Multi-source monitoring damage assessment method and system for composite chain disaster of power grid

Abstract

The invention provides a power grid composite chain disaster multi-source monitoring damage assessment method and system, and relates to the technical field of data processing, and the method comprises the steps: dividing a reference geographic coverage area, generating a plurality of independent analysis units, enabling each three-dimensional discrete point to belong to the corresponding independent analysis unit, and carrying out the calculation to obtain a unit space calibration coefficient; performing regional correction on the fused disaster monitoring image data set by using the unit space calibration coefficient to obtain calibrated disaster monitoring image data; based on the calibrated disaster monitoring image data, disaster damage features are extracted through a deep learning semantic segmentation network; and based on the disaster damage characteristics, carrying out spatial diffusion path simulation and time evolution process analysis of disaster influence to obtain disaster spatio-temporal evolution characteristics, and carrying out comprehensive quantitative evaluation of damage degree and risk level on power grid facilities to generate a comprehensive damage evaluation result. According to the invention, the accuracy of power grid disaster damage monitoring and the timeliness of evaluation are improved, and the risk early warning and coping capability of power grid facilities is enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a power grid composite chain disaster multi-source monitoring damage evaluation method and system. BACKGROUND

[0002] The power transmission corridor of the Jinshajiang section of the Hengduan Mountain in the Sichuan-Tibet interconnected project (east from Xiangcheng County in Ganzi, Sichuan, to Mangkang section in Changdu City, Tibet, with a total length of about 420 kilometers) is located in the hinterland of the Hengduan Mountain fault zone, with an average elevation of 3850 meters along the line, the highest tower position reaches 4980 meters, the terrain slope is mostly concentrated in 30 to 45 degrees, the valley is deeply cut, the rock mass is broken, and the loose accumulation layer is widely distributed, which is a high-risk section of the West-to-East Power Transmission backbone channel in China; Influenced by the southwest airflow and terrain uplift, short-time heavy rain occurs frequently every year from July to August, and the single rainfall intensity of some sections can reach more than 50 mm / hour, which is easy to cause shallow landslide (the thickness of landslide body is mostly 0.8 to 3 meters), which may impact the excavation foundation or pile foundation of the tower along the line, and if it causes the conductor to break or the tower to tilt, it may also contact with the flammable vegetation such as Yunnan pine and rhododendron around, which has the risk of inducing landslide and forest fire composite disaster.

[0003] At present, the disaster damage evaluation of this section mainly relies on the cooperative monitoring and data fusion technology of satellite optical image (resolution of about 1 meter) and unmanned aerial vehicle laser radar (LiDAR) point cloud (point cloud density of 20 to 30 points per square meter); but in actual application, due to multiple factors, the spatial registration and fusion of the two types of data are easy to have slight deviation: the flight attitude of unmanned aerial vehicle is easy to fluctuate in high-altitude strong wind environment, which causes slight deviation of LiDAR point cloud collection; the difference between the central projection of satellite optical image and the distance projection characteristics of LiDAR, which is easy to produce projection difference contradiction in steep slope area with slope greater than 35 degrees; combined with GNSS positioning system error (such as QX-CORS positioning method, the maximum plane error is 0.984 meters, and the maximum height error is 0.553 meters), the fusion data may have ghosting or spatial misplacement in some steep slopes and valleys, which may cause deviation in landslide body boundary interpretation, error in accurate distance measurement of tower foundation and landslide body, and further affect the accuracy of disaster damage grade determination and risk assessment. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a power grid composite chain disaster multi-source monitoring damage evaluation method and system, which can effectively resist the influence of load change and temperature drift, and ensure stable output power and high energy conversion efficiency.

[0005] To solve the above technical problems, the technical scheme of the present application is as follows: In a first aspect, a power grid composite chain disaster multi-source monitoring damage evaluation method, the method comprising: Step 1, obtaining multi-source remote sensing monitoring data; Step 2, fusing the multi-source remote sensing monitoring data to generate a fused disaster monitoring image data set, and converting it into an initial three-dimensional discrete point set; calculating the initial three-dimensional discrete point set, extracting position extreme points, determining an initial spatial convex polyhedron structure based on the position extreme points, sequentially determining the position relationship between the remaining candidate data points and the initial spatial convex polyhedron structure, generating a spatial envelope surface, and optimizing to obtain a three-dimensional discrete point set; Step 3, based on the three-dimensional discrete point set, constructing a spatial surface, and defining a reference geographic coverage area according to the distribution range of the power grid facilities; Step 4, dividing the reference geographic coverage area according to the density distribution of the three-dimensional discrete point set in the reference geographic coverage area to generate multiple independent analysis units, attributing each three-dimensional discrete point to the corresponding independent analysis unit, and calculating a unit space calibration coefficient; Step 5, using the unit space calibration coefficient to regionally correct the fused disaster monitoring image data set to obtain a calibrated disaster monitoring image data; Step 6, based on the calibrated disaster monitoring image data, extracting disaster damage features through a deep learning semantic segmentation network; Step 7, based on the disaster damage features, simulating the spatial diffusion path of the disaster impact and analyzing the time evolution process to obtain disaster spatio-temporal evolution features, and comprehensively quantifying the damage degree and risk level of the power grid facilities to generate a comprehensive damage assessment result.

[0006] In a second aspect, a power grid composite chain disaster multi-source monitoring damage assessment system includes: The acquisition module is configured to obtain multi-source remote sensing monitoring data. The calculation module is configured to fuse the multi-source remote sensing monitoring data to generate a fused disaster monitoring image data set, and convert it into an initial three-dimensional discrete point set; calculate the initial three-dimensional discrete point set, extract position extreme points, determine an initial spatial convex polyhedron structure based on the position extreme points, sequentially determine the position relationship between the remaining candidate data points and the initial spatial convex polyhedron structure, generate a spatial envelope surface, and optimize to obtain a three-dimensional discrete point set. The construction module is configured to construct a spatial surface based on the three-dimensional discrete point set, and define a reference geographic coverage area according to the distribution range of the power grid facilities. The division module is configured to divide the reference geographic coverage area according to the density distribution of the three-dimensional discrete point set in the reference geographic coverage area to generate multiple independent analysis units, attribute each three-dimensional discrete point to the corresponding independent analysis unit, and calculate a unit space calibration coefficient. The correction module is configured to perform regional correction on the fused disaster monitoring image data set by using the unit space calibration coefficient to obtain calibrated disaster monitoring image data. The extraction module is configured to extract disaster damage features by using a deep learning semantic segmentation network based on the calibrated disaster monitoring image data. The evaluation module is configured to simulate a spatial diffusion path of disaster influence and analyze a time evolution process based on the disaster damage features, obtain disaster spatio-temporal evolution features, and comprehensively quantitatively evaluate damage degrees and risk levels of power grid facilities to generate a comprehensive damage evaluation result.

[0007] In a third aspect, a computing device includes: one or more processors; a memory device storing one or more programs that, when executed by the one or more processors, cause the one or more processors to implement the method.

[0008] In a fourth aspect, a computer-readable storage medium stores a program that, when executed by a processor, implements the method.

[0009] The above scheme of the present application at least has the following beneficial effects: By constructing an initial spatial convex polyhedron by extracting position extreme points, generating a spatial envelope surface to optimize a three-dimensional discrete point set, and combining a reference geographic coverage division and regional correction of the unit space calibration coefficient, the influence caused by flight attitude fluctuation of a UAV in a high-altitude strong wind environment, differences in projection characteristics of satellite optical images and LiDAR, and GNSS positioning errors can be effectively compensated, the ghosting and spatial misplacement amplitude of fused data in steep slope and valley regions can be reduced, and the spatial registration accuracy of multi-source remote sensing data can be improved, thereby reducing landslide body boundary interpretation deviation and distance measurement error of tower foundations and landslide bodies. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 FIG. 1 is a flowchart of a power grid composite chain disaster multi-source monitoring damage evaluation method according to an embodiment of the present application; Figure 2 FIG. 2 is a schematic diagram of a power grid composite chain disaster multi-source monitoring damage evaluation system according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Exemplary embodiments of the present disclosure will be described below in greater detail with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, the embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0012] As Figure 1 shown, the embodiment of the present application proposes a power grid composite chain disaster multi-source monitoring damage evaluation method, the method comprising the following steps: Step 1, acquiring multi-source remote sensing monitoring data; Step 2, fusing the multi-source remote sensing monitoring data, generating a fused disaster monitoring image data set, and converting it into an initial three-dimensional discrete point set; calculating the initial three-dimensional discrete point set, extracting position extreme points, determining an initial spatial convex polyhedron structure based on the position extreme points, sequentially determining the position relationship between the remaining candidate data points and the initial spatial convex polyhedron structure, generating a spatial envelope surface, and optimizing to obtain a three-dimensional discrete point set; Step 3, based on the three-dimensional discrete point set, constructing a spatial surface, and defining a reference geographic coverage area according to the distribution range of the power grid facilities; Step 4, dividing the reference geographic coverage area according to the density distribution of the three-dimensional discrete point set within the reference geographic coverage area, generating a plurality of independent analysis units, and attributing each three-dimensional discrete point to the corresponding independent analysis unit, and calculating the unit space calibration coefficient; Step 5, using the unit space calibration coefficient to correct the fused disaster monitoring image data set regionally to obtain the calibrated disaster monitoring image data; Step 6, based on the calibrated disaster monitoring image data, extracting disaster damage features through a deep learning semantic segmentation network; Step 7, based on the disaster damage features, simulating the spatial diffusion path of the disaster impact and analyzing the time evolution process to obtain the disaster spatio-temporal evolution characteristics, and comprehensively quantifying the damage degree and risk level of the power grid facilities to generate a comprehensive damage evaluation result.

[0013] In the embodiment of the present application, by extracting position extreme points to construct an initial spatial convex polyhedron, generating a spatial envelope surface to optimize a three-dimensional discrete point set, combining reference geographic coverage area division and regional correction of unit space calibration coefficient, the influence of flight attitude fluctuation in unmanned aerial vehicle high-altitude strong wind environment, satellite optical image and LiDAR projection characteristic difference and GNSS positioning error can be effectively compensated, the ghosting and spatial misplacement amplitude of fused data in steep slope and valley area is reduced, and the spatial registration accuracy of multi-source remote sensing data is improved; further reducing the landslide body boundary interpretation deviation and the distance measurement error of tower foundation and landslide body.

