoring data without human intervention. This self-adaptive adjustment mechanism can quickly respond to changes in the monitoring area, improving the intelligent level of the geological deformation monitoring method based on remote sensing data.

[0117] In the embodiment, after the adjusted risk threshold is used as the target risk threshold, the method further includes:

[0118] Based on the deep generative model, the historical multi-source remote sensing data corresponding to the target area and the real-time multi-source remote sensing data are input to generate a respective predicted evolution scenario for each of the sub-regions. The predicted evolution scenario includes the development trend of geological deformation and the motion trajectory of the external damage hidden danger target.

[0119] Based on the pre-trained discriminative model, each of the generated predicted evolution scenarios is subjected to risk assessment, and a predicted risk level corresponding to each of the predicted evolution scenarios is calculated.

[0120] Based on the target risk threshold, high-risk scenarios in each of the sub-regions are screened out.

[0121] For each of the screened high-risk scenarios, based on the specific content of the high-risk scenario, warning and protection information is obtained.

[0122] In this embodiment, first, historical multi-source remote sensing data and real-time multi-source remote sensing data of the target area are collected; the historical data includes past optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data and LiDAR point cloud data, and the real-time data is the latest monitoring data; the historical multi-source remote sensing data and the real-time multi-source remote sensing data are input into a deep generation model, which can be a generative adversarial network (GAN) or other similar generation model, and the purpose is to generate a predicted evolution scene; the deep generation model generates a predicted evolution scene for each sub-area according to the input historical and real-time data, and these scenes include the development trend of geological deformation and the motion trajectory of external damage hidden targets.

[0123] The generated predicted evolution scene is risk assessed using a discriminant model; the discriminant model can be a machine learning-based model (such as random forest, support vector machine) or a deep learning model (such as convolutional neural network), which is used to identify and assess risks in the scene; the discriminant model can calculate the predicted risk level of each scene according to the features (such as deformation rate, location and motion trajectory of external damage hidden targets) in the predicted evolution scene.

[0124] The calculated risk level is compared with a target risk threshold, high-risk scenes with a risk level higher than the target risk threshold are screened out, and relevant information (such as location, risk level, main risk factors, etc.) of these scenes is recorded.

[0125] According to the specific content of the high-risk scene, corresponding warning information is generated. The warning information includes risk type, risk level, predicted occurrence time, impact range, etc. Combined with the characteristics of the risk scene, specific protection suggestions are provided, such as strengthening inspection, suspending nearby construction, reinforcing foundation, etc. The warning information and protection suggestions are pushed to the operation and maintenance personnel, and information transmission can be carried out through various channels (such as SMS, email, mobile application push), to ensure that the operation and maintenance personnel receive and take measures in time.

[0126] The deep generation model generates a predicted evolution scene, which can simulate the future development of geological deformation and external damage hidden targets, provide forward-looking risk prediction, and through the generation of predicted evolution scenes in advance and risk assessment, timely warning can be given before the risk occurs, providing sufficient time for operation and maintenance personnel to take preventive measures and reducing the actual occurrence probability of risk events.

[0127] In the embodiment, before the real-time multi-source remote sensing data corresponding to the target area of the substation is acquired, the method further includes:

[0128] Based on the federated learning framework, the update information of each data source node corresponding to the local model is uploaded to the center server for updating the global model; the update information includes the gradient of the model parameters or the updated model parameters; the local model is used for preliminary processing and feature extraction of the initial multi-source remote sensing data collected locally;

[0129] The model parameters of the updated global model are distributed back to each data source node, so that each data source node corresponds to the local model, and adjusts and optimizes according to the update of the global model;

[0130] Based on homomorphic encryption technology, the update information of the local model is encrypted, and the updated local model is used to process the initial multi-source remote sensing data collected locally on each data source node to generate local monitoring results, and the local monitoring results are aggregated to the center server to form the global real-time multi-source remote sensing data.

[0131] Each data source node (such as different sensor networks or data collection platforms) can independently run a local model to preliminarily process and extract features from the collected initial multi-source remote sensing data. The local model can be a machine learning model for data classification, feature extraction or anomaly detection. The update information is extracted from the local model, which includes the gradient of the model parameters or the updated model parameters, and the update information reflects the learning result of the model on the current data.

