Aerial work emergency scheduling method and system

Abstract

The invention provides a high-altitude operation emergency scheduling method and system, and the method comprises the steps: firstly obtaining a real-time video data set of a high-altitude operation region, carrying out the preprocessing of the real-time video data set, and generating a standardized video data flow, and then calling a video analysis model to extract an operation environment feature set and an operator behavior feature set; and then constructing a graph structure based on the working environment feature set, generating an environment state feature vector through graph convolutional network multi-level feature aggregation, and generating a dynamic emergency scheduling strategy in combination with the environment state feature vector and the working personnel behavior feature set. And finally, generating an emergency scheduling instruction set after strategy verification, and transmitting the emergency scheduling instruction set to the operation terminal equipment to trigger emergency response operation, so that a targeted emergency scheduling strategy can be dynamically generated according to actual conditions, automation and intelligence of emergency scheduling are realized, and the emergency response efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-altitude operation safety guarantee and emergency management, in particular to a high-altitude operation emergency dispatching method and system. BACKGROUND

[0002] In the field of high-altitude operation, the operation environment is complex and changeable, and the behavior of the operation personnel directly relates to the safety of the operation. Once a sudden situation occurs, such as sudden intensification of bad weather, equipment failure, etc., timely and effective emergency dispatching is crucial. At present, the emergency dispatching of high-altitude operation mainly relies on manual experience judgment and simple feedback of on-site personnel. Manual experience judgment has the problems of strong subjectivity and difficulty in fully and accurately grasping complex and changeable environmental conditions, and cannot quickly and accurately formulate emergency dispatching strategies. The feedback of on-site personnel often has delays and inaccuracies, and it is difficult to integrate information globally for the entire operation environment.

[0003] In addition, the existing emergency dispatching method lacks deep mining and analysis of the characteristics of the operation environment and the behavior characteristics of the operation personnel, and cannot dynamically generate targeted emergency dispatching strategies according to the actual situation, resulting in slow emergency response speed and unreasonable dispatching when facing sudden situations, increasing the risk of high-altitude operation and seriously threatening the safety of the operation personnel. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a high-altitude operation emergency dispatching method, which comprises: obtaining a set of real-time video data of a high-altitude operation area, performing a preprocessing operation on the set of real-time video data to generate a standardized video data stream; calling a video analysis model to perform feature extraction on the standardized video data stream to generate a set of operation environment features and a set of operation personnel behavior features; constructing an operation environment graph structure based on the set of operation environment features, calling a graph convolution network to perform a multi-level feature aggregation operation on the operation environment graph structure to generate an environment state feature vector; generating a dynamic emergency dispatching strategy according to the environment state feature vector and the set of operation personnel behavior features; performing an effectiveness verification operation on the dynamic emergency dispatching strategy to obtain a strategy verification result, and generating a set of emergency dispatching instructions based on the strategy verification result, and transmitting the set of emergency dispatching instructions to an operation terminal device to trigger an emergency response operation.

[0005] In still another aspect, the embodiment of the present application further provides a high-altitude operation emergency dispatching system, comprising a processor, a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above-mentioned method.

[0006] Based on the above aspects, the embodiment of the present application generates a standardized video data stream by acquiring a real-time video data set of a high-altitude operation area and performing preprocessing, extracts a work environment feature set and a work personnel behavior feature set by calling a video analysis model, comprehensively captures key information in a high-altitude operation scene, constructs a work environment graph structure based on the work environment feature set and performs multi-level feature aggregation by using a graph convolution network, and the generated environment state feature vector can accurately reflect the comprehensive state of the work environment, generates a dynamic emergency dispatching strategy in combination with the environment state feature vector and the work personnel behavior feature set, fully considers the interaction between the work environment and personnel behavior, makes the dynamic emergency dispatching strategy more targeted and adaptive, verifies the effectiveness of the dynamic emergency dispatching strategy and generates an emergency dispatching instruction set to be transmitted to a work terminal device to trigger an emergency response operation, realizes the automation and intelligentization of emergency dispatching, greatly improves the speed and accuracy of emergency response, effectively reduces the risk of high-altitude operation, and protects the life safety of work personnel. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is an execution flow diagram of the high-altitude operation emergency dispatching method provided by the embodiment of the present application.

[0008] Figure 2 is a schematic diagram of exemplary hardware and software components of the high-altitude operation emergency dispatching system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0009] The present application will be described in detail below with reference to the accompanying drawings, Figure 1 is a flow diagram of the high-altitude operation emergency dispatching method provided by an embodiment of the present application, and the high-altitude operation emergency dispatching method will be described in detail below.

[0010] Step S110: acquiring a real-time video data set of a high-altitude operation area, performing a preprocessing operation on the real-time video data set to generate a standardized video data stream.

[0011] In this embodiment, the high-altitude maintenance work scene of a large wind turbine is taken as an example. In this scene, multiple video acquisition devices such as cameras C1, C2, C3, etc. are arranged around the high-altitude work area of the wind turbine, which real-time shoot the work area from different angles, and the video data collected by these cameras jointly constitute a real-time video data set. Due to the differences in performance and settings of different cameras, these video data may have inconsistencies in resolution, frame rate, color gamut, etc., and therefore need to be preprocessed to generate standardized video data streams suitable for subsequent analysis.

[0012] Step S111: Perform resolution equalization processing operation on the real-time video data set to generate initial video data frames in a uniform resolution format.

[0013] Suppose the video resolution of camera C1 is a1xb1, the video resolution of camera C2 is a2xb2, and the video resolution of camera C3 is a3xb3. To achieve resolution equalization, first determine a uniform target resolution of AxB. For the video collected by camera C1, an image scaling algorithm is used. This algorithm processes each video frame, for example, for each pixel position in the original resolution a1, according to its relationship with adjacent pixels, the pixel value at the corresponding position under the target resolution A is calculated according to the set interpolation rule. For example, if a linear interpolation method is used, for original pixel position i, new pixel position j, the new pixel value is calculated according to the relative position relationship between i and j through the pixel values of original pixel positions i and i+1. The vertical pixels are calculated in the same way. Through such processing, the video resolution collected by camera C1 is adjusted to AxB. The same method is applied to the videos collected by cameras C2 and C3, and finally initial video data frames in a uniform resolution format of AxB are generated.

[0014] Step S112: Call a moving target detection algorithm to perform noise filtering operation on the initial video data frames to remove non-work target interference pixel regions and obtain filtered video data frames.