[0014] In a preferred embodiment of the application, in step 1, the multi-source remote sensing monitoring data is obtained, specifically comprising: obtaining satellite optical image data of the Hengduan Mountain Jinsha River section of the Sichuan-Tibet interconnected power transmission corridor, the image completely covering a 420 km area from Sichuan Ganzi Xiancheng County in the east to Mangkang in Changdu City, Tibet in the west, the data containing the latitude and longitude geographic coordinate information and radiation brightness information of each pixel, the image spatial resolution being fixed at 1 meter to ensure clear capture of landslide body boundaries and tower foundation details; at the same time, a professional surveying and mapping unmanned aerial vehicle with strong wind resistance is selected, which is equipped with a laser radar detection device, and performs data collection tasks along the pre-planned polyline flight route of the power transmission corridor, the flight height is set to 150 meters from the ground, the flight speed is controlled at 8 meters per second, and the wind speed is monitored in real time during flight, when the wind speed exceeds 5 meters per second, the gimbal stability parameters are automatically adjusted to keep the unmanned aerial vehicle flight attitude stable, and ensure that the collected LiDAR point cloud data point cloud density is stable between 20 and 30 points per square meter, the point cloud data contains X-axis plane coordinates, Y-axis plane coordinates, Z-axis elevation coordinates and reflection intensity information of each detection point; before collection, the power transmission corridor is divided into 84 5 km long collection sections, each section covers the tower base periphery 50 meters, the power transmission line both sides 30 meters and the historical landslide high incidence area, the flight trajectory is calibrated in real time through the satellite positioning system to ensure that there is no data collection omission area.

[0015] In this embodiment, by obtaining two types of multi-source remote sensing monitoring data suitable for the characteristics of high-risk areas of the power transmission corridor, the disaster-prone areas are fully covered, and complete and actual scene-matching basic data support is provided for subsequent disaster monitoring and damage assessment.

[0016] In a preferred embodiment of the application, in step 2, the multi-source remote sensing monitoring data is fused and processed to generate a fused disaster monitoring image data set, and is converted into an initial three-dimensional discrete point set; the initial three-dimensional discrete point set is calculated, the position extreme point is extracted, the initial spatial convex polyhedron structure is determined based on the position extreme point, the remaining candidate data points are sequentially subjected to position relationship determination with the initial spatial convex polyhedron structure, a spatial envelope surface is generated, and a three-dimensional discrete point set is obtained by optimization, which can include: Step 201, spatial registration and radiation normalization processing is performed on the multi-source remote sensing monitoring data to obtain a registered and normalized multi-source remote sensing monitoring data set, which specifically includes: taking satellite optical images as the reference data, selecting 20 evenly distributed fixed ground objects along the power transmission corridor as the same name control points, the fixed ground objects are specifically road intersection corners, bridge end points, tower base corners, the straight line distance between each control point is not more than 25 kilometers, the X-axis plane coordinate, Y-axis plane coordinate and Z-axis elevation value of each control point in the satellite optical image are read respectively, the X-axis plane coordinate, Y-axis plane coordinate and Z-axis elevation value of the control point in the LiDAR point cloud data are read, the X-axis coordinate deviation of each control point is obtained by subtracting the X-axis plane coordinate in the satellite optical image from the X-axis plane coordinate in the LiDAR point cloud data, the Y-axis coordinate deviation of each control point is obtained by subtracting the Y-axis plane coordinate in the satellite optical image from the Y-axis plane coordinate in the LiDAR point cloud data, the X-axis average deviation is obtained by adding the X-axis coordinate deviations of the 20 control points and dividing by 20, the preliminary corrected X-axis plane coordinate is obtained by subtracting the X-axis average deviation from the original X-axis plane coordinate of each LiDAR point cloud data point, the corrected X-axis coordinate deviation of each control point is recalculated, if there is a control point with a deviation greater than 0.1 meters, the secondary correction deviation is obtained by adding the over-standard X-axis deviations and dividing by the number of over-standard control points, the preliminary corrected X-axis plane coordinate is subtracted by the secondary correction deviation, the deviation statistics and coordinate adjustment operation are repeatedly performed until the X-axis coordinate deviations of all control points are less than 0.1 meters; the Y-axis plane coordinate is processed according to the same process, the Y-axis average deviation of the 20 control points is calculated first, the preliminary correction value is obtained by subtracting the Y-axis average deviation from the original Y-axis plane coordinate of the LiDAR point cloud data point, the secondary correction deviation is calculated according to the over-standard deviation and adjusted until the Y-axis coordinate deviations of all control points are less than 0.1 meters.

[0017] For the height coordinate, taking the height information of the control points in the satellite optical image as the reference, the reference Z-axis height value of each control point in the satellite optical image is read, the Z-axis height value of the control point in the LiDAR point cloud data is read, the Z-axis height value in the LiDAR point cloud data is subtracted from the reference Z-axis height value in the satellite optical image to obtain the Z-axis height deviation of each control point, the Z-axis height deviations of all control points are added and then divided by 20 to obtain the average Z-axis deviation, the original Z-axis height value of each LiDAR point cloud data point is subtracted by the average Z-axis deviation to obtain the corrected Z-axis height value, and the accurate alignment of the two types of data in the height direction is realized; the radiation normalization processing selects three uniform reflecting surfaces with an area of 100 square meters each along the power transmission corridor as the radiation calibration region, which are the exposed granite region located in the Xiangcheng County section, the cement maintenance road surface in the Mangkang section, and the sandstone construction site in the middle section, the radiation brightness values of all pixels in the satellite optical image in each calibration region are extracted, the radiation brightness values of all pixels in the region are added and then divided by the total number of pixels to obtain the average radiation brightness value of the satellite optical image, the reflection intensity values of all points in the LiDAR point cloud data in each calibration region are extracted, the reflection intensity values of all points in the region are added and then divided by the total number of points to obtain the average reflection intensity value of the LiDAR point cloud data, the average radiation brightness value of the satellite optical image in each calibration region is divided by the average reflection intensity value of the LiDAR point cloud in the region to obtain three radiation calibration coefficients, the three radiation calibration coefficients are added and then divided by 3 to obtain the average radiation calibration coefficient, and the reflection intensity value of each point in the LiDAR point cloud data is multiplied by the average radiation calibration coefficient to obtain the radiation data with the same dimension as the radiation brightness of the satellite optical image, and the radiation normalization processing of the two types of data is completed.

[0018] Step 202, the registered and normalized multi-source remote sensing monitoring data set is fused to generate a fused disaster monitoring image data set; the fused disaster monitoring image data set is converted into an initial three-dimensional discrete point set, specifically including: adopting a weighted average fusion method to point-by-point fuse the registered and normalized satellite optical image data and LiDAR radiation data, the weight value range of the satellite optical image data is set to be between 0.3 and 0.5, and 0.4 is selected this time; the weight value range of the LiDAR radiation data is between 0.5 and 0.7, and 0.6 is selected this time, the same spatial position is determined through coordinate matching, that is, the X-axis and Y-axis coordinates of a certain pixel in the satellite optical image and the X-axis and Y-axis coordinates of a certain point in the LiDAR point cloud data are less than 0.3 meters, which is regarded as the same spatial position, the satellite optical image radiation brightness value of the spatial position is multiplied by 0.4, and the LiDAR radiation data value of the spatial position is multiplied by 0.6, the two product results are added to obtain the fusion radiation value of the spatial position, the X-axis plane coordinate, the Y-axis plane coordinate and the Z-axis elevation coordinate of the spatial position are retained, the fusion radiation value and the three-dimensional coordinate information are recorded one by one according to the spatial position, and the fused disaster monitoring image data set covering the entire monitoring area is formed; information is extracted from the fused disaster monitoring image data set point by point, each data point corresponds to a complete set of information, including X-axis coordinate, Y-axis coordinate and Z-axis coordinate accurate to centimeter level and fusion radiation value accurate to three decimal places, all data points are arranged in order from small to large according to the X-axis coordinate, each data point is independently numbered, and all numbered data points are combined to form an initial three-dimensional discrete point set.

[0019] Step 203, based on the initial three-dimensional discrete point set, the position extreme points are extracted along the three-dimensional coordinate axis direction; based on the position extreme points, the initial space convex polyhedral structure is constructed; based on the initial three-dimensional discrete point set and the position extreme points, the candidate data point set is determined, specifically including: traversing all data points in the initial three-dimensional discrete point set in the order of numbering, comparing the X-axis coordinate values of all data points, finding out the data point with the maximum X-axis coordinate value and the data point with the minimum X-axis coordinate value, similarly comparing the Y-axis coordinate values of all data points, finding out the data point with the maximum Y-axis coordinate value and the data point with the minimum Y-axis coordinate value, comparing the Z-axis coordinate values of all data points, finding out the data point with the maximum Z-axis coordinate value and the data point with the minimum Z-axis coordinate value, these six data points are the basic extreme points; again traversing all data points in the initial three-dimensional discrete point set in the order of numbering, for each coordinate axis direction, the deviation proportion of the data point and the corresponding basic extreme point is calculated, taking the X-axis as an example, the absolute value of the X-axis coordinate of the data point minus the X-axis coordinate of the X-axis maximum basic extreme point is divided by the X-axis coordinate of the X-axis maximum basic extreme point, to obtain the deviation proportion of the data point relative to the X-axis maximum basic extreme point, similarly, the deviation proportion relative to the X-axis minimum basic extreme point is calculated, the Y-axis and the Z-axis are calculated in the same way, the data points with the deviation proportion of all coordinate axes within 5% are determined as the secondary extreme points, the six basic extreme points and the six secondary extreme points selected are integrated to form a position extreme point set containing 12 data points; the 12 position extreme points are sorted in the order of the sum of X-axis Y-axis Z-axis coordinates from small to large, the adjacent two extreme points are connected in turn according to the sorting result, finally the extreme point at the last position of the sorting is connected with the extreme point at the first position of the sorting, forming a closed polyhedral structure, the closed structure is the initial space convex polyhedron; all the 12 data points belonging to the position extreme point set are deleted from the initial three-dimensional discrete point set one by one, and the remaining all data points are reserved in the original numbering order to jointly constitute the candidate data point set.

[0020] At step 204, based on the candidate data point set and the initial spatial convex polyhedral structure, a spatial envelope surface is generated by point-by-point position judgment and structure updating; based on the spatial envelope surface, the initial three-dimensional discrete point set is screened and optimized to obtain a three-dimensional discrete point set, specifically including: according to the original number order of the candidate data point set, the position of each data point is judged in turn, and the ray method is used to execute the judgment operation; a virtual ray with a direction vector of 100 is emitted from the current candidate data point, the ray is parallel to the positive direction of the X axis and extends infinitely, the intersection of the ray with each surface of the initial spatial convex polyhedron is judged one by one, if the ray passes through a certain surface, it is counted once, and finally the total intersection point number of the ray and the surface of the initial spatial convex polyhedron is obtained; if the total intersection point number is odd, it means that the candidate data point is located inside the initial spatial convex polyhedron and does not participate in the spatial convex polyhedron structure updating; if the total intersection point number is even or zero, it means that the candidate data point is located outside the initial spatial convex polyhedron, the data point is added to the vertex set of the initial spatial convex polyhedron, the spatial distance between all vertices is recalculated, the edges with a distance greater than 1.5 times the average distance of adjacent vertices are deleted, and the adjacent vertices are reconnected to form a new closed polyhedral structure.