[0132] After the center server receives the update information uploaded by each data source node, it uses an aggregation algorithm (such as federated averaging algorithm) to integrate these update information. The federated averaging algorithm weights and averages the gradient of the model parameters or the updated model parameters according to the data volume or quality weight of the data source node. According to the aggregated update information, the center server updates the parameters of the global model and optimizes the performance of the global model.

[0133] The center server distributes the updated global model parameters back to each data source node through a secure communication channel. Each data source node receives the distributed global model parameters and updates the local model with them, so that the local model can indirectly learn valuable information from the data of other data source nodes.

[0134] Select a suitable homomorphic encryption algorithm (such as additive homomorphic encryption, multiplicative homomorphic encryption or full homomorphic encryption) to encrypt the update information of the local model. The selected homomorphic encryption algorithm is used to encrypt the update information to ensure the confidentiality of the update information during data transmission and processing. The encrypted update information is uploaded to the center server as ciphertext, and the center server performs aggregation operation on the ciphertext without decryption.

[0135] Each data source node uses the updated local model to process and analyze newly collected initial multi-source remote sensing data, generating more accurate local monitoring results. Each data source node uploads the generated local monitoring results to the central server, and the central server aggregates all local monitoring results of the data source nodes to form a global real-time multi-source remote sensing data, providing comprehensive data support for subsequent analysis and decision-making.

[0136] The update information of the local model is encrypted by homomorphic encryption technology, ensuring that the data is not leaked during transmission and processing, effectively protecting the data privacy of the data source nodes. The federated learning framework only shares model update information rather than raw data, reducing the risk of data leakage and further enhancing the data privacy protection capability of the entire system. Each data source node can indirectly share the data features and patterns of other nodes through the update of the global model, enabling the model to learn more extensive data features and enhancing the generalization ability and accuracy of the model. Aggregating the update information of multiple data source nodes to optimize the global model can more effectively capture the commonality and regularity in the data, improving the overall performance and stability of the model.

[0137] In the embodiment, before uploading the update information of each data source node's corresponding local model to the central server for updating the global model based on the federated learning framework, the method further comprises:

[0138] For each data source node, the initial multi-source remote sensing data collected by the data source node is subjected to quality evaluation, and the evaluation result corresponding to the data source node is obtained. The evaluation indicators of the quality evaluation include data integrity, noise level, and data consistency.

[0139] Based on each evaluation result, the allocation weight corresponding to each data source node is dynamically adjusted, and the global model is updated based on each allocation weight. The data source node with a higher weight contributes more to the global model.

[0140] Data quality is evaluated using indicators such as data integrity, noise level, and data consistency. For example, data integrity can be measured by checking if the data is missing or damaged; noise level can be evaluated by calculating the signal-to-noise ratio (SNR); and data consistency can be determined by comparing the differences between data collected at different times or by different sensors.

[0141] According to the data quality evaluation result, a weight is assigned to each data source node, and the calculation of the weight can be performed according to a preset rule; for example, a node with high data quality is assigned a higher weight, and a node with low data quality is assigned a lower weight, and the weight is dynamically adjusted to reflect the current data quality of the data source node, and the weight adjustment can be performed periodically or triggered when a significant change in data quality is detected.

[0142] During the global model update, the contribution of each data source node is weighted using the weight of the node, and the node with a high weight contributes more to the update of the global model. The update information aggregated by weighting is used to optimize the parameters of the global model, so that the global model can better reflect the characteristics and patterns of high-quality data source nodes.

[0143] During the update process of the global model, the update information of the data source node with a high weight has a greater impact on the model parameters, thereby ensuring that the global model is more inclined to learn and reflect the characteristics of high-quality data source nodes. By giving high-quality data source nodes a greater contribution, the performance and accuracy of the global model are improved, and the negative impact of low-quality data on the model is reduced.

[0144] Through the quality evaluation step, low-quality data can be discovered and processed in a timely manner, ensuring that the data input into the model has high reliability and usability. Dynamic adjustment of the weight makes the global model rely more on the update information of high-quality data source nodes, thereby improving the reliability and accuracy of the model. This weight adjustment mechanism based on data quality helps to build a more stable and accurate model, which can dynamically adjust the weight according to the real-time evaluation result of data quality, adapt to the quality changes of data source nodes, and maintain good performance in the case of data quality fluctuations.