[0015] In this embodiment, the initial video data frames may contain various non-work target interferences such as wind-blown debris, images of other irrelevant facilities in the distance, etc. The moving target detection algorithm identifies the interference by analyzing the motion characteristics of the pixel regions in the initial video data frames, which specifically compares the changes of the same position pixel regions in adjacent frames frame by frame.

[0016] For example, for a pixel region, if its position change, color change or texture change between adjacent frames do not conform to the motion law of the operation target, such as the motion speed being too fast or too slow, or the motion trajectory being chaotic, it is determined that the pixel region is a non-operation target interference pixel region. For the pixel region determined to be interference, various processing methods are adopted.

[0017] One way is to set its pixel value to a value similar to the surrounding background. A suitable pixel value can be calculated by analyzing the color, brightness, etc. of the surrounding background pixels to replace the pixel value of the interference region. Another way is to perform interpolation calculation according to the pixel value of the surrounding operation target, to generate a pixel value that is more integrated with the surrounding operation scene, thereby removing the interference pixel region and obtaining the filtered video data frame.

[0018] Step S113: performing time sequence alignment processing operation on the filtered video data frame to generate a standardized video data stream with continuous time stamps.

[0019] Since the start time and frame rate of different cameras C1, C2, C3 may differ, resulting in inconsistent video data in time. Let the start collection time of camera C1 be t1, the start collection time of camera C2 be t2, and the start collection time of camera C3 be t3, and their frame rates be f1, f2, f3 respectively.

[0020] In order to generate a standardized video data stream with continuous time stamps, a unified start time T0 is first determined. For the video data collected by camera C1, the time difference Δt1n of each frame relative to T0 is calculated (n represents the frame number). Time sequence alignment is achieved by adjusting the frame rate or inserting / deleting frames. For example, if the frame rate f1 of camera C1 is higher than the desired unified frame rate F, and the time difference Δt1n of a certain frame is calculated to be significantly different from the time point calculated according to the unified frame rate, some frames can be deleted appropriately. When deleting frames, the frames that have less impact on the overall video information will be selected for deletion according to the content relationship between the previous and next frames. If the frame rate f1 is lower than the unified frame rate F, some intermediate frames are inserted through interpolation algorithm. The interpolation algorithm calculates the pixel value of the inserted frame according to the pixel value and motion trend of the previous and next frames.

[0021] The same method is applied to the video data collected by cameras C2 and C3. Finally, all video data are arranged in time sequence, and each video frame corresponds to a continuous time stamp, forming a standardized video data stream with continuous time stamps.

[0022] Step S114: performing a color gamut conversion operation on the standardized video data stream to map the original color gamut space to a preset standardized color gamut space, and performing an illumination compensation processing operation on the converted video data stream to eliminate the shadow area and overexposure area in the initial video data frame, to obtain a compensated video data stream.

[0023] In this embodiment, the original color gamut spaces of the videos captured by different cameras can be different. Let the original color gamut space of camera C1 be G1, the original color gamut space of camera C2 be G2, and the original color gamut space of camera C3 be G3. The preset standardized color gamut space is G0. A color gamut conversion algorithm is used to convert the color value of each pixel in the standardized video data stream from its original color gamut space to the preset standardized color gamut space G0. For each pixel, the red (R), green (G), and blue (B) color components are calculated according to the conversion relationship between the original color gamut space and G0.

[0024] For example, for the video captured by camera C1, if the value of the red component in the original color gamut space G1 is R1, the value of the red component in G0 is calculated according to the conversion function from G1 to G0. After color gamut conversion, illumination compensation processing is performed. Because in the work site, some areas can be overexposed due to direct sunlight, and some areas can be in the shadow due to equipment blocking. The illumination compensation algorithm analyzes the brightness value L of each pixel. For pixels with excessively high brightness value (overexposure), the brightness is reduced. For example, by an adjustment coefficient k1 (0 < k1 < 1), the R, G, and B component values of the overexposed pixel are multiplied by k1. For pixels with excessively low brightness value (shadow), the brightness is increased. For example, by an adjustment coefficient k2 (k2 > 1), the R, G, and B component values of the shadow pixel are multiplied by k2, thereby eliminating the shadow area and overexposure area in the initial video data frame, to obtain a compensated video data stream.

[0025] Step S115: performing a frame rate stability detection operation on the compensated video data stream to eliminate video jitter and generate a time-continuous standardized video data stream.

[0026] In the high-altitude operation scene of a wind turbine, the camera may produce jitter due to strong wind, equipment vibration and other factors, resulting in unstable video frame rate. The frame rate stability detection algorithm analyzes the time interval and image content change between adjacent frames in the compensated video data stream to determine the frame rate stability. Let the time interval between adjacent frames be Δt, and the average time interval be ΔT. If the deviation of the time interval Δt of a certain frame from the average time interval ΔT exceeds the set threshold, it indicates that there is a frame rate instability. At the same time, the video jitter is detected by analyzing the displacement and deformation of the image content of adjacent frames. For the part with frame rate instability and video jitter, appropriate processing measures are taken. If the frame rate of a certain section of video is too fast, some intermediate frames are inserted appropriately. The pixel value of the inserted intermediate frame is calculated by interpolating the pixel values of the previous and next frames.

[0027] For example, using the bilinear interpolation method, the pixel value of the corresponding position of the intermediate frame is calculated according to the four adjacent pixel values of the corresponding position pixels of the previous and next frames. If the frame rate of a certain section of video is too slow, some redundant frames are deleted appropriately. When deleting frames, frames that have less impact on the continuity of video content, such as frames with small changes in pictures, are preferentially selected. After these operations, the video jitter phenomenon is eliminated, and a time-continuous standardized video data stream is generated.

[0028] Step S120: calling a video analysis model to perform feature extraction on the standardized video data stream to generate a set of job environment features and a set of job personnel behavior features.

[0029] In the high-altitude maintenance operation scene of a large wind turbine, the trained video analysis model is used to extract features from the standardized video data stream to obtain feature information related to the job environment and the job personnel.

[0030] Step S121: dividing the standardized video data stream into a plurality of equal-length video segments, and performing a three-dimensional convolution feature extraction operation on each video segment to generate a spatiotemporal correlation feature matrix.