[0021] The above point-by-point judgment and structure updating operation is repeatedly executed until all data points in the candidate data point set complete the position judgment and no new vertex is added, and the outer surface of the finally formed closed convex polyhedron is the spatial envelope surface; the initial three-dimensional discrete point set is screened based on the spatial envelope surface, the vertical distance of each data point in the initial three-dimensional discrete point set to the nearest surface of the spatial envelope surface is calculated, the internal redundant data points with a vertical distance greater than 0.2 meters are removed, and all vertices on the spatial envelope surface and the effective data points with a vertical distance less than or equal to 0.2 meters are retained; the retained effective data points are de-duplicated, and the X axis, Y axis and Z axis coordinates of each data point are compared one by one; if the X axis coordinate difference, Y axis coordinate difference and Z axis coordinate difference of two data points are all less than 0.01 meters, it is determined that the two data points are duplicate data points, the data point with a smaller number is retained, and the data point with a larger number is deleted, and finally an optimized three-dimensional discrete point set is obtained In this embodiment, the spatial envelope surface is formed by point-by-point judgment and updating, redundant data is effectively removed, and the optimized three-dimensional discrete point set can accurately reflect the spatial profile of the monitoring area, thereby improving the accuracy of subsequent disaster damage assessment.

[0022] In a preferred embodiment of the application, step 3, based on the three-dimensional discrete point set, a spatial curved surface is constructed, and a reference geographic coverage area is defined according to the distribution range of the power grid facility, which can include: Step 301, based on the three-dimensional discrete point set, a continuous spatial surface is calculated and generated; based on the spatial surface, a meshing process is performed to obtain a spatial surface mesh; based on the known geographic distribution range data of the power grid facility, a two-dimensional polygon is generated, the two-dimensional polygon is extended to three-dimensional space and processed to obtain a facility distribution surface mesh, which specifically includes: based on the optimized three-dimensional discrete point set, a distance weighted interpolation method is used to generate a continuous spatial surface, the interpolation grid spacing is first set to 0.5 meters, a uniform interpolation point array is constructed in the monitoring area, for each interpolation point, search for three-dimensional discrete points within a 5-meter range around it, assign weights according to the principle that the closer the distance, the greater the weight, the discrete points within 1 meter of the interpolation point have a weight value range of 0.3 to 0.5, the weight value range decreases by 0.05 to 0.1 for every 1 meter increase, and the discrete points at a distance of 5 meters have a weight value range of 0.05 to 0.1. Multiply the Z-axis elevation value of each peripheral discrete point by the corresponding weight, and add all the product results to obtain the elevation value of the interpolation point. The elevation values of all interpolation points are spliced to form a continuous spatial surface that completely covers the monitoring area and fits the steep slope and valley terrain of the Sichuan-Tibet interconnected power transmission corridor.

[0023] The known geographic distribution range data of the power grid facility is collected, including the center coordinates of the bottom base of all towers, the direction coordinates of the transmission line, and the tower spacing data. Taking the center line of the transmission corridor as the reference, each side is expanded by 50 meters as the facility protection range, combined with the distribution boundary of the tower along the line, the closed two-dimensional polygon is outlined, and the planar range of the polygon completely covers all power grid facilities and the surrounding potential disaster affected areas. The two-dimensional polygon is expanded along the elevation direction, and the expanded elevation range extends from the lowest elevation of 3200 meters along the transmission corridor to the highest elevation of 5100 meters, forming a three-dimensional column structure. The outer surface of the three-dimensional column structure is divided into quadrilaterals with a side length of 0.5 meters, ensuring that the grid density after division is consistent with the spatial surface mesh, and each grid unit is a regular quadrilateral facet. Finally, the facility distribution surface mesh is obtained.

[0024] At step 302, based on the space surface grid and the facility distribution surface grid, a set of intersected triangular face elements is calculated; based on the set of intersected triangular face elements, a minimum circumscribed rectangular region is calculated, specifically including: extracting each quadrilateral face element in the space surface grid one by one, recording the complete three-dimensional coordinates of the four vertices of the face element, i.e., the X-axis plane coordinate, the Y-axis plane coordinate and the Z-axis elevation value of each vertex, and at the same time, extracting each quadrilateral face element in the facility distribution surface grid in the same way, and recording the three-dimensional coordinates of the four vertices completely; pairing and comparing the two groups of face elements one by one, each pair of comparison objects being a space surface quadrilateral face element and a facility distribution surface quadrilateral face element, judging whether there is a spatial intersection between the two, and the judgment process being divided into two steps, the first step being to check the vertex attribution, comparing the four vertices of the space surface face element with the coordinate range of the facility distribution surface face element respectively, calculating the X-axis maximum coordinate, the X-axis minimum coordinate, the Y-axis maximum coordinate, the Y-axis minimum coordinate, the Z-axis maximum coordinate and the Z-axis minimum coordinate of the four vertices of the facility distribution surface face element, and if the X-axis coordinate of a vertex of the space surface face element is between the X-axis maximum and minimum coordinates of the facility distribution surface face element, the Y-axis coordinate is between the corresponding Y-axis maximum and minimum coordinates, and the Z-axis coordinate is between the corresponding Z-axis maximum and minimum coordinates, then it is determined that the vertex falls within another face element, and there is an intersection between the two.

[0025] If the first step does not determine the intersection, the second step of edge intersection checking is performed, four edges of the space surface panel are extracted, each edge corresponds to the three-dimensional coordinates of two vertices, and four edges of the facility distribution surface panel are also extracted. Each edge of the space surface panel is paired with each edge of the facility distribution surface panel, the X-axis coordinate variation range and the Y-axis coordinate variation range of each edge are calculated, if the X-axis range of the two edges overlaps and the Y-axis range overlaps, the intersection point of the projection of the two edges in the same X-Y plane is calculated, if the intersection point falls within the range of the line segment of the two edges, it is determined that the two panels have an intersection; all space surface panels and facility distribution surface panels that have been determined to have an intersection through the above two steps are screened out to form an intersection panel set; all vertices of all panels in the intersection panel set are extracted, and the X-axis coordinate list and the Y-axis coordinate list containing all vertices are obtained by sorting; all values in the X-axis coordinate list are compared one by one, the first value in the list is selected as the temporary maximum value and the temporary minimum value, and then each subsequent value in the list is compared with the temporary maximum value and the temporary minimum value, if the value is greater than the temporary maximum value, the temporary maximum value is updated, if the value is less than the temporary minimum value, the temporary minimum value is updated, after all values are traversed, the final X-axis maximum value and the X-axis minimum value are obtained; the Y-axis coordinate list is traversed in the same way to obtain the final Y-axis maximum value and the Y-axis minimum value; a closed rectangular area is constructed by connecting four vertices in clockwise order, the four vertices are (X-axis minimum value, Y-axis minimum value), (X-axis minimum value, Y-axis maximum value), (X-axis maximum value, Y-axis maximum value), and (X-axis maximum value, Y-axis minimum value), the rectangular area is the minimum circumscribed rectangular area of the intersection panel set.

[0026] In step 303, the reference geographic coverage area is obtained by projection mapping based on the minimum circumscribed rectangular region and the spatial curved surface, and specifically includes: determining that the projection direction is the negative direction of the Z-axis perpendicular to the ground, that is, the projection line is perpendicular to the ground and points to the ground; the boundaries of the minimum circumscribed rectangular region are equally divided at intervals of 0.5 meters to obtain a plurality of boundary division points, and the X-axis and Y-axis coordinates of each division point are recorded, and the grid nodes uniformly distributed in the rectangular region at intervals of 0.5 meters are also retained, and the X-axis and Y-axis coordinates of each node are also recorded; performing a projection operation on each division point and node, keeping the X-axis and Y-axis coordinates unchanged, finding the Z-axis elevation value corresponding to the same X-axis and Y-axis coordinates on the continuous spatial curved surface, and assigning the elevation value to the corresponding division point or node to form a projected spatial point; extracting the three-dimensional coordinates of all projected boundary division points, arranging the projected boundary sequence in the order of the original rectangular boundary, and combining the steep slope and valley topographic features of the Sichuan-Tibet network project transmission corridor to smooth the projected boundary. For each boundary point, select the two adjacent boundary points before and after it, add the Z-axis elevation values of the five points and divide by 5 to obtain the smoothed elevation value, replace the original boundary point elevation value with the smoothed elevation value, and sequentially process all boundary points to eliminate boundary sharp corners and breakpoints caused by terrain changes and ensure smooth boundary continuity. The last boundary point after smoothing is connected to the first boundary point according to the terrain trend to form a complete closed region, and the projected nodes inside the region are supplemented to be complete in terms of spatial position. The region not only completely covers the overlapping range of the power grid facility and the disaster monitoring region, but also completely matches the actual topographic features of the curved surface, that is, the reference geographic coverage area.

[0027] In this embodiment, by screening intersecting surface elements and constructing a minimum circumscribed rectangle, the overlapping core range of the power grid facility and the disaster area is focused, and the data interference of irrelevant areas is reduced.

[0028] In a preferred embodiment of the present application, in step 4, the reference geographic coverage area is divided according to the density distribution of the three-dimensional discrete point set in the reference geographic coverage area to generate a plurality of independent analysis units, and each three-dimensional discrete point is attributed to the corresponding independent analysis unit, and the unit space calibration coefficient is calculated, which can include: Step 401, based on the reference geographic coverage area and the three-dimensional discrete point set, the spatial density distribution of the three-dimensional discrete point set in the reference geographic coverage area is calculated and generated, specifically including: taking the planar range of the reference geographic coverage area as a statistical area, first determining the X-axis minimum coordinate, the X-axis maximum coordinate, the Y-axis minimum coordinate and the Y-axis maximum coordinate of the area, taking the X-axis minimum coordinate and the Y-axis minimum coordinate as the starting point, and dividing the statistical grid uniformly along the X-axis and Y-axis direction in turn according to the size of 10m x 10m, the X-axis range of each grid is the starting X-coordinate plus the grid serial number multiplied by 10m to the starting X-coordinate plus (grid serial number plus 1) multiplied by 10m, and the Y-axis range is calculated in the same way, the whole reference geographic coverage area is divided into 840 statistical grids, and the area of each grid is fixed at 100 square meters; read the X-axis plane coordinate and Y-axis plane coordinate of each data point in the three-dimensional discrete point set one by one, judge whether the data point falls in a certain statistical grid, the judgment standard is that the X-axis coordinate of the data point is greater than or equal to the X-axis starting coordinate of the grid and less than the X-axis ending coordinate of the grid, and the Y-axis coordinate is greater than or equal to the Y-axis starting coordinate of the grid and less than the Y-axis ending coordinate of the grid, count the total number of data points meeting the conditions in each grid; divide the total number of discrete points in each grid by the grid area of 100 square meters to get the spatial density of each statistical grid, that is, the number of discrete points per unit area; sort all the spatial density values of the statistical grids according to the density size, and clearly define the specific distribution range of the high-density area (spatial density greater than or equal to 3 points per square meter), the medium-density area (spatial density greater than or equal to 1 point per square meter and less than 3 points per square meter) and the low-density area (spatial density less than 1 point per square meter), mark that the high-density area is concentrated around the steep slope and valley and near the tower base, and the low-density area is mostly open and flat land.