[0145] In the embodiment, before the quality evaluation of the initial multi-source remote sensing data collected by each data source node is performed to obtain the evaluation result corresponding to the data source node, the method further includes:

[0146] Based on a preset time window, it is determined whether there is a data source node that has not collected initial multi-source remote sensing data within the time window; the time window is a parameter representing the time range of data collection;

[0147] For each data source node that does not collect initial multi-source remote sensing data within the time window, obtain the historical multi-source remote sensing data of the data source node; if there is historical multi-source remote sensing data related to the current monitoring task, generate first replacement data based on the historical multi-source remote sensing data; if there is no historical multi-source remote sensing data related to the current monitoring task, determine whether there are other data source nodes associated with the data source node, if there are other data source nodes, generate second replacement data based on the relevant data collected by the other data source nodes through data interpolation or machine learning methods; if there are no other data source nodes, set the allocation weight of the data source node to zero and do not participate in the update of the global model; the first replacement data or the second replacement data is used as the initial multi-source remote sensing data corresponding to the data source node, and the quality evaluation of the initial multi-source remote sensing data collected by the data source node is performed to obtain the evaluation result corresponding to the data source node.

[0148] For each data source node that collects initial multi-source remote sensing data within the time window, the quality evaluation of the initial multi-source remote sensing data collected by the data source node is performed to obtain the evaluation result corresponding to the data source node.

[0149] The time window can be the last 24 hours or 48 hours, which is the time range of data collection, checking whether each data source node successfully collects and uploads initial multi-source remote sensing data within the time window, identifying data source nodes that do not collect data within the time window, and recording the identification information of these nodes.

[0150] Access the data center or data storage system to query the historical multi-source remote sensing data of the data source nodes that do not collect data, evaluate the relevance of the historical data to the current monitoring task, and ensure that the historical data can effectively replace the current missing data. If there is relevant historical data, use these historical data to generate first replacement data; the generation method can include directly copying the historical data or performing simple time series interpolation.

[0151] Check whether the data source nodes that do not collect data are associated with other data source nodes, such as adjacent geographical positions or overlapping monitoring areas, if there are associated other data source nodes, obtain the relevant data collected by these nodes within the time window, and use the data of the other data source nodes obtained to generate second replacement data through data interpolation or machine learning methods (such as K-nearest neighbor interpolation, linear regression, etc.).

[0152] Set the allocation weight of the data source node that does not collect data and has no historical data and associated node data to zero, ensure that these nodes do not participate in the current global model update, record the status of these nodes for subsequent maintenance or checking to ensure the reliability of data collection.

[0153] The missing initial multi-source remote sensing data is replaced with the generated first replacement data or second replacement data, the replacement data is integrated into the data set of the data source node, the integrity and consistency of the data set are ensured, the replacement data is evaluated, and a quality evaluation result of the replacement data is generated to ensure that the quality of the replacement data can meet the requirements of the monitoring task.

[0154] By generating replacement data, the problem of data missing caused by the fact that some data source nodes do not collect data is reduced, the integrity and efficiency of data collection are improved, and the global model can be updated according to the plan without waiting for all data source nodes to collect data. The delay caused by incomplete data is avoided, historical data and data of other associated nodes are fully utilized, repeated collection and storage of data are avoided, resource allocation is optimized, additional data collection and processing costs caused by data missing are reduced, and operating costs are reduced.

[0155] Embodiment 2:

[0156] The application also provides a geological deformation monitoring device based on remote sensing data, as shown in the figure, the geological deformation monitoring device 200 based on remote sensing data mainly comprises: Figure 2

[0157] The first acquisition module 201 is configured to acquire real-time multi-source remote sensing data corresponding to a target area of a substation, and the real-time multi-source remote sensing data comprises optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data, and LiDAR point cloud data.

[0158] The feature extraction module 202 is configured to perform fusion analysis based on the real-time multi-source remote sensing data and extract texture features corresponding to external damage hidden danger targets, wherein the external damage hidden danger targets comprise construction machinery, illegal stacking objects, temporary buildings, and vegetation intrusion areas in the target area.