[0031] The standardized video data stream is segmented into video segments S1, S2, S3, etc. of a fixed length L. For each video segment Si, a three-dimensional convolution feature extraction operation is performed. The three-dimensional convolution operation performs convolution calculation in the time dimension (with a length of t), the horizontal spatial dimension (with a length of x), and the vertical spatial dimension (with a length of y) of the video segment. Let the convolution kernel be K, with a length of kt in the time dimension, a length of kx in the horizontal spatial dimension, and a length of ky in the vertical spatial dimension. For each frame in the video segment Si, the convolution kernel K slides at different positions and performs weighted calculation on the pixel values at each position. For example, for the nth pixel position of the mth frame in the video segment Si, the convolution kernel K covers the position and the surrounding pixels, and according to the weight w of each element in the convolution kernel K, the new pixel value Pnew is obtained by multiplying the weight w with the pixel value p of the covered position and summing the results, i.e. Pnew = Σ(w*p). In the time dimension, the convolution kernel K will span several adjacent frames for calculation, thereby capturing the spatio-temporal correlation information in the video segment.

[0032] Through the above three-dimensional convolution operation, each video segment Si generates a spatio-temporal correlation feature matrix Fi, which contains the feature information of the video segment in the spatio-temporal dimension.

[0033] Step S122: performing an attention weight distribution operation on the spatio-temporal correlation feature matrix to generate a visual focus distribution map with region importance annotation, and performing an edge tracking operation on the work equipment contour based on the visual focus distribution map to extract equipment operation state features and spatial displacement trajectory features.

[0034] In this embodiment, the attention weight distribution operation analyzes the importance of the region represented by each element in the spatio-temporal correlation feature matrix Fi to the work scene. For regions related to the work equipment, a higher weight is assigned, and for irrelevant background regions, a lower weight is assigned. For example, by analyzing the elements in the feature matrix related to the device shape, color, and other features, the importance weight is determined. In this way, a visual focus distribution map Vi with region importance annotation is generated.

[0035] In the wind turbine high-altitude maintenance operation scenario, edge tracking operation is performed on the outline of the operation equipment based on the visual focus distribution map Vi. The approximate position of the operation equipment is determined from the visual focus distribution map Vi, and then from this position, along the edge of the equipment, the outline of the equipment is tracked point by point according to the change of pixel value. In the tracking process, the position information of each point on the equipment outline is recorded, so as to extract the equipment running state features, such as whether the equipment is in a stationary, rotating or other running state, and if it is in a rotating state, the rotating speed. At the same time, by comparing the position information of the equipment outline at different time points, the spatial displacement trajectory features of the equipment are extracted, for example, the equipment moves from position P1 to position P2 in a period of time.

[0036] Step S123: Perform skeleton key point detection operation on the body action of the operator to generate a set of action amplitude features and motion speed features.

[0037] In the wind turbine high-altitude maintenance operation scenario, the operator performs various operations, and the skeleton key point detection method is used to analyze the image of the operator in the video data to identify the key points of each bone of the human body, such as wrist, elbow, shoulder, etc. According to the position change of these skeleton key points in different frames, the amplitude of the operator's body action is calculated.

[0038] For example, the change of the distance between the wrist key point and the elbow key point in adjacent frames is calculated to obtain the related data of the action amplitude. At the same time, according to the displacement of the skeleton key points in unit time, the motion speed of the operator's body action is calculated. Assuming that a certain skeleton key point moves from position P1 to position P2 in a time interval Δt, the distance d between the two points is calculated and divided by the time interval Δt to obtain the motion speed v=d / Δt. These action amplitude data and motion speed data are sorted to generate a set of action amplitude features and motion speed features.

[0039] Step S124: Converge the equipment running state features, spatial displacement trajectory features, action amplitude features and motion speed features to generate the operator behavior feature set.

[0040] The equipment running state features and spatial displacement trajectory features extracted from the operation equipment are summarized with the action amplitude features and motion speed features extracted from the body action of the operator. According to the set rules, these different types of features are integrated.

[0041] For example, the device running state features are first classified according to different types of devices, and then the spatial displacement trajectory features are correspondingly associated with them. For the action amplitude features and the motion speed features, they are classified and arranged according to different operation stages of the operating personnel. Then, these classified and arranged features are arranged in a set order, or are integrated after weight distribution according to their correlation degrees with operation safety, efficiency, etc., to finally generate the operating personnel behavior feature set.

[0042] Step S125: performing a texture analysis operation on the fixed obstacles in the operating environment to extract surface roughness features and geometric shape features.

[0043] In the high-altitude operating environment of a wind turbine generator set, there are some fixed obstacles, such as protruding parts of the tower drum, fixed supports of the equipment, etc. Texture analysis operation is performed on these fixed obstacles. By analyzing the gray level change, color distribution, etc. of the obstacle surface pixels, the surface roughness features are extracted. For example, the standard deviation of the surface pixel gray value is calculated, and the larger the standard deviation, the higher the surface roughness. At the same time, by identifying and analyzing the outline of the obstacle, its geometric shape features are determined, such as whether it is a cylinder, a cuboid, etc., and related size information such as length, width, height, etc.

[0044] Step S126: fusing the surface roughness features, geometric shape features and device running state features through a graph neural network to construct a dynamic interaction model of material properties and device kinematics parameters, and generating an environment interaction risk feature set.

[0045] The extracted surface roughness features, geometric shape features of the fixed obstacles and the device running state features of the operating equipment are fused by using a graph neural network. The graph neural network represents the relationship between different features through nodes and edges. The surface roughness features, geometric shape features and device running state features are taken as different nodes, and the connection mode and weight of the edges are determined according to the actual correlation between them. For example, if the device running state features have a close correlation with the geometric shape features of the obstacles in actual operation, then the edge weight connecting these two feature nodes is higher. In this way, a dynamic interaction model of material properties (reflected by surface roughness, etc.) and device kinematics parameters (reflected by device running state) is constructed. In this dynamic interaction model, the interaction between different features is analyzed, for example, how the device running state is affected by the material and shape of the obstacles. According to the analysis results, an environment interaction risk feature set is generated, which contains feature information related to risks that may be generated due to the interaction between the device and the obstacles in the operating environment, such as the possibility of collision risk, the influence degree of the risk, etc.

[0046] Step S130: constructing a job environment graph structure based on the set of job environment features, calling a graph convolution network to perform a multi-level feature aggregation operation on the job environment graph structure, and generating an environment state feature vector.

[0047] In the scenario of high-altitude maintenance of large wind turbine generators, a job environment graph structure is constructed according to the extracted set of job environment features, and a multi-level feature aggregation operation is performed on the graph convolution network to generate a feature vector that can reflect the environment state.