[0029] At step 402, the reference geographic coverage area is divided based on the spatial density distribution to generate a plurality of independent analysis units; based on the three-dimensional discrete point set and the independent analysis units, each three-dimensional discrete point is attributed to a corresponding independent analysis unit through spatial position matching, and a three-dimensional discrete point subset is extracted, specifically including: dividing the independent analysis units by using an adaptive density division method according to the spatial density distribution characteristics, first determining a density division threshold, a spatial density greater than or equal to 3 points per square meter is a high-density area, the plane side length of the independent analysis unit in this area is in a range of 4 meters to 6 meters, and is set to 5 meters in this case, so as to ensure that each unit contains at least 125 data points to support subsequent statistical calculation; a spatial density greater than or equal to 1 point per square meter and less than 3 points per square meter is a medium-density area, and the plane side length of the unit is in a range of 8 meters to 12 meters, and is set to 10 meters in this case; a spatial density less than 1 point per square meter is a low-density area, and the plane side length of the unit is in a range of 18 meters to 22 meters, and is set to 20 meters in this case, so as to avoid insufficient data representativeness due to excessively large units; when dividing, the minimum coordinate of the X-axis and the minimum coordinate of the Y-axis of the reference geographic coverage area are taken as the starting point, the high-density area is divided into rectangular units along the X-axis and the Y-axis at a step length of 5 meters, the medium-density area is divided at a step length of 10 meters, and the low-density area is divided at a step length of 20 meters, so as to ensure that all units do not overlap and completely cover the reference geographic coverage area, and the plane coordinate range (minimum X-axis, maximum X-axis, minimum Y-axis, maximum Y-axis) and the elevation range (minimum elevation to maximum elevation of all potential data points in the unit) of each unit are recorded; after the division is completed, unit verification is performed, if the number of expected data points in a unit is less than 30, the unit is combined with an adjacent unit of the same density type, and the coordinate range of the combined unit is re-determined; the X-axis and Y-axis plane coordinates of each data point in the three-dimensional discrete point set are read one by one, the independent analysis unit to which the coordinates belong is judged, if the X-axis coordinate of the data point is between the minimum and maximum values of the X-axis of a unit, and the Y-axis coordinate is between the minimum and maximum values of the Y-axis of the unit, the data point is attributed to the unit, if the data point falls on the boundary line of two units, the attribution is performed according to the X-axis coordinate priority principle, that is, the X-axis coordinate closer to the center of which unit belongs to which unit; according to the division of the independent analysis units, all three-dimensional discrete points attributed to the same unit are sorted to form a three-dimensional discrete point subset corresponding to each unit In step 403, based on the three-dimensional discrete point subset, the geometric feature statistics and the spectral reflection feature statistics of the three-dimensional discrete point subset are calculated; based on the geometric feature statistics and the spectral reflection feature statistics, the unit space calibration coefficient is generated through weighted fusion calculation, specifically including: calculating the geometric feature statistics of the three-dimensional discrete point subset, first calculating the elevation mean, adding all the Z-axis elevation values of all data points in the point subset to obtain the elevation sum, dividing the elevation sum by the number of data points in the subset to obtain the elevation mean of the unit; calculating the elevation standard deviation, first taking out the elevation mean of the subset, subtracting the elevation mean from the Z-axis elevation value of each data point to obtain the elevation deviation of each data point, performing square operation on all elevation deviations respectively to obtain the square value of the deviation, adding all the square values of the deviation to obtain the square sum, dividing the square sum by the number of data points in the subset to obtain the square mean of the deviation, performing arithmetic square root operation on the square mean of the deviation, which is the elevation standard deviation; calculating the mean value of the facet slope, extracting all the quadrilateral facets corresponding to the point subset, for each facet, first reading the three-dimensional coordinates of its four vertices, calculating the horizontal projection area and the vertical height difference of the facet, obtaining the slope value (unit: degree) of the facet by dividing the vertical height difference by the horizontal projection distance, adding the slope values of all facets to obtain the slope total, dividing the slope total by the total number of facets to obtain the mean value of the facet slope.

[0030] The facet roughness is calculated as follows: first, the average slope of the subset is calculated, and the slope deviation of each facet is obtained by subtracting the average slope from the slope of each facet. The sum of the squares of all slope deviations is then calculated, and the average of the squared deviations is obtained by dividing the sum by the total number of facets. The average is then square rooted to obtain the facet roughness. The spectral reflectance characteristic statistics are calculated as follows: first, the average radiance is calculated by summing the fusion radiance values of all data points in the subset and dividing the sum by the number of data points. The standard deviation of the radiance is then calculated by subtracting the average radiance from the fusion radiance value of each data point, summing the squares of all deviations, dividing the sum by the number of data points, and square rooting the average. The range of the radiance is obtained by comparing the fusion radiance values of all data points in the subset and subtracting the minimum value from the maximum value. The geometric feature statistics and spectral reflectance characteristic statistics are normalized as follows: for each statistic type (e.g., the average elevation of all units), the maximum and minimum values of the statistic in all independent analysis units are found, and the normalized value of the statistic is obtained by subtracting the minimum value from the statistic value of the unit and dividing the result by the difference between the maximum and minimum values. All normalized values are adjusted to be between 0 and 1. The total weight of the geometric feature statistics is set to be in the range of 0.5 to 0.7, and 0.6 is selected in this case. The weight of the average elevation is in the range of 0.15 to 0.25, and 0.2 is selected in this case. The weight of the standard deviation of the elevation is in the range of 0.15 to 0.25, and 0.2 is selected in this case. The weight of the average slope of the facet is in the range of 0.25 to 0.35, and 0.3 is selected in this case. The weight of the facet roughness is in the range of 0.25 to 0.35, and 0.3 is selected in this case.

[0031] The normalized value of each geometric feature statistic is multiplied by the corresponding weight to obtain the geometric weighted value of each statistic, all geometric weighted values are added to obtain the geometric feature weighted sum; the total weight value range of the spectral reflectance feature statistic is set to 0.3 to 0.5, and 0.4 is selected this time, wherein the weight value range of the radiation brightness mean is 0.35 to 0.45, and 0.4 is selected this time, the weight value range of the radiation brightness standard deviation is 0.25 to 0.35, and 0.3 is selected this time, the weight value range of the radiation brightness range is 0.25 to 0.35, and 0.3 is selected this time; the normalized value of each spectral reflectance feature statistic is multiplied by the corresponding weight to obtain the spectral weighted value of each statistic, all spectral weighted values are added to obtain the spectral feature weighted sum; the geometric feature weighted sum is multiplied by 0.6 to obtain the geometric feature contribution value, the spectral feature weighted sum is multiplied by 0.4 to obtain the spectral feature contribution value, and the two contribution values are added to obtain the unit space calibration coefficient of the independent analysis unit, and the coefficient value range is 0 to 1.

[0032] In this embodiment, by comprehensively calculating the geometric feature and the spectral reflectance feature statistic, and combining the unit space calibration coefficient obtained by weighted fusion, the data quality of each unit can be accurately reflected.

[0033] In a preferred embodiment of the present application, in step 5, the unit space calibration coefficient is used for regional correction of the fused disaster monitoring image data set to obtain the calibrated disaster monitoring image data, which can include: Step 501, based on the fusion disaster monitoring image data set and the unit space calibration coefficient, the spatial range of each independent analysis unit and the corresponding unit space calibration coefficient are established. The registration relationship includes: traversing the X-axis coordinates of all three-dimensional discrete points in the unit, comparing all X-axis coordinate values, finding the minimum value as the X-axis minimum coordinate, and finding the maximum value as the X-axis maximum coordinate. The coordinate value is accurate to the centimeter level (retaining two decimal places); traverse all data points in the unit Y-axis coordinates in the same way, determine the Y-axis minimum coordinate and Y-axis maximum coordinate, also accurate to the centimeter level; traverse all data points in the unit Z-axis elevation value, extract the minimum value as the Z-axis minimum elevation, and extract the maximum value as the Z-axis maximum elevation. The elevation value is accurate to the centimeter level, and the six parameters are recorded as the spatial range information of the unit; the identification number of each unit is bound with the corresponding unit space calibration coefficient, and the above six spatial range parameters are associated to form a registration relationship list containing eight core contents of unit number, X-axis minimum coordinate, X-axis maximum coordinate, Y-axis minimum coordinate, Y-axis maximum coordinate, Z-axis minimum elevation, Z-axis maximum elevation, and calibration coefficient. Each record in the list uniquely corresponds to an independent analysis unit; the registration relationship list is double-checked, the first check is the spatial range, and the coordinate comparison tool is used to check each unit one by one. Taking unit A and unit B as an example, first get the X-axis maximum coordinate, X-axis minimum coordinate, Y-axis maximum coordinate, and Y-axis minimum coordinate of unit A, then get the corresponding coordinate parameters of unit B, judge whether the X-axis maximum coordinate of unit A is greater than the X-axis minimum coordinate of unit B, and whether the X-axis minimum coordinate of unit A is less than the X-axis maximum coordinate of unit B, and whether the Y-axis maximum coordinate of unit A is greater than the Y-axis minimum coordinate of unit B, and whether the Y-axis minimum coordinate of unit A is less than the Y-axis maximum coordinate of unit B. If the four conditions are met at the same time, it is determined that the spatial range of the two units overlaps, and the X-axis center line and Y-axis center line of the overlapping region of the two units are taken as the boundary, the X-axis and Y-axis coordinate range of the two units is adjusted, and then the adjustment is checked again until the spatial range of all units does not overlap; the second check is the calibration coefficient, which is checked one by one to confirm whether the calibration coefficient value is between 0 and 1 and has no missing value. If there is a value outside the range or missing record, return to recalculate the geometric feature statistics and spectral reflectance feature statistics of the unit, and then regenerate the unit space calibration coefficient; after the double check, the registration relationship list is stored in the local database in the order of unit number from small to large, and an index is established with the unit number as the key, which is convenient for subsequent data extraction and quick query of the corresponding six spatial range parameters and calibration coefficient through the number.