[0159] The second acquisition module 203 is configured to acquire an external damage hidden danger probability distribution map corresponding to the external damage hidden danger targets based on a dynamic change model and the texture features, and acquire a human disturbance area based on the external damage hidden danger probability distribution map.

[0160] The third acquisition module 204 is configured to process the SAR satellite images based on the human disturbance area and time series InSAR technology, and acquire surface deformation monitoring data of the target area, wherein the surface deformation monitoring data comprises a deformation rate map and a cumulative deformation amount.

[0161] The reference optimization module 205 is configured to acquire Beidou positioning data based on a Beidou corner reflector, and optimize a spatial reference of the SAR satellite images based on the Beidou positioning data to obtain a three-dimensional deformation field. ​

[0162] The report generation module 206 is configured to construct a real three-dimensional base map corresponding to the substation based on the LiDAR point cloud data and the unmanned aerial photogrammetry data, and generate an interactive visualization report based on the real three-dimensional base map, the external damage hidden danger probability distribution map, the geological deformation monitoring data and the three-dimensional deformation field.

[0163] Further, the first acquisition module 201, the feature extraction module 202, the second acquisition module 203, the third acquisition module 204, the reference optimization module 205, and the report generation module 206 implement the steps of the functions, which are described in Embodiment 1.

[0164] In one example, the modules in any of the above apparatuses can be one or more integrated circuits configured to implement the above methods, for example, one or more application specific integrated circuits (ASICs), or one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs), or a combination of at least two of these integrated circuit forms.

[0165] For another example, when the modules in the apparatus can be implemented in the form of a processing element scheduler, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor that can invoke programs. For another example, the modules can be integrated together to be implemented in the form of a system-on-a-chip (SOC).

[0166] Embodiment 3

[0167] The present application also provides an electronic device, as shown in the figure, the electronic device 300 includes a processor 301 and a memory 302, and can further include one or more of an information input / output (I / O) interface 303, a communication component 304, and a communication bus 305. Figure 3

[0168] ​The processor 301 is configured to control overall operations of the electronic device 300 to complete all or part of the steps of the above-described geological deformation monitoring method based on remote sensing data. The memory 302 is configured to store various types of data to support operations of the electronic device 300. For example, the data can include instructions for any application or method operating on the electronic device 300, and application-related data. The memory 302 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as one or more of a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0169] The I / O interface 303 provides an interface between the processor 301 and other interface modules, which can be a keyboard, a mouse, a button, and the like. The buttons can be virtual buttons or physical buttons. The communication component 304 is configured to test wired or wireless communication between the electronic device 300 and other devices. The wireless communication, such as Wi-Fi, Bluetooth, near field communication (NFC), 2G, 3G, or 4G, or a combination of one or more of them, so the corresponding communication component 304 can include a Wi-Fi component, a Bluetooth component, and an NFC component.

[0170] The communication bus 305 can include a path for transmitting information between the above-mentioned components. The communication bus 305 can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, or the like. The communication bus 305 can be divided into an address bus, a data bus, a control bus, and the like.

[0171] The electronic device 300 can be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic elements for performing the remote sensing data-based geological deformation monitoring method given in the above-described embodiments.

[0172] The electronic device 300 can include, but is not limited to, a mobile terminal of a digital broadcast receiver, a PDA (Personal Digital Assistant), a PMP (Portable Multimedia Player), and the like, and a fixed terminal such as a digital TV, a desktop computer, and the like, and can also be a server or the like.

[0173] Based on the same technical concept, the present application also provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the remote sensing data-based geological deformation monitoring method described above.

[0174] The computer readable storage medium can include a U disk, a mobile hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0175] In the present application, various messages / information / equipment / network elements / systems / devices / actions / operations / processes / concepts and other types of objects that can appear in the present application are named. It can be understood that these specific names do not constitute a limitation on the related objects, and the names can be changed according to the scene, context or usage habits, etc. The technical meaning of the technical terms in the present application should be mainly determined according to the function and technical effect embodied / implemented in the technical scheme.

[0176] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and module described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0177] Those skilled in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0178] The term "comprising" or "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or other elements inherent to such process, method, article or apparatus.