[0048] Step S131: setting the job equipment nodes, environmental obstacle nodes, and personnel position nodes as a set of basic nodes of the graph structure.

[0049] In this scenario, maintenance equipment of the wind turbine generator, such as a maintenance basket and a tool lifting platform, are respectively set as the job equipment nodes; environmental obstacles such as protruding structures of the tower and surrounding fixed facilities are set as the environmental obstacle nodes; and the positions of the workers at different times are set as the personnel position nodes. These nodes together constitute the set of basic nodes of the graph structure.

[0050] Step S132: constructing a first type of edge connection relationship according to the actual distance between the equipment nodes and the obstacle nodes, and constructing a second type of edge connection relationship according to the operation relevance between the personnel position nodes and the equipment nodes.

[0051] For the job equipment nodes and the environmental obstacle nodes, the actual distance between them is determined by measurement or a video image-based calculation method, thereby constructing the first type of edge connection relationship. For example, a laser ranging sensor or a video image-based calculation method is used to calculate the distance between the equipment nodes and the obstacle nodes according to the scale relationship and the principle of triangulation in the video image. The edge connection weight between nodes with a relatively short distance is relatively high, indicating that the spatial relationship between them is closer. For the personnel position nodes and the equipment nodes, the second type of edge connection relationship is constructed according to the operation relevance factors such as the operation frequency and operation duration of the workers on the equipment. If the worker frequently and for a long time operates a certain equipment, the edge connection weight between the personnel position node and the corresponding equipment node is high, indicating that the operation relevance between them is strong.

[0052] Step S133: constructing a set of node attribute features based on the environmental monitoring sensor data, the set of node attribute features including equipment operating parameters, obstacle material attribute, and personnel state parameters, the personnel state parameters including personnel protective equipment state parameters and personnel vital sign state parameters.

[0053] In the wind turbine high-altitude maintenance operation scene, various environmental monitoring sensors are arranged. Through these sensors, equipment operation parameters such as the lifting speed of the maintenance basket, the carrying weight of the tool lifting platform, and the like are obtained; the material properties of obstacles such as whether the obstacle is a metal material or a composite material and the like related information are obtained; the personnel state parameters are obtained, wherein the personnel protection equipment state parameters include the tightening degree of the safety belt, whether the safety helmet is worn in a standard manner, and the like, and the personnel vital sign state parameters include heart rate, blood pressure, and the like. These data are integrated to construct a node attribute feature set.

[0054] Step S134: combining the base node set, edge connection relationship, and node attribute feature set to generate the job environment graph structure.

[0055] Step S134-1: performing a vectorization encoding operation on the equipment operation parameters in the node attribute feature set to generate an equipment attribute vector.

[0056] For the equipment operation parameters in the node attribute feature set, such as the lifting speed of the maintenance basket, the carrying weight of the tool lifting platform, and the like, a vectorization encoding mode is adopted. Each parameter is mapped to a vector space, and each parameter is assigned a dimension. For example, assuming that the equipment operation parameters include lifting speed V and carrying weight W, the lifting speed V is mapped to the first dimension of the vector, the carrying weight W is mapped to the second dimension of the vector, and so on, to generate an equipment attribute vector.

[0057] Step S134-2: performing a texture feature extraction operation on the obstacle material properties in the node attribute feature set to generate an obstacle attribute vector.

[0058] The texture features of the obstacle material properties are extracted. By analyzing the gray level change mode and color distribution law of the obstacle surface pixels, the texture features are extracted. For example, a gray level co-occurrence matrix is calculated to describe the spatial relationship between pixels, and feature values such as contrast, correlation, and entropy are extracted from the gray level co-occurrence matrix, thereby forming a vector from these feature values to generate an obstacle attribute vector.

[0059] Step S134-3: performing a vectorization encoding operation on the personnel state parameters in the node attribute feature set to generate a personnel attribute vector.

[0060] For the personnel state parameters, including the personnel protective equipment state parameters and the personnel vital sign state parameters. The personnel protective equipment state parameters, such as whether the safety belt is fastened (which can be represented as 0 or 1, 0 indicating not fastened and 1 indicating fastened), whether the safety helmet is worn correctly (also represented as 0 or 1), and the personnel vital sign state parameters, such as heart rate H, blood pressure B, etc., are vectorized and coded. These parameters are arranged in a set order, such as setting the safety belt fastening state as the first element of the vector, the safety helmet wearing state as the second element, the heart rate H after a certain normalization processing (assuming the normalization function is Norm, which normalizes the heart rate H to 0-1, i.e. H_norm = Norm(H)), as the third element, and the blood pressure B after similar normalization processing (assuming B_norm = Norm(B)) as the fourth element, and so on, to generate the personnel attribute vector. The normalization processing here is to ensure that parameters of different dimensions are processed on the same scale, ensuring the uniformity of the subsequent calculation dimension and the matching of the feature dimension.

[0061] Step S134-4: After the device attribute vector, the obstacle attribute vector and the personnel attribute vector are respectively mapped to a unified feature dimension space through a fully connected layer, they are spliced in the channel dimension to generate a node feature matrix.

[0062] The initial dimensions of the device attribute vector, the obstacle attribute vector and the personnel attribute vector may not be consistent due to the different parameters they contain. In order to effectively fuse them, they need to be mapped to a unified feature dimension space.

[0063] Taking the device attribute vector as an example, its initial dimension is d1. Through a fully connected layer, the fully connected layer contains multiple neurons, each neuron is connected to each element of the device attribute vector and performs weighted calculation through a weight matrix W1. For each element in the device attribute vector, multiply the corresponding weight in the weight matrix W1 and add it, and then pass it through an activation function (assuming Act) for processing, such as Act(∑(Wi1*ei1)), where Wi1 is the weight in the weight matrix W1, ei1 is the element in the device attribute vector, to obtain a new vector processed through the fully connected layer, whose dimension is set as D. The same method is applied to the obstacle attribute vector and the personnel attribute vector, which are processed through the fully connected layer using weight matrices W2 and W3, respectively, to obtain new vectors with a dimension of D.

[0064] Then, the three vectors processed through the fully connected layer are spliced in the channel dimension. Assuming that the device attribute vector processed is V1, the obstacle attribute vector processed is V2, and the personnel attribute vector processed is V3, they are arranged in order, such as [V1, V2, V3], to generate a node feature matrix.

[0065] Step S134-5: Calculate the spatial correlation degree between the device nodes and the obstacle nodes according to the actual distance value of the first type of edge connection relationship, and generate a first type of edge weight matrix.