[0034] At step 502, based on the registration relationship, the original data block in the spatial range of each independent analysis unit is extracted from the fused disaster monitoring image data set, specifically including: according to the unit spatial range parameters recorded in the registration relationship list, locking the geographical boundary of each independent analysis unit one by one, and determining the effective range of the X-axis minimum coordinate, the X-axis maximum coordinate, the Y-axis minimum coordinate, the Y-axis maximum coordinate, the Z-axis minimum elevation, and the Z-axis maximum elevation of the unit; filtering point by point from the fused disaster monitoring image data set according to the data point number, first judging whether the X-axis coordinate of the data point is greater than or equal to the X-axis minimum coordinate of the unit and less than or equal to the X-axis maximum coordinate, then judging whether the Y-axis coordinate is greater than or equal to the Y-axis minimum coordinate of the unit and less than or equal to the Y-axis maximum coordinate, and finally judging whether the Z-axis elevation is greater than or equal to the Z-axis minimum elevation of the unit and less than or equal to the Z-axis maximum elevation, all three conditions are met, then the data point is retained; arranging all the filtered data points in order of X-axis coordinate from small to large and Y-axis coordinate from small to large, each data point retaining complete three-dimensional coordinates (accurate to centimeter level) and fused radiation value (accurate to three decimal places), forming the original data block corresponding to the independent analysis unit.

[0035] At step 503, based on the original data block, the unit space calibration coefficient corresponding to the original data block is determined according to the registration relationship, and the unit space calibration coefficient is used to perform geometric position correction and radiation value correction to obtain a correction result, which specifically includes: through the unit number marked by the original data block, the corresponding unit space calibration coefficient is quickly queried in the index of the registration relationship list; geometric position correction is performed, the original data block is uniformly divided into 10 small areas in the X-axis and Y-axis directions, 1 center data point in each small area is selected as a reference point, 10 uniformly distributed reference points are selected in total, corresponding coordinates (i.e. reference data) of these reference points in the satellite optical image are extracted, X-axis coordinate deviation (original X-axis coordinate minus reference X-axis coordinate), Y-axis coordinate deviation (original Y-axis coordinate minus reference Y-axis coordinate), and Z-axis elevation deviation (original Z-axis elevation minus reference Z-axis elevation) of each reference point are calculated, the X-axis coordinate deviation of the 10 reference points is added and divided by 10 to obtain the X-axis average coordinate deviation, the Y-axis average coordinate deviation and the Z-axis average elevation deviation are calculated in the same way; the unit space calibration coefficient is multiplied by the X-axis average coordinate deviation, the Y-axis average coordinate deviation, and the Z-axis average elevation deviation respectively to obtain the X-axis coordinate correction amount, the Y-axis coordinate correction amount, and the Z-axis elevation correction amount; the original X-axis coordinate of each data point is added by the X-axis coordinate correction amount, the original Y-axis coordinate is added by the Y-axis coordinate correction amount, and the original Z-axis elevation is added by the Z-axis elevation correction amount to complete the geometric position correction; radiation value correction is performed, the fusion radiation brightness value of all data points in the original data block is added and divided by the total number of data points to obtain the average radiation brightness value; the unit space calibration coefficient is subtracted by 1 to obtain the radiation correction proportion, and the average radiation brightness value is multiplied by the radiation correction proportion to obtain the radiation correction amount; the original radiation brightness value of each data point is added by the radiation correction amount to obtain the corrected radiation brightness value, which ensures that the radiation value conforms to the actual ground feature; the corrected three-dimensional coordinates of each data point and the corrected radiation brightness value are corresponded one by one to form the correction result corresponding to each original data block.

[0036] At step 504, based on the correction result, a regional corrected data block is generated; based on the regional corrected data block, spatial seamless splicing and data fusion are performed to obtain calibrated disaster monitoring image data, specifically including: performing outlier rejection on the correction result of each original data block, setting a reasonable range of radiation brightness value as 0 to 1000 (determined according to the statistical characteristics of the surface radiation in the Sichuan-Tibet region), determining that the coordinate deviating from the single unit space range by more than 0.5 meters as an abnormal data point, checking the radiation brightness value and coordinate of each data point one by one, after removing the abnormal data points, the remaining effective data points are sorted in the order of X-axis, Y-axis and Z-axis coordinates to form a regional corrected data block; for the corrected data blocks of adjacent independent analysis units, the data in the overlapping region is extracted, the width of the overlapping region is fixed as 1 meter, a grid is established in the overlapping region at an interval of 0.1 meters in X-Y coordinates, the corresponding data points in the same grid are matched, the corrected radiation brightness values of all the matched data points are added and then divided by the number of data points to obtain the radiation brightness average value of the overlapping region, and the radiation brightness values of the original data points in the overlapping region of the two blocks are replaced by the average value to eliminate the radiation difference of the unit boundary; in the geographical order from west to east and from north to south according to the reference geographical coverage area, starting from the starting boundary with the minimum X-axis and Y-axis, all the regional corrected data blocks are spliced in the spatial position in turn, and the coordinates of the boundary data points of adjacent blocks are checked one by one during splicing to ensure that the coordinates are continuous without breakpoints and overlapping deviation; the overall data after splicing is re-combed according to a spatial grid of 0.5m x 0.5m, the X-axis coordinate of the center point of each grid is the minimum coordinate of the grid X-axis plus 0.25m, the Y-axis center point coordinate is the same, and the Z-axis coordinate is the elevation average value of all data points in the grid, if there are multiple data points in the grid, the radiation brightness average value of these data points is taken as the radiation brightness value of the grid, and finally the calibrated disaster monitoring image data covering the entire monitoring area, spatially continuous and uniformly radiated is formed.

[0037] In this embodiment, by establishing an accurate registration relationship and regional correction, the spatial deviation and radiation difference after multi-source data fusion are effectively corrected, and the accuracy and consistency of the disaster monitoring image data are improved.

[0038] In a preferred embodiment of the present application, step 6, based on the calibrated disaster monitoring image data, the disaster damage features are extracted through a deep learning semantic segmentation network, which can include: At step 601, based on the calibrated disaster monitoring image data, image slicing and standardization processing are performed to generate standardized sample data; based on the standardized sample data, multi-level spatial spectral features are extracted through a multi-level convolution structure of a deep learning semantic segmentation network to obtain a high-level semantic feature set, specifically including: performing image slicing on the calibrated disaster monitoring image data, setting the slicing size to 512 pixels x 512 pixels, starting from the top left corner of the image, sliding and slicing along the X-axis and Y-axis directions at a step size of 412 pixels (with an overlapping area of 100 pixels between adjacent slices) to avoid loss of edge disaster features, and after slicing, numbering according to the rules of region number, row number and column number to ensure that the geographical position of each slice is traceable; performing standardization processing on each slice, first adding the radiation brightness values of all pixels in the slice and then dividing by the total number of pixels (512 x 512 = 262144 pixels) to obtain the radiation brightness mean value of the slice; then subtracting the radiation brightness mean value from the radiation brightness value of each pixel to obtain the radiation brightness deviation of each pixel, squaring and adding the radiation brightness deviations of all pixels to obtain the sum of squared deviations; dividing the sum of squared deviations by the total number of pixels to obtain the variance, and performing arithmetic square root operation on the variance to obtain the standard deviation; subtracting the mean value from the radiation brightness value of each pixel and dividing by the standard deviation to convert all pixel values to values within the range of standard normal distribution, generating standardized sample data; constructing a deep learning semantic segmentation network, the network adopts an encoder-decoder structure, the encoder includes 4 convolution modules, each convolution module is composed of 2 3x3 convolution layers, 1 ReLU activation function and 1 2x2 max pooling layer, the output channel number of the convolution layer of the first convolution module is 64, that of the second is 128, that of the third is 256, and that of the fourth is 512, the padding mode of all convolution layers is SAME, and the step size is set to 1, the step size of the max pooling layer is set to 2, which is used to reduce the feature map size and retain key features; the decoder includes 4 deconvolution modules, each deconvolution module is composed of 1 2x2 deconvolution layer, 2 3x3 convolution layers and 1 ReLU activation function, the output channel number of the deconvolution layer of the first deconvolution module is 256, that of the second is 128, that of the third is 64, and that of the fourth is 32, and the step size of the deconvolution layer is set to 2, which is used to restore the spatial resolution of the feature map.

[0039] In training the network, a data set formed by marking 5-year historical disaster monitoring data of the Sichuan-Tibet networking project transmission corridor is selected as the training sample, the data set contains 10,000 image slices, and is marked as five categories of landslide submergence, forest fire influence, structural deformation, electrical fault and normal area. Before training, the sample is subjected to data enhancement operations such as random horizontal flip, vertical flip, rotation ± 10 degrees, brightness ± 10% fine tuning, etc. to improve the generalization ability of the model; the data set is divided into training set, validation set and test set according to the ratio of 7:2:1, the batch size is set to 8 during training, the Adam optimizer is used, the initial learning rate is 0.001, the learning rate is reduced to 0.5 of the original every 20 rounds, the cross entropy loss function (the loss value is obtained by calculating the logarithmic difference between the predicted category probability and the true labeled category, summing and averaging) is used to calculate the loss value of each iteration, the network parameters are adjusted by gradient descent method through back propagation, when the number of iterations reaches 100 rounds and the validation set loss value is stable below 0.01 for 10 consecutive rounds, the training is stopped, the test set is used to verify the model accuracy after the training is completed, and the classification accuracy is ensured to be not less than 95%; the standardized sample data is input into the trained deep learning semantic segmentation network, the basic texture features such as terrain edge and vegetation contour are extracted through the first convolution module of the encoder, the shallow shape features such as landslide shape and tower contour are extracted through the second convolution module, the middle layer spatial structure features such as terrain fluctuation rule and facility spatial layout are extracted through the third convolution module, and the deep layer semantic features such as the association between disaster affected area and facility are extracted through the fourth convolution module. The features extracted by each module are transmitted and retained in turn, and finally a high-level semantic feature set containing different levels of information is formed.