[0179] In addition, the terms "first", "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0180] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in combination with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0181] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for geologic deformation monitoring based on remote sensing data, characterized in that, The method comprises the following steps: acquiring real-time multi-source remote sensing data corresponding to a target area of a substation; the real-time multi-source remote sensing data comprises optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data and LiDAR point cloud data; performing fusion analysis based on the real-time multi-source remote sensing data to extract texture features corresponding to external damage hidden danger targets; the external damage hidden danger targets comprise construction machinery, illegal stacking objects, temporary buildings and vegetation intrusion areas in the target area; based on a dynamic change model and the texture features, acquiring an external damage hidden danger probability distribution map corresponding to the external damage hidden danger targets, and based on the external damage hidden danger probability distribution map, acquiring a human interference area; based on the human interference area and time-series InSAR technology processing the SAR satellite images, acquiring ground surface deformation monitoring data of the target area; the ground surface deformation monitoring data comprises a deformation rate map and a cumulative deformation amount; based on a Beidou corner reflector, acquiring Beidou positioning data, and based on the Beidou positioning data, optimizing a spatial reference of the SAR satellite images to obtain a three-dimensional deformation field; based on the LiDAR point cloud data and the unmanned aerial vehicle oblique photogrammetry data, constructing a real scene three-dimensional base map corresponding to the substation, and based on the real scene three-dimensional base map, the external damage hidden danger probability distribution map, the geological deformation monitoring data and the three-dimensional deformation field, generating an interactive visual report.

2. The method according to claim 1, wherein, After the spatial reference of the SAR satellite images is optimized based on the Beidou positioning data to obtain the three-dimensional deformation field, the method further comprises the following steps: acquiring geological survey data, the geological survey data comprising geological drilling data and construction period foundation settlement monitoring reports, the geological drilling data comprising soil layer compression coefficients; and performing spatial matching between the deformation rate map and the soil layer compression coefficients to calculate a foundation stability index; based on a risk assessment model, acquiring the foundation stability index, the external damage hidden danger probability distribution map and the three-dimensional deformation field, and real-time risk levels of a plurality of sub-areas in the target area corresponding to the three; for each of the real-time risk levels, judging whether the real-time risk level is higher than a target risk threshold value; if the real-time risk level is higher than the target risk threshold value, the sub-area corresponding to the real-time risk level is automatically pushed to an operation and maintenance platform, and an adaptive protection strategy is started for the sub-area.

3. The method according to claim 2, wherein, Before judging whether the real-time risk level is higher than the target risk threshold value for each of the real-time risk levels, the method further comprises the following steps: performing normalization processing on the real-time multi-source remote sensing data, and based on the processed real-time multi-source remote sensing data, extracting a current state space; the current state space comprises a foundation settlement rate, distances between the external damage hidden danger targets and key facilities, moving speeds of the external damage hidden danger targets or amplitudes of geological deformation; based on a deep Q network, calculating Q values of each preset action corresponding to the current state space; the preset action corresponds to an adjustment amplitude of a risk threshold value; selecting an action with the largest Q value as a current decision action, adjusting the current risk threshold value according to the current decision action, and taking the adjusted risk threshold value as the target risk threshold value.

4. The method according to claim 3, wherein, After the adjusted risk threshold is taken as the target risk threshold, further comprising: Based on the deep generative model, input the historical multi-source remote sensing data and real-time multi-source remote sensing data corresponding to the target area, generate a prediction evolution scene corresponding to each of the sub-regions; the prediction evolution scene includes the development trend of geological deformation and the motion trajectory of external hidden danger targets; Based on the discriminant model, the generated each prediction evolution scene is respectively evaluated, and the prediction risk level corresponding to each prediction evolution scene is calculated; Based on the target risk threshold, filter out high-risk scenes in each of the sub-regions; For each filtered high-risk scene, based on the specific content of the high-risk scene, obtain early warning and protection information.

5. The method of claim 1, wherein, Before the real-time multi-source remote sensing data corresponding to the target area of the substation is obtained, further comprising: Based on the federated learning framework, upload the update information of the local model corresponding to each data source node to the center server to update the global model; the update information includes the gradient of the model parameter or the updated model parameter; the local model is used to preliminarily process and feature extract the initial multi-source remote sensing data collected locally; Distribute the model parameters of the updated global model back to each data source node, so that each data source node corresponds to the local model, and adjusts and optimizes according to the update of the global model; Based on homomorphic encryption technology, encrypt the update information of the local model, and use the updated local model to process the initial multi-source remote sensing data collected locally on each data source node to generate local monitoring results, and then aggregate the local monitoring results to the center server to form global real-time multi-source remote sensing data.