[0066] For the actual distance value between the device nodes and the obstacle nodes in the first type of edge connection relationship, let the device node be Ni, the obstacle node be Nj, and the distance between them be dij. To calculate the spatial correlation degree, a distance-related function (denoted as Func1) is used, which returns a value representing the spatial correlation degree according to the size of the distance value. For example, Func1(dij)=1 / (1+dij), the smaller the distance, the greater the spatial correlation degree. For each pair of device nodes and obstacle nodes, the spatial correlation degree value is calculated through the function. Arrange these spatial correlation degree values in the order of node pairs to form a matrix, which is the first type of edge weight matrix. For example, if there are n device nodes and m obstacle nodes, the size of the first type of edge weight matrix is n×m, and the element wij in the first type of edge weight matrix represents the spatial correlation degree weight between the device node Ni and the obstacle node Nj.

[0067] Step S134-6: Calculate the interaction intensity value between the personnel location nodes and the device nodes according to the operation correlation parameters of the second type of edge connection relationship, and generate a second type of edge weight matrix.

[0068] For the operation correlation parameters between the personnel location nodes and the device nodes in the second type of edge connection relationship, let the personnel location node be Nk, the device node be Ni, and the operation correlation parameters include operation frequency fki, operation duration tki, etc. A comprehensive function (denoted as Func2) is used to calculate the interaction intensity value. For example, Func2(fki, tki)=α*fki+β*tki, where α and β are weight coefficients set according to actual conditions, used to balance the influence of operation frequency and operation duration on interaction intensity. For each pair of personnel location nodes and device nodes, the interaction intensity value is calculated through the function. Arrange these interaction intensity values in the order of node pairs to form a matrix, which is the second type of edge weight matrix. Assuming that there are p personnel location nodes and n device nodes, the size of the second type of edge weight matrix is p×n, and the element zki in the matrix represents the interaction intensity weight between the personnel location node Nk and the device node Ni.

[0069] Step S134-7: Perform graph topology mapping on the node feature matrix, the first type of edge weight matrix, and the second type of edge weight matrix to generate initial graph structure data.

[0070] Each element in the node feature matrix corresponds to a node in the graph structure, and each row of the node feature matrix represents a feature vector of a node. The elements in the first type of edge weight matrix and the second type of edge weight matrix correspond to the edges between the corresponding node pairs in the graph structure, respectively. For example, the i-th row of the node feature matrix corresponds to the node Ni in the graph structure, the element wji in the i-th row and the j-th column of the first type of edge weight matrix corresponds to the weight of the edge between the device node Ni and the obstacle node Nj, and the element zki in the k-th row and the i-th column of the second type of edge weight matrix corresponds to the weight of the edge between the personnel position node Nk and the device node Ni. Through this correspondence, the matrix data is mapped to the graph structure to generate initial graph structure data. In this process, the correspondence between the node features and the edge weights is ensured to be accurate to reflect the actual relationship between the elements in the work environment.

[0071] Step S134-8: Perform inter-node reachability verification on the initial graph structure data, delete edge connection relationships that do not meet the preset connectivity threshold, supplement missing key edge connection relationships, and generate a fully connected work environment graph structure.

[0072] The inter-node reachability verification is performed on the initial graph structure data, that is, it is checked whether all other nodes can be reached from any one node through a series of edge connections. Let the preset connectivity threshold be T. For each edge connection relationship, it is judged whether the connectivity requirement is met according to the edge weight. For example, if the weight of an edge is lower than the threshold T, it is considered that the edge connection relationship is weak and may have a small contribution to the connectivity of the overall graph structure, and it is deleted. At the same time, through the analysis of the logical relationship between the nodes and the actual demand of the work scene, it is determined whether there is a missing key edge connection relationship. For example, in the maintenance scene of a wind turbine generator, if there is no direct connection between a device node and another device node that has an important influence on the operation of the device node, but there should be an association from the work logic, then this edge connection relationship needs to be supplemented. By continuously adjusting the edge connection relationship, a fully connected work environment graph structure is finally generated, which ensures that the graph structure can accurately reflect the connectivity and mutual relationship between the elements in the work environment.

[0073] Step S134-9: Perform dynamic incremental update on the fully connected work environment graph structure according to the real-time update frequency of the device operation parameters, and generate a dynamic work environment graph structure with a timestamp version identifier.

[0074] The device operating parameters change in real time as the work progresses, and the update frequency is set to f update. Whenever the device operating parameters are updated, the corresponding device operating parameter values in the node attribute feature set are first updated. For example, if the lifting speed of the maintenance basket changes, the parameter value representing the lifting speed in the node attribute feature set is modified. Then, based on the updated device operating parameters, the relevant quantities are recalculated. For example, the actual distance from the obstacle node is recalculated based on the new position of the device node, and the weight of the first-type edge connection relationship is recalculated according to the method of calculating the first-type edge weight matrix as described above. For a fully connected work environment graph structure, a dynamic incremental update method is used, i.e., only the affected part is updated, rather than rebuilding the entire graph structure. During the update process, a timestamp version identifier is added for each update, such as recording the update times t1, t2, etc., to generate a dynamic work environment graph structure with a timestamp version identifier. In this way, the changes in the work environment graph structure over time can be recorded, facilitating subsequent analysis and decision-making.

[0075] Step S135: performing a feature aggregation operation on the directly adjacent nodes of each node in the work environment graph structure at the initial convolution layer of the graph convolution network to generate a primary aggregation feature vector.

[0076] At the initial convolution layer of the graph convolution network, for each node in the work environment graph structure, let the node be N. The directly adjacent nodes refer to the nodes directly connected to the node through edges. For node N, the feature vectors of its directly adjacent nodes are collected, let the directly adjacent nodes be N1, N2,..., Nx, and their feature vectors be V1, V2,..., Vx, respectively. At the same time, the weights of the edges between node N and these directly adjacent nodes are obtained, let them be w1, w2,..., wx. An aggregation method (let it be Aggregate1) is used, such as the weighted summation method, to calculate the primary aggregation feature vector V primary. That is, V primary = Aggregate1(w1*V1+w2*V2+...+wx*Vx), where the weighting is to consider the importance of different adjacent nodes to the current node, and the greater the edge weight, the greater the contribution of the corresponding adjacent node feature in aggregation. In this way, a primary aggregation feature vector is generated for each node in the work environment graph structure, which contains the comprehensive feature information of the node and its directly adjacent nodes.