[0040] At step 602, based on the high-level semantic feature set, spatial information recovery and detail enhancement are performed to obtain detail enhancement feature data; based on the detail enhancement feature data, a probability distribution is calculated through a classification output layer to obtain initial pixel-level classification result data, specifically including: performing spatial information recovery on the high-level semantic feature set, adopting a dilated convolution operation, setting three different dilated rates of 2, 4 and 8, corresponding to three 3x3 convolution kernels, the sampling interval of the convolution kernel on the feature map is 2 when the dilated rate is 2, and the dilated rates of 4 and 8 are the same, the channel number of the convolution results of the three scales is 32, multi-scale convolution is performed on the high-level semantic features respectively, the feature receptive field is expanded, and long-distance spatial correlation information is captured; the features after three-scale convolution are spliced according to the channel dimension, the total channel number is 96, the features after spatial information recovery are obtained, and the lost spatial details in the encoding process are compensated; performing detail enhancement processing, introducing a residual connection structure, adding the 96-channel features after spatial information recovery and the shallow features (64-channel texture features of the first convolution module and 128-channel shape features of the second convolution module) extracted by the corresponding level of the encoder through 1x1 convolution to adjust the channel number to 96, then adding the corresponding level features, strengthening the expression of edge, corner and other detail features, obtaining the detail enhancement feature data; inputting the detail enhancement feature data into the classification output layer, the classification output layer adopts a Softmax activation function, calculating the probability value of each pixel belonging to five categories of landslide inundation, mountain fire influence, structure deformation, electrical fault and normal area, and the sum of the probability values of the five categories is 1; the class with the maximum probability value of each pixel is taken as the preliminary classification result of the pixel, and the preliminary classification results of all pixels are combined according to the spatial position of the slice to form the initial pixel-level classification result data.

[0041] Step 603, the initial pixel-level classification result data is processed to eliminate discrete noise points and merge adjacent same pixel regions, and refined semantic segmentation result data is obtained; based on the refined semantic segmentation result data, the connected space range classified as the power grid facility flooded area, the structure deformation area and the electrical fault area is recognized and extracted, and the disaster damage feature is generated, specifically including: morphological processing is performed on the initial pixel-level classification result data, first, an erosion operation is performed, a 3*3 structure element (a square region composed of a center pixel and 8 adjacent pixels around the center pixel) is selected, each pixel is traversed, the number of pixels with the same class as the pixel within the 3*3 range around the pixel is counted, if the number is less than 5, the pixel is determined as a discrete noise point, and the pixel is corrected to the class with the largest number within the 3*3 range around the pixel; then, an inflation operation is performed, and a 3*3 structure element is also selected, each pixel is traversed, if there is at least one pixel with the same class as the pixel within the 3*3 range around the pixel, the class range of the pixel is expanded to the 3*3 range around the pixel, and the small holes in the class area with no more than 2 pixels are filled; adjacent region merging is performed on the processed classification result data, the Euclidean distance (the square of the X-axis coordinate difference plus the square of the Y-axis coordinate difference, and then the arithmetic square root operation is performed) of the nearest two points on the boundary of two same pixel regions is calculated, if the distance is less than 3 pixels, the two regions are merged into one continuous region; the refined semantic segmentation result data is obtained through the above processing; the pixels classified as the power grid facility flooded area, the structure deformation area and the electrical fault area are identified one by one from the data, the three-dimensional coordinates of these pixels are extracted, a boundary point is taken every 0.5 meters in the order of X-axis from small to large and Y-axis from small to large, and the boundary contour of each region is outlined; the area (the number of pixels in the boundary contour multiplied by the actual area of each pixel 0.25 square meters, because the space grid is 0.5m*0.5m), the center point three-dimensional coordinates (the center point X-axis coordinate is obtained by summing the X-axis coordinates of all pixels in the region and dividing by the number of pixels, the Y-axis and Z-axis coordinates are calculated in the same way), and the elevation range (the minimum value to the maximum value of the Z-axis elevation of all pixels in the region) of each region are calculated, and the information is sorted according to the region category (flooded, deformed, fault), and the disaster damage feature is generated.

[0042] In this embodiment, with the multi-level feature extraction and detail enhancement of the deep learning semantic segmentation network, the accurate identification of the disaster damage area is realized, and the refinement of the disaster damage feature extraction is ensured.

[0043] In a preferred embodiment of the application, step 7, based on the disaster damage feature, the spatial diffusion path simulation and time evolution process analysis of the disaster influence are performed, the disaster space-time evolution feature is obtained, and the comprehensive quantitative evaluation of the damage degree and risk level of the power grid facility is performed, and the comprehensive damage evaluation result is generated, which can include: Step 701, based on the disaster damage characteristics, extract the spatial position information and damage type information of the power grid facility damage area as the disaster initial state data, specifically including: extracting the spatial position information of each damage area from the disaster damage characteristics, taking a sampling point every 0.5 meters in the clockwise direction for the boundary coordinates, recording the X-axis, Y-axis and Z-axis coordinates of each sampling point, with the coordinate values accurate to the centimeter level (retaining two decimal places), all sampling points are connected in sequence to form a continuous boundary coordinate chain, ensuring that the distance between adjacent two points on the chain is not more than 0.5 meters; the center point three-dimensional coordinate calculation method is to add the X-axis coordinates of all pixels in the region one by one to obtain the total sum of the X-axis coordinates, and then divide the total sum by the total number of pixels in the region to obtain the X-axis coordinate of the center point, the Y-axis coordinate is calculated according to the same logic, and the Z-axis coordinate is obtained by adding the Z-axis elevation of all pixels and then dividing by the total number of pixels; At the same time, record the power grid facility number covered by the region, the numbering rule is unified as tower number T+region code+3-digit serial number (such as T-CN-001), transmission line section number L+region code+start and end tower number (such as L-CN-T001-T005), insulator group number I+tower number+2-digit installation serial number (such as I-TCN001-01), to clearly show the one-to-one correspondence between each damage area and the power grid facility; distinguish the damage type of each damage area, accurately label as power grid facility flooded area, structural deformation area or electrical fault area, and record the damage details in detail, such as the depth of the tower base soaked in the flooded area (calculated by the Z-axis elevation difference of the pixel), the inclination angle of the tower in the deformation area (calculated by the coordinate deviation between the base and the top), and the radiation abnormal value range of the conductor connection point in the fault area; integrate the above spatial position information (boundary coordinate chain, center point coordinate), damage type information, facility number and associated details to form structured disaster initial state data.

[0044] Step 702, calculate the disaster initial state data to get the disaster spatial diffusion path data; analyze the time evolution of the disaster spatial diffusion path data to get the time evolution process data of the disaster impact, which specifically includes: constructing a disaster spatial diffusion model, input parameters include disaster initial state data (initial location, damage type), topographic data of Sichuan-Tibet interconnected power transmission corridor (slope, slope direction, elevation, extracted from 1:5000 topographic map), vegetation distribution data (coverage degree is inverted from remote sensing image, vegetation types are divided into coniferous forest, alpine meadow and desert) and power grid facility layout data (facility spacing is recorded according to actual account, facility types are divided into tower, transmission line and insulator group); the model training process is as follows: collect the diffusion case data of 100 typical disasters (60 landslides and 40 forest fires) in Sichuan-Tibet region in the past 5 years, each case contains initial location, topographic and vegetation parameters, facility layout, actual diffusion path and impact range, divide the case data into training set and validation set in the ratio of 7:3, take the average Euclidean distance of the predicted diffusion path and the actual path as the loss target, use gradient descent method to optimize parameters, set the initial learning rate to 0.001, iterate 500 times, optimize the slope weight (final value 0.4), vegetation resistance weight (final value 0.3) and facility resistance weight (final value 0.3), stop training when the average deviation on the validation set is less than 1 meter; use the trained model to calculate the disaster spatial diffusion path, the landslide disaster diffusion direction calculation is to take the center point of the initial damage area as the origin, divide 360 degrees evenly into 36 directions, take the topographic elevation data within 100 meters in each direction, calculate the slope value of each direction (vertical elevation difference divided by horizontal distance, such as the elevation from 3000 meters to 2980 meters within 100 meters in a certain direction, the slope value = 20 meters ÷ 100 meters = 0.2), select the three directions with the largest slope value as the main diffusion direction; the forest fire disaster diffusion direction calculation is to use vector composition method, first convert the topographic slope direction and the dominant wind direction into azimuth angle (azimuth angle takes north as 0 degree, clockwise increasing), the dominant wind direction in the east section of Sichuan-Tibet is southeast wind corresponding to azimuth angle 135 degrees, the middle section is southwest wind corresponding to azimuth angle 225 degrees, and the west section is northwest wind corresponding to azimuth angle 315 degrees, the slope direction is south corresponding to azimuth angle 180 degrees, and the east direction is corresponding to azimuth angle 90 degrees; take the slope direction azimuth angle as the first vector and the dominant wind direction azimuth angle as the second vector, respectively decompose into x-axis (east-west direction, east is positive) and y-axis (north-south direction, north is positive) components, x-axis component = vector size × cos (azimuth angle radian), y-axis component = vector size × sin (azimuth angle radian), vector size is uniformly taken as 1; for example, the slope direction is south (azimuth angle 180 degrees) and the dominant wind direction is southeast wind (azimuth angle 135 degrees), the first vector x component = 1 × cos (π) = -1, y component = 1 × sin (π) = 0; the second vector x component = 1 × cos (3π / 4) ≈ -0.707, y component = 1 × sin (3π / 4) ≈ 0.707; add the x components of the two vectors (-1 + (-0.707) ≈ -1.707), the y component is added (0+0.707=0.707), and the x and y components of the resultant vector are obtained; the resultant azimuth is calculated by arctan2 (y component, x component), in this case arctan2 (0.707, -1.707)=156 degrees, corresponding to 24 degrees south by east, which is the direction of the spread of the forest fire disaster; at the same time, the diffusion resistance of each diffusion direction is calculated: the resistance value in the area with a coniferous forest coverage of more than 60% is 0.9, the resistance value in the alpine meadow area with a coverage of more than 60% is 0.7, the resistance value in the desert area with a coverage of less than 30% is 0.3, the resistance value in the iron tower group area is 0.7, and the resistance value in the open area of the power transmission line is 0.4; the diffusion resistances are sorted from small to large to determine three main diffusion paths, and the coordinates of the key nodes are recorded at intervals of 50 meters on each path, the total length of the path (the sum of the distances between the key nodes) is calculated, and the numbers of the power grid facilities passed through are marked; the time step is set to 1 hour, and the time evolution analysis is carried out based on the diffusion model, and the diffusion distance in each time step is calculated according to the initial speed x (1-diffusion resistance) (the initial speed of the landslide is 15 meters / hour, and the initial speed of the forest fire is 20 meters / hour), for example, if the diffusion resistance of the landslide in a certain direction is 0.3, the diffusion distance in this time step is =15x (1-0.3)=10.5 meters; the diffusion area is calculated by the grid method, and the specific steps are to take the minimum coordinate of the X axis and the minimum coordinate of the Y axis of the coordinate chain of the diffusion area boundary as the starting point, divide the square grid according to the specification of 1 meter x 1 meter, and cover the entire diffusion area and the surrounding 0.5 meter range; whether each grid is in the diffusion area is judged by the ray method, and a ray is emitted from the center point of the grid in any direction (such as the positive right direction), and the number of intersection points of the ray and the boundary coordinate chain is counted; if the number of intersection points is odd, the grid is in the area, and if the number of intersection points is even, the grid is outside the area; for the grid completely in the area, 1 square meter is counted into the area; for the grid overlapping with the boundary (the number of intersection points of the ray is 0 or the grid is judged to be ambiguous), the area is estimated by calculating the proportion of the pixels in the grid that belong to the area; each grid is divided into 100x100 pixels, the number of pixels in the area is counted, the proportion is obtained by dividing the number by 10000, and the contribution area of the grid is obtained by multiplying the proportion by 1 square meter; the total area of the diffusion area is obtained by adding the contribution areas of all the grids; the area increment is the current step area minus the last step area, the diffusion distance increment is the distance between the current step end and the initial position minus the corresponding distance of the last step, and the number of newly added affected facilities is counted by judging whether the facility coordinates are in the current diffusion area; the diffusion speed variation trend is analyzed, if the diffusion speed of a certain time step deviates from the average speed of the historical similar cases by more than 20%, the diffusion resistance value of the corresponding direction is adjusted (the resistance value is increased by 0.1 when the deviation is too large) and recalculated; finally, the time evolution process data including the diffusion area, diffusion distance, affected facility number, and diffusion area boundary coordinate chain of each time step (0 to 24 hours) are formed, and the disaster spreading law is completely presented.