6. The method according to claim 5, wherein, Before the update information of the local model corresponding to each data source node is uploaded to the center server based on the federated learning framework to update the global model, further comprising: For each data source node, perform quality assessment on the initial multi-source remote sensing data collected by the data source node to obtain an evaluation result corresponding to the data source node; the evaluation indexes of the quality assessment include data integrity, noise level and data consistency; Based on each of the evaluation results, dynamically adjust the allocation weight corresponding to each of the data source nodes, so as to update the global model based on each of the allocation weights.

7. The method according to claim 6, wherein, Before the quality assessment of the initial multi-source remote sensing data collected by each data source node is performed to obtain the evaluation result corresponding to the data source node, further comprising: Based on a preset time window, determine whether there is a data source node that has not collected initial multi-source remote sensing data within the time window; the time window is a parameter representing the time range of data collection; For each data source node that does not collect initial multi-source remote sensing data within the time window, historical multi-source remote sensing data of the data source node is obtained; if there is historical multi-source remote sensing data related to the current monitoring task, first substitute data is generated based on the historical multi-source remote sensing data; if there is no historical multi-source remote sensing data related to the current monitoring task, it is determined whether there are other data source nodes associated with the data source node; if there are other data source nodes, second substitute data is generated based on the relevant data collected by the other data source nodes through data interpolation or machine learning methods; if there are no other data source nodes, the allocation weight of the data source node is set to zero and the data source node does not participate in the update of the global model; the first substitute data or the second substitute data is used as the initial multi-source remote sensing data corresponding to the data source node, and the quality of the initial multi-source remote sensing data collected by the data source node is evaluated to obtain the evaluation result corresponding to the data source node; For each data source node that collects initial multi-source remote sensing data within the time window, the quality of the initial multi-source remote sensing data collected by the data source node is evaluated to obtain the evaluation result corresponding to the data source node.

8. A device for monitoring geological deformation based on remote sensing data, characterized in that, The device is applied to the method of any one of claims 1-7, comprising: A first obtaining module is configured to obtain real-time multi-source remote sensing data corresponding to a target area of a substation, wherein the real-time multi-source remote sensing data includes optical satellite images, SAR satellite images, unmanned aerial vehicle oblique photogrammetry data, and LiDAR point cloud data. A feature extraction module is configured to perform fusion analysis based on the real-time multi-source remote sensing data and extract texture features corresponding to external damage hidden danger targets, wherein the external damage hidden danger targets include construction machinery, illegal stacking objects, temporary buildings, and vegetation intrusion areas in the target area. A second obtaining module is configured to obtain an external damage hidden danger probability distribution map corresponding to the external damage hidden danger targets based on a dynamic change model and the texture features, and obtain a human disturbance area based on the external damage hidden danger probability distribution map. A third obtaining module is configured to process the SAR satellite images based on the human disturbance area and time-series InSAR technology, and obtain ground surface deformation monitoring data of the target area, wherein the ground surface deformation monitoring data includes a deformation rate map and a cumulative deformation amount. A reference optimization module is configured to obtain Beidou positioning data based on a Beidou corner reflector, and optimize a spatial reference of the SAR satellite images based on the Beidou positioning data to obtain a three-dimensional deformation field. A report generation module is configured to construct a real scene three-dimensional base map corresponding to the substation based on the LiDAR point cloud data and the unmanned aerial vehicle oblique photogrammetry data, and generate an interactive visualization report based on the real scene three-dimensional base map, the external damage hidden danger probability distribution map, the ground surface deformation monitoring data, and the three-dimensional deformation field.

9. An electronic device, comprising: The electronic device comprises a processor and a memory, and the processor is coupled to the memory. The processor is configured to execute a computer program stored in the memory, so that the electronic device performs the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, comprising computer programs or instructions, which, when executed on a computer, cause the computer to perform the method according to any one of claims 1-7.

Images