[0077] Step S136: performing a cross-level feature propagation operation on the indirectly adjacent nodes separated by two edge connection relationships at the intermediate convolution layer of the graph convolution network to generate a middle aggregation feature vector.

[0078] In the intermediate convolutional layer of the graph convolutional network, the indirect adjacent relationship of the nodes in the job environment graph structure is considered, which refers to the indirect adjacent nodes separated by two edge connection relationships. Let the current node be N, and the indirect adjacent nodes separated by two edge connection relationships be M1, M2,..., My.

[0079] First, the path from the node N to these indirect adjacent nodes is determined. For example, the indirect adjacent node M is reached from the node N through the intermediate node N'. For each indirect adjacent node M, its feature vector VM is obtained, as well as the edge weight on the path from the node N to M. Let the edge weight from the node N to the intermediate node N' be w1, and the edge weight from the intermediate node N' to the indirect adjacent node M be w2. A cross-level feature propagation method (set as Propagate1) is used to propagate the features of the indirect adjacent nodes to the current node N, for example, by multiplying the weights and transforming the feature vectors. That is, for each indirect adjacent node M, the propagated feature vector VM' is calculated as Propagate1(w1*w2*VM).

[0080] Then, the propagated feature vectors of all indirect adjacent nodes are aggregated (set the aggregation method as Aggregate2), for example, by weighted summation or other appropriate aggregation methods, to generate the intermediate aggregation feature vector V_intermediate. That is, V_intermediate=Aggregate2(VM1'+VM2'+...+VMy'), which contains the comprehensive influence feature information of the indirect adjacent nodes separated by two edge connection relationships on the current node, enriching the feature representation of the node.

[0081] Step S137: In the final convolutional layer of the graph convolutional network, the primary aggregation feature vector and the intermediate aggregation feature vector are fused to generate an environment state feature vector containing global dependency relationships.

[0082] In the final convolutional layer of the graph convolutional network, the primary aggregation feature vector generated by the initial convolutional layer and the intermediate aggregation feature vector generated by the intermediate convolutional layer are fused. Let the primary aggregation feature vector be V_primary, and the intermediate aggregation feature vector be V_intermediate. A fusion method (set as Fuse1) is used to fuse them, for example, by concatenation or weighted addition (the weighting coefficient is determined according to the actual situation). If the concatenation method is used, the primary aggregation feature vector and the intermediate aggregation feature vector are concatenated in order to form a new vector V_fused.

[0083] Then, the fused vector is further processed by a fully connected layer (with weight matrix W fuse) and an activation function (set as Act fuse) to obtain an environment state feature vector V environment containing global dependency relationships. That is, V environment = Act fuse(W fuse*V fused). The environment state feature vector integrates the feature information of the direct adjacent nodes and the indirect adjacent nodes connected by two edges, reflecting the global dependency relationships between the nodes in the job environment graph structure.

[0084] Step S138: Perform dimensionality reduction visualization processing on the spatial coordinate encoding information embedded in the environment state feature vector to generate an environment risk heat map and label the high-risk area coordinates.

[0085] The environment state feature vector contains spatial coordinate encoding information of each element in the job environment, denoted as S. Since the dimensionality of this information may be high, it is not conducive to intuitive analysis and needs to be processed for dimensionality reduction visualization. A dimensionality reduction algorithm (denoted as Reduce1) is used, such as a method similar to principal component analysis (PCA), to convert the high-dimensional spatial coordinate encoding information S into low-dimensional representation S low.

[0086] Then, the environment risk heat map is generated according to the dimensionality-reduced information. For each position in space, a color value is assigned to represent the risk level according to its corresponding dimensionality-reduced feature value, with darker colors indicating higher risk. At the same time, by analyzing the color distribution in the heat map, the high-risk area is determined. The determination threshold of the high-risk area is denoted as T risk, and for the area in the heat map whose color value exceeds the threshold T risk, its coordinate position is determined and labeled. For example, the top-left corner coordinates and the bottom-right corner coordinates of the high-risk area are labeled to clearly indicate the position of the high-risk area in the job environment.

[0087] Step S139: Perform spatial clustering analysis on the coordinate data of the environment risk heat map to identify risk aggregation areas and generate region boundary labeling information.

[0088] The coordinate data of the environmental risk heat map is subjected to spatial clustering analysis, and a spatial clustering algorithm (referred to as Cluster1) such as a method similar to DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is used. According to the distance and density relationship between the coordinate data points, the points in space are divided into different clusters. For each cluster, it is determined whether it is a risk aggregation area. If the points in a cluster correspond to generally high risk values in the heat map, and the size of the cluster reaches a certain degree (the minimum cluster size threshold is set as Min_size), it is considered that the cluster is a risk aggregation area. For each risk aggregation area, the area boundary is determined. For example, the convex hull of the outermost points in the cluster is calculated to determine the area boundary, and area boundary annotation information is generated. The annotation information includes the vertex coordinates of the area boundary, the shape description of the boundary, and the like, so as to accurately represent the range and shape of the risk aggregation area.

[0089] Step S1390: updating the node attribute feature set of the job environment graph structure based on the area boundary annotation information, and generating an optimized environmental state feature vector.

[0090] According to the area boundary annotation information of the risk aggregation area, the node attribute feature set of the job environment graph structure is updated. For nodes located in the risk aggregation area or nodes closely associated with the risk aggregation area, the node attribute features are adjusted. For example, if a device node is located in the risk aggregation area, the risk-related attribute value of the device node can be increased. Let the node attribute feature set be Attributes, and for the relevant node Ni, according to the situation of the risk aggregation area, the attributes of the node are adjusted by an adjustment function (referred to as Adjust1), that is, Attributes[Ni]=Adjust1(Attributes[Ni], area boundary annotation information).

[0091] Then, based on the updated node attribute feature set, the relevant operations in the graph convolution network are recalculated, such as the feature vectors of the nodes and the weights of the edges. Through the same graph convolution network processing flow as before, from the initial convolution layer to the final convolution layer, an optimized environmental state feature vector is generated. The optimized environmental state feature vector more accurately reflects the influence of the risk distribution in the job environment on each node.

[0092] Step S140: generating a dynamic emergency scheduling strategy according to the environmental state feature vector and the job personnel behavior feature set.

[0093] In the high-altitude maintenance operation scene of large wind turbine generators, based on the obtained environment state feature vector and the operation personnel behavior feature set, a dynamic emergency scheduling strategy adapting to the actual situation of the operation site is generated.

[0094] Step S141: After inputting the environment state feature vector and the operation personnel behavior feature set into the linear projection layer respectively to align to the same feature dimension, input the pre-trained spatio-temporal correlation model, perform feature interaction verification on the node attribute features in the environment state feature vector and the dynamic parameters in the operation personnel behavior feature set through the cross-attention mechanism, and generate an interaction matrix.

[0095] The environment state feature vector and the operation personnel behavior feature set may not be consistent in feature dimension due to different sources and compositions. First, set the environment state feature vector as V_env and the operation personnel behavior feature set as Set_behave. Input them into the linear projection layer respectively. For the environment state feature vector V_env, the linear projection layer performs weighted calculation through the weight matrix W1, that is, V_env_projected=W1*V_env, so that its dimension becomes the same as that of the operation personnel behavior feature set after processing, that is, dimension D.

[0096] For the operation personnel behavior feature set Set_behave, suppose it contains multiple feature vectors, such as V1_behave, V2_behave,..., Vn_behave. For each feature vector, use the weight matrix W2 through the linear projection layer to obtain V1_behave_projected=W2*V1_behave, V2_behave_projected=W2*V2_behave,..., Vn_behave_projected=W2*Vn_behave, and combine these projected feature vectors into a vector V_behave_projected with the same dimension as the projected environment state feature vector.

[0097] Then, the projected environment state feature vector V_env_projected and the projected worker behavior feature set V_behave_projected are input into a pre-trained spatio-temporal correlation model. The spatio-temporal correlation model verifies the feature interaction between the two through a cross-attention mechanism. For each node attribute feature in the environment state feature vector, denoted as Attr_env, and each dynamic parameter in the worker behavior feature set, denoted as Param_behave, the spatio-temporal correlation model calculates the attention weight between them. For example, through an attention calculation function (denoted as Attention1), the attention weight w between Attr_env and Param_behave is calculated. w = Attention1(Attr_env, Param_behave). The attention weights between all node attribute features and dynamic parameters are combined into a matrix, i.e., an interaction matrix Matrix_interaction. The interaction matrix reflects the degree of mutual correlation between the environment state and the worker behavior.

[0098] Step S142: Based on the correlation weight values of each node and behavior feature in the interaction matrix, the risk correlation degree parameters between the high-risk nodes in the work environment graph structure and the abnormal behavior features of the workers are extracted.

[0099] From the interaction matrix Matrix_interaction, for each node in the work environment graph structure, denoted as Nj, find the corresponding correlation weight value. These correlation weight values reflect the degree of correlation between the node and the worker behavior features. For those nodes marked as high-risk nodes (for example, nodes in the high-risk area determined in the previous generation of environment risk heat map and subsequent analysis), denote the high-risk node set as Set_high-risk, extract their correlation weight values with each behavior feature in the worker behavior feature set from the interaction matrix.

[0100] For each behavior feature in the behavior feature set of the worker, denoted as Behave attr, the associated weight value of each node Nj in the high-risk node set Set high-risk and Behave attr is obtained from the interaction matrix, denoted as wNj_Behave attr. By analyzing and calculating these associated weight values, the risk correlation degree parameter is extracted. For example, a comprehensive calculation method (denoted as Compute1) is used to calculate the associated weight values of all nodes in the high-risk node set Set high-risk and a behavior feature Behave attr, such as calculating their average value or weighted average value (if different weights are given according to the importance of the nodes and other factors). Denote the weight of node Nj as aNj, then for each node Nj in Set high-risk, that is, for each node Nj in Set high-risk, Risk_correlation_Behave_attr=Compute1 (Σ(αNj*wNj_Behave_attr), thus the risk correlation degree parameter set between the high-risk nodes in the work environment graph structure and the abnormal behavior features of the workers is obtained, each parameter corresponds to a behavior feature of the worker, and reflects the risk correlation closeness between the high-risk nodes and the corresponding behavior features.

[0101] Step S143: performing an adaptive scaling operation on the node space coordinate features in the environment state feature vector according to the risk correlation degree parameters, to generate a risk-weighted spatial distribution feature vector.

[0102] For the environment state feature vector V_env, which contains node space coordinate features, denoted as Coords. According to the risk correlation degree parameter set calculated above, an adaptive scaling operation is performed on each node space coordinate in Coords. Denote the risk correlation degree parameter set as Set_Risk_correlation, and for each node coordinate in Coords, denoted as Coord_i, find the corresponding risk correlation degree parameter Risk_correlation_i (for example, by node identification or some kind of correspondence).

[0103] A scaling function (set as Scale1) is used to scale the node coordinates according to the risk correlation parameter. For example, for two-dimensional coordinates (x, y), the new coordinates (x_new, y_new) = Scale1 (x, y, Risk_correlation_i). The scaling function Scale1 here may scale the coordinates in different directions to different degrees according to the value of the risk correlation parameter. If the risk correlation parameter is larger, it means that the area where the node is located has a higher risk correlation with the personnel behavior anomaly feature, and the coordinates may be scaled to a greater extent to highlight the importance of the area in the spatial distribution. By performing such operations on all node coordinates in Coords, a risk-weighted spatial distribution feature vector V_risk_weighted is generated, which reflects the relationship between different areas and risk correlation in the spatial distribution.

[0104] Step S144: input the risk-weighted spatial distribution feature vector and the edge connection relationship of the work environment graph structure into the graph attention network, and generate a topological optimization feature map containing the emergency obstacle avoidance path between nodes through an edge weight dynamic adjustment algorithm.

[0105] The risk-weighted spatial distribution feature vector V_risk_weighted and the edge connection relationship of the work environment graph structure (including complete edge connection information composed of the first type of edge connection relationship and the second type of edge connection relationship) are input into the graph attention network. The graph attention network will dynamically adjust the edge weight in the graph structure according to the input information.

[0106] For each edge in the graph structure, set as Edge_ij, and the initial edge weight is w_ij. The graph attention network combines the information of the nodes in the risk-weighted spatial distribution feature vector V_risk_weighted and the features of the nodes at both ends of the edge through an edge weight dynamic adjustment algorithm (set as Adjust_weight1) to recalculate the weight of the edge. For example, the algorithm Adjust_weight1 will consider the positional relationship of the nodes at both ends of the edge in the risk-weighted spatial distribution, the attribute features of the nodes themselves, and the correlation with other nodes, and calculate the new edge weight w_ij_new = Adjust_weight1 (V_risk_weighted, Edge_ij, w_ij).

[0107] A new graph structure representation is generated by adjusting all edge weights. In the new graph structure, the emergency obstacle avoidance paths between nodes are found. A path search algorithm (set as Search_path1), for example, a method similar to A* algorithm, is used to search the shortest or optimal emergency obstacle avoidance paths from one node to other nodes according to the adjusted edge weights and the connectivity between nodes. The path information is integrated into a feature map to generate a topology-optimized feature map Feature_map_topology containing the emergency obstacle avoidance paths between nodes. The topology-optimized feature map shows the emergency obstacle avoidance path information between nodes in the job environment graph structure considering the risk-weighted spatial distribution and dynamic adjustment of edge weights.

[0108] Step S145: Based on the path connectivity parameters in the topology-optimized feature map, the Dijkstra algorithm is called to calculate the shortest safe movement path sequence of the job terminal device in the job environment graph structure in combination with the edge weights of the graph attention network, and the device motion parameter constraints corresponding to the shortest safe movement path sequence are generated.

[0109] The path connectivity parameters are extracted from the topology-optimized feature map Feature_map_topology, which describe the connectivity state, length, risk level, and other information of each path in the graph structure. Set the path connectivity parameter set as Set_connectivity.

[0110] The Dijkstra algorithm is called in combination with the adjusted edge weights of the graph attention network (set the edge weight matrix as W_edges) to calculate the shortest safe movement path sequence of the job terminal device from the current position to the target position (for example, a safe area or a position where an emergency operation needs to be performed) in the job environment graph structure. The Dijkstra algorithm starts from the starting node and gradually explores the nodes in the graph structure, and calculates the distance to each node according to the edge weights (here, the distance can be understood as the path cost from the starting node to the node, which is related to risk, distance, etc.).

[0111] For each node in the graph structure, set as Node_k, the shortest path distance from the starting node to Node_k (set as Distance_k) and the predecessor node on the path (set as Predecessor_k) are recorded during the execution process. By continuously updating these values, the shortest safe movement path sequence from the starting node to the target node is finally determined. Set the shortest safe movement path sequence as Sequence_path, which is composed of a series of nodes, for example, Sequence_path=[Node_start, Node_1, Node_2, …, Node_end].

[0112] According to the position relationship of nodes in the shortest safe mobile path sequence Sequence_path and the actual situation of the working environment, device motion parameter constraints corresponding to the path sequence are generated. For example, the distance between nodes, the change of angle, and the physical limitations of the device are considered. Let the device motion parameters include speed V, acceleration A, turning angle θ, etc. For adjacent nodes Node_i and Node_i+1 in the path sequence, according to their spatial position relationship, the approximate speed range, acceleration range, and turning angle range required for the device to move from Node_i to Node_i+1 are calculated. For example, by calculating the distance d = Distance(Node_i, Node_i+1) between the two points and the expected moving time t, the speed range V_min <= V <= V_max can be obtained, where V_min and V_max are determined according to the distance d, time t, and device performance, etc. The acceleration range and turning angle range are also obtained by similar calculation methods based on node position relationship and device characteristics. These device motion parameter constraints are arranged into a set, denoted as Set_constraints.

[0113] Step S146: Perform dynamic compatibility check on the device motion parameter constraints and real-time device running parameters, eliminate path segments that exceed the physical limits of the device according to the check results, and generate a dynamic emergency scheduling strategy containing feasible path coordinates and device control instruction set.

[0114] Get real-time device running parameters, denoted as Set_real-time, which includes device current speed V_real, acceleration A_real, turning angle θ_real, etc. Perform dynamic compatibility check on the device motion parameter constraint set Set_constraints and the real-time device running parameter set Set_real-time.

[0115] For each device motion parameter constraint in Set_constraints, let the speed constraint be V_min <= V <= V_max, the acceleration constraint be A_min <= A <= A_max, and the turning angle constraint be θ_min <= θ <= θ_max. Compare the speed V_real, acceleration A_real, and turning angle θ_real in the real-time device running parameters with these constraints respectively.

[0116] If a parameter in the real-time device running parameters exceeds the corresponding constraint range, for example, V_real < V_min or V_real > V_max, it means that the current path segment may pose a risk to the device or exceed the physical limits of the device. For path segments that exceed the physical limits of the device, eliminate them from the shortest safe mobile path sequence Sequence_path.

[0117] After the check and elimination operations on all path segments, a path sequence containing feasible path coordinates is obtained, which is set as Sequence_feasible_path. According to the feasible path sequence and the device motion parameter constraint conditions, a device control instruction set is generated. For example, for each path segment in the feasible path sequence, according to the corresponding device motion parameter constraint, the corresponding speed control instruction, acceleration control instruction, turning angle control instruction, etc. are generated. Set the device control instruction set as Set_commands, which contains a series of control instructions for different motion parameters of the device.

[0118] Combine the feasible path coordinates and the device control instruction set to generate a dynamic emergency scheduling strategy Strategy_emergency containing feasible path coordinates and device control instruction set. The dynamic emergency scheduling strategy considers the real-time running state and physical limit of the device, and provides specific guidance for the safe movement of the job terminal device in emergency situations.

[0119] Step S150: Perform validity verification operation on the dynamic emergency scheduling strategy to obtain strategy verification result, and generate emergency scheduling instruction set based on the strategy verification result, and transmit the emergency scheduling instruction set to the job terminal device to trigger emergency response operation.

[0120] In the high-altitude maintenance operation scenario of large wind turbine generators, the generated dynamic emergency scheduling strategy Strategy_emergency is verified for effectiveness to ensure that the dynamic emergency scheduling strategy can be effectively executed and is safe and reliable in actual application.

[0121] Step S151: Extract the feasible path coordinate sequence and device motion parameter constraint conditions in the dynamic emergency scheduling strategy, and construct a three-dimensional dynamic scene reconstruction model according to the spatial coordinate system of the job environment graph structure.

[0122] Extract the feasible path coordinate sequence Sequence_feasible_path and the device motion parameter constraint condition set Set_constraints from the dynamic emergency scheduling strategy Strategy_emergency. According to the spatial coordinate system defined by the job environment graph structure (set the spatial coordinate system as Coordinate_system, which determines the position representation of each point in the job environment), construct a three-dimensional dynamic scene reconstruction model Model_reconstruction.

[0123] In constructing the model, first, the nodes in the job environment graph structure (including job equipment nodes, environmental obstacle nodes, and personnel position