[0045] Step 703, the disaster space diffusion path data and the time evolution process data of disaster influence are fused to obtain disaster space-time evolution characteristic data, specifically including: determining the corresponding relationship of each time step and space parameter, such as 0 hour corresponding to the boundary coordinate chain of the initial state of disaster, the center point coordinate, the diffusion distance 0 meters, 1 hour corresponding to the boundary coordinate chain of the first path extended 15 meters (landslide), the new center point coordinate, the diffusion distance 15 meters, at the same time, the area increment and the number of new influence facilities of the time step are associated; constructing a disaster space-time evolution characteristic matrix, the rows of the matrix are arranged in time step from 0 hour to 24 hours in turn, and the columns include space characteristic parameters and influence parameters, the space characteristic parameters are specifically diffusion area boundary coordinate chain, center point X axis / Y axis / Z axis coordinate, current length of three main paths, total area of diffusion area, and the influence parameters are specifically total number of cumulative affected facilities, area proportion of three types of regions of submergence / deformation / failure (area of a certain type of region ÷ total diffusion area), and list of affected facility numbers; the matrix data is filled in time step order one by one, and when filling, it is ensured that the space parameters and influence parameters of each time step are matched, such as the number of new influence facilities must be the facilities whose coordinates are in the diffusion area of the time step; the matrix data is subjected to consistency check, and the check method is: extracting the diffusion area boundary coordinate chain of a certain time step, calculating the space range of the region, and checking whether the facility coordinates in the facility list of the time step are in the range one by one, if there are more than 3 facility coordinates out of range, then the diffusion path and influence facility of the time step are recalculated; after the check is passed, complete disaster space-time evolution characteristic data is formed.

[0046] At step 704, based on the disaster spatio-temporal evolution characteristic data, the exposure index and the physical vulnerability index of each power grid facility unit under the influence of the disaster are calculated, and facility damage potential quantification data is generated, specifically including: calculating the exposure index, first, the spatial parameters of each type of power grid facility unit are determined, the tower unit is calculated according to the base footprint area of 4 meters x 4 meters (16 square meters), the transmission line unit is calculated according to the line width of 0.5 meters multiplied by the actual length (such as a line section of 1000 meters long, the footprint area is 500 square meters), and the insulator group unit is calculated according to the installation area of 0.5 meters x 0.5 meters (0.25 square meters); the area covered by the disaster in each time step of the facility unit is extracted through the disaster spatio-temporal evolution characteristic data, and the covered area is calculated according to the polygon area of the overlapping part of the facility unit and the diffusion area (such as the tower unit has 8 square meters of overlap in a certain time step, the covered area of this step is 8 square meters); the total covered area is obtained by adding the covered areas of all time steps, and the exposure index is obtained by dividing the total covered area by the footprint area of the facility unit (such as the total covered area of the tower unit is 12 square meters, and the exposure index = 12 ÷ 16 = 0.75); calculating the physical vulnerability index, setting the basic vulnerability value according to the facility type, the tower base is 0.8, the transmission conductor is 0.6, and the insulator group is 0.7, combined with the disaster type adjustment: the landslide disaster increases the basic vulnerability value of the tower base by 0.15 (the adjusted value is 0.95), and the forest fire disaster increases the basic vulnerability value of the transmission conductor by 0.15 (the adjusted value is 0.75); the time influence coefficient is determined according to the disaster influence duration, 1 to 3 hours take 1.0, 3 to 6 hours take 1.2, 6 to 12 hours take 1.5, and 12 hours or more take 1.8; the physical vulnerability index is obtained by multiplying the basic vulnerability value by the time influence coefficient (such as the tower base is affected by the landslide for 8 hours, and the physical vulnerability = 0.95 x 1.5 = 1.425); the exposure index of each facility unit is multiplied by the physical vulnerability index to obtain the facility damage potential quantification data (such as the exposure index of the tower unit is 0.75 x 1.425 = 1.06875).

[0047] Step 705, the facility damage potential quantification data is calculated and loss estimation is obtained, and the damage degree quantification value and risk level are obtained, specifically including: based on the facility damage potential quantification data, the loss estimation is calculated, the direct loss calculation method is that the facility damage potential quantification data is multiplied by the actual cost of the facility, the cost of a single iron tower is 500,000 yuan, the cost of a 1000-meter power transmission line section is 200,000 yuan, and the cost of a single insulator group is 5,000 yuan (such as the damage potential of an iron tower unit is 1.06875, the direct loss = 1.06875 x 50 = 53.4375 thousand yuan); the indirect loss calculation method is that the disaster influence duration is multiplied by the unit time power failure loss, the unit time power failure loss is determined according to the regional type, the industrial concentrated area section (such as Changdu industrial zone) is 50,000 yuan per hour, the residential area section is 10,000 yuan per hour, and the remote section is 5,000 yuan per hour (such as the iron tower unit is affected for 8 hours, and is located in the industrial area section, the indirect loss = 8 x 5 = 40,000 yuan); the direct loss and the indirect loss are added to obtain the total loss amount of each power grid facility unit, that is, the damage degree quantification value (such as the above-mentioned iron tower unit 53.4375 + 40 = 93.4375 thousand yuan); the risk level division standard is set, the damage degree quantification value of the first risk (low risk) is 0 to 5,000 yuan, the prevention and control suggestion is to be included in the daily monitoring, and the monthly inspection is 1 time; the damage degree quantification value of the second risk (low-medium risk) is 5 to 20,000 yuan, the prevention and control suggestion is to be inspected once every half month, and the emergency disposal plan is developed; the damage degree quantification value of the third risk (medium-high risk) is 20 to 100,000 yuan, the prevention and control suggestion is to be inspected once every week, and the local repair is immediately carried out; the damage degree quantification value of the fourth risk (high risk) is more than 100,000 yuan, and the prevention and control suggestion is to be disposed of in 24 hours, and the damaged parts are replaced.

[0048] At step 706, based on the damage degree quantitative value and the risk level, regional comprehensive analysis and weighted superposition calculation are performed to generate a comprehensive damage assessment result, specifically including: dividing the Sichuan-Tibet power transmission corridor into three geographical sub-regions according to the terrain and administrative regions, the east section is the Changdu-Linzhi section, the middle section is the Linzhi-Lhasa section, and the west section is the Lhasa-Shigatse section, each sub-region contains a plurality of power grid facility units; calculating the total damage quantitative value of each sub-region, and adding the damage degree quantitative values of all power grid facility units in the sub-region to obtain the total (for example, the east section contains 100 facility units, and the total damage quantitative value = the sum of the damage values of each unit); counting the number of settings of each risk level in each sub-region, and obtaining the proportion of high-risk facilities by dividing the number of settings of the fourth risk level by the total number of settings in the sub-region (for example, the east section has 10 fourth risk level facilities and a total of 100 facilities, and the proportion = 10 ÷ 100 = 0.1); setting the weight of the total damage quantitative value of the region as 0.6, and the weight of the proportion of high-risk facilities as 0.4; calculating the regional damage proportion by dividing the total damage quantitative value of the region by the total cost of all facilities in the sub-region (for example, the total damage quantitative value of the east section is 80 million yuan, and the total cost is 800 million yuan, and the damage proportion = 8000 ÷ 80000 = 0.1); multiplying the regional damage proportion by 0.6 to obtain the damage proportion contribution value (0.1 x 0.6 = 0.06), and multiplying the proportion of high-risk facilities by 0.4 to obtain the risk proportion contribution value (0.1 x 0.4 = 0.04); adding the two contribution values to obtain the comprehensive evaluation value of the region (0.06 + 0.04 = 0.1); integrating the comprehensive evaluation values of all regions, the damage degree quantitative values of each facility unit and the risk levels to form a comprehensive damage assessment result.

[0049] In this embodiment, through the spatio-temporal evolution analysis of disasters, the disaster diffusion path and influence process are clearly presented, and the quantitative calculation of exposure and vulnerability indexes improves the guarantee capability of power grid safe operation.

[0050] As shown in Figure 2 The embodiment of the present application also provides a power grid composite chain disaster multi-source monitoring damage assessment system, which comprises: A collection module is configured to acquire multi-source remote sensing monitoring data; A calculation module is configured to perform fusion processing on the multi-source remote sensing monitoring data, generate a fused disaster monitoring image data set, and convert the fused disaster monitoring image data set into an initial three-dimensional discrete point set; calculate the initial three-dimensional discrete point set, extract position extreme points, determine an initial spatial convex polyhedron structure based on the position extreme points, sequentially determine the position relationship between the remaining candidate data points and the initial spatial convex polyhedron structure, generate a spatial envelope surface, and optimize the three-dimensional discrete point set; A construction module is configured to construct a spatial surface based on the three-dimensional discrete point set, and define a reference geographical coverage area according to the distribution range of the power grid facilities; The dividing module is configured to divide the reference geographic coverage area according to the density distribution of the three-dimensional discrete point set in the reference geographic coverage area, generate a plurality of independent analysis units, attribute each three-dimensional discrete point to a corresponding independent analysis unit, and calculate a unit space calibration coefficient; The correction module is configured to correct the fused disaster monitoring image data set regionally by using the unit space calibration coefficient to obtain the calibrated disaster monitoring image data. The extraction module is configured to extract disaster damage features by using a deep learning semantic segmentation network based on the calibrated disaster monitoring image data. The evaluation module is configured to simulate a spatial diffusion path of disaster influence and analyze a time evolution process based on the disaster damage features, obtain disaster spatio-temporal evolution features, and comprehensively quantitatively evaluate damage degrees and risk levels of power grid facilities to generate a comprehensive damage evaluation result.

[0051] It should be noted that the system corresponds to the above method, and all implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0052] Embodiments of the present application also provide a computing device, comprising a processor and a memory storing a computer program, wherein the computer program is executed by the processor to perform the above method. All implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0053] Embodiments of the present application also provide a computer readable storage medium storing instructions, wherein the instructions are executed on a computer to make the computer perform the above method. All implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0054] The above is the preferred embodiment of the present application, and it should be noted that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A multi-source monitoring and damage assessment method for complex chain-related disasters in power grids, characterized in that, The method includes: Step 1: Acquire multi-source remote sensing monitoring data; Step 2: The multi-source remote sensing monitoring data is fused to generate a fused disaster monitoring image dataset, which is then converted into an initial three-dimensional discrete point set. The initial three-dimensional discrete point set is calculated to extract the location extreme points. Based on the location extreme points, the initial spatial convex polyhedron structure is determined. The remaining candidate data points are then sequentially compared with the initial spatial convex polyhedron structure to determine their positional relationship, generating a spatial envelope surface. The three-dimensional discrete point set is then optimized. Step 3: Based on the three-dimensional discrete point set, construct a spatial surface and define a benchmark geographical coverage area according to the distribution range of power grid facilities; Step 4: Based on the density distribution of the three-dimensional discrete point set within the baseline geographic coverage area, the baseline geographic coverage area is divided into multiple independent analysis units. Each three-dimensional discrete point is assigned to the corresponding independent analysis unit, and the unit spatial calibration coefficient is calculated. Step 5: Using the unit space calibration coefficient, perform regional correction on the fused disaster monitoring image dataset to obtain the calibrated disaster monitoring image data; Step 6: Based on the calibrated disaster monitoring image data, extract disaster damage features using a deep learning semantic segmentation network; Step 7: Based on the characteristics of disaster damage, simulate the spatial diffusion path and analyze the temporal evolution process of the disaster impact to obtain the spatiotemporal evolution characteristics of the disaster, and conduct a comprehensive quantitative assessment of the degree of damage and risk level of the power grid facilities to generate a comprehensive damage assessment result.

2. The method for multi-source monitoring and damage assessment of power grid compound chain disasters according to claim 1, characterized in that, Multi-source remote sensing monitoring data are fused to generate a fused disaster monitoring image dataset, which is then converted into an initial three-dimensional discrete point set. The initial 3D discrete point set is calculated to extract location extrema. Based on these extrema, the initial spatial convex polyhedron structure is determined. The remaining candidate data points are then sequentially compared with the initial spatial convex polyhedron structure to determine their positional relationships, generating a spatial envelope surface. This optimization yields the 3D discrete point set, including: Spatial registration and radiometric normalization were performed on the multi-source remote sensing monitoring data to obtain the registered and normalized multi-source remote sensing monitoring dataset. The registered and normalized multi-source remote sensing monitoring datasets are fused to generate a fused disaster monitoring image dataset; the fused disaster monitoring image dataset is then converted into an initial three-dimensional discrete point set. Based on the initial set of three-dimensional discrete points, extract the positional extreme points along the three-dimensional coordinate axes; based on the positional extreme points, construct the initial spatial convex polyhedron structure; based on the initial set of three-dimensional discrete points and the positional extreme points, determine the set of candidate data points; Based on the candidate data point set and the initial spatial convex polyhedron structure, a spatial envelope surface is generated by point-by-point position determination and structure update; based on the spatial envelope surface, the initial three-dimensional discrete point set is screened and optimized to obtain a three-dimensional discrete point set.

3. The method for multi-source monitoring and damage assessment of power grid compound chain disasters according to claim 2, characterized in that, Based on a three-dimensional discrete point set, a spatial surface is constructed, and a baseline geographical coverage area is defined according to the distribution range of power grid facilities, including: Based on a three-dimensional discrete point set, a continuous spatial surface is calculated and generated; based on the spatial surface, a meshing process is performed to obtain a spatial surface mesh; based on the known geographical distribution range data of power grid facilities, two-dimensional polygons are generated, and the two-dimensional polygons are extended to three-dimensional space and processed to obtain a facility distribution surface mesh; Based on the spatial surface mesh and the facility distribution surface mesh, the set of intersecting triangular facets is calculated; based on the set of intersecting triangular facets, the minimum bounding rectangle region is calculated. Based on the minimum bounding rectangle region and the spatial surface, the baseline geographic coverage area is obtained through projection mapping.

4. The method for multi-source monitoring and damage assessment of power grid compound chain disasters according to claim 3, characterized in that, Based on the density distribution of the 3D discrete point set within the baseline geographic coverage area, the baseline geographic coverage area is divided into multiple independent analysis units. Each 3D discrete point is assigned to its corresponding independent analysis unit, and the unit spatial calibration coefficients are calculated, including: Based on the baseline geographic coverage area and the three-dimensional discrete point set, the spatial density distribution of the three-dimensional discrete point set within the baseline geographic coverage area is calculated and generated. Based on spatial density distribution, the baseline geographic coverage area is divided to generate multiple independent analysis units; Based on a set of three-dimensional discrete points and independent analysis units, each three-dimensional discrete point is assigned to a corresponding independent analysis unit through spatial location matching, and a subset of three-dimensional discrete points is extracted. Based on a subset of three-dimensional discrete points, the geometric feature statistics and spectral reflectance feature statistics of the subset are calculated; based on the geometric feature statistics and spectral reflectance feature statistics, the cell space calibration coefficient is generated through weighted fusion calculation.

5. The method for multi-source monitoring and damage assessment of power grid compound chain disasters according to claim 4, characterized in that, Using cell spatial calibration coefficients, regional corrections are performed on the fused disaster monitoring image dataset to obtain calibrated disaster monitoring image data, including: Based on the fused disaster monitoring image dataset and unit spatial calibration coefficients, a registration relationship between the spatial range of each independent analysis unit and the corresponding unit spatial calibration coefficients is established. Based on the registration relationship, the original data blocks within the spatial range of each independent analysis unit are extracted from the fused disaster monitoring image dataset; Based on the original data blocks, the cell space calibration coefficients corresponding to the original data blocks are determined according to the registration relationship, and geometric position correction and radiometric value correction are performed using the cell space calibration coefficients to obtain the correction results; Based on the calibration results, regional calibrated data blocks are generated; based on the regional calibrated data blocks, spatial seamless stitching and data fusion are performed to obtain calibrated disaster monitoring image data.

6. The method for multi-source monitoring and damage assessment of power grid compound chain disasters according to claim 5, characterized in that, Based on calibrated disaster monitoring image data, disaster damage features are extracted using a deep learning semantic segmentation network, including: Based on the calibrated disaster monitoring image data, image slicing and standardization processing are performed to generate standardized sample data; based on the standardized sample data, multi-level spatial spectral features are extracted through the multi-level convolutional structure of a deep learning semantic segmentation network to obtain a high-level semantic feature set. Based on the high-level semantic feature set, spatial information recovery and detail enhancement are performed to obtain detail-enhanced feature data; based on the detail-enhanced feature data, the probability distribution is calculated through the classification output layer to obtain the initial pixel-level classification result data; The initial pixel-level classification results are processed to eliminate discrete noise points and merge adjacent pixel regions of the same type to obtain refined semantic segmentation results. Based on the refined semantic segmentation results, the connected spatial ranges classified as power grid facility flooding areas, structural deformation areas, and electrical fault areas are identified and extracted to generate disaster damage features.

7. The method for multi-source monitoring and damage assessment of power grid compound chain disasters according to claim 6, characterized in that, Based on the characteristics of disaster damage, spatial diffusion paths and temporal evolution processes of disaster impacts are simulated and analyzed to obtain the spatiotemporal evolution characteristics of the disaster. A comprehensive quantitative assessment of the damage degree and risk level of power grid facilities is then conducted, generating a comprehensive damage assessment result, including: Based on the characteristics of disaster damage, spatial location information and damage type information of the damaged area of ​​power grid facilities are extracted as initial disaster state data. The initial state data of the disaster are calculated to obtain the spatial diffusion path data of the disaster; the temporal evolution analysis of the spatial diffusion path data of the disaster is performed to obtain the temporal evolution process data of the disaster impact. Data on the spatial diffusion path of disasters and the temporal evolution of disaster impacts are fused to obtain spatiotemporal evolution characteristic data of disasters. Based on the spatiotemporal evolution characteristics of disasters, the exposure index and physical vulnerability index of each power grid facility unit under the impact of disasters are calculated to generate quantitative data on the potential damage to the facilities. The potential damage data of the facilities is used to calculate and estimate the loss, so as to obtain the quantitative value of the damage level and the risk level. Based on the quantitative value of damage degree and risk level, a comprehensive regional analysis and weighted overlay calculation are performed to generate a comprehensive damage assessment result.

8. A multi-source monitoring and damage assessment system for complex chain-related disasters in power grids, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire multi-source remote sensing monitoring data; The calculation module is used to fuse multi-source remote sensing monitoring data, generate a fused disaster monitoring image dataset, and convert it into an initial three-dimensional discrete point set; The initial three-dimensional discrete point set is calculated to extract the positional extreme points. Based on the positional extreme points, the initial spatial convex polyhedron structure is determined. The remaining candidate data points are sequentially compared with the initial spatial convex polyhedron structure to determine their positional relationship, generate a spatial envelope surface, and optimize to obtain the three-dimensional discrete point set. The module is used to construct a spatial surface based on a three-dimensional discrete point set and define a baseline geographic coverage area according to the distribution range of power grid facilities. The partitioning module is used to partition the baseline geographic coverage area according to the density distribution of the three-dimensional discrete point set within the baseline geographic coverage area, generate multiple independent analysis units, assign each three-dimensional discrete point to the corresponding independent analysis unit, and calculate the unit spatial calibration coefficient. The calibration module is used to perform regional calibration on the fused disaster monitoring image dataset using the cell space calibration coefficient to obtain calibrated disaster monitoring image data; The extraction module is used to extract disaster damage features based on calibrated disaster monitoring image data using a deep learning semantic segmentation network. The assessment module is used to simulate the spatial diffusion path and analyze the temporal evolution process of disaster impact based on disaster damage characteristics, obtain the spatiotemporal evolution characteristics of disaster, and conduct a comprehensive quantitative assessment of the degree of damage and risk level of power grid facilities, generating a comprehensive damage assessment result.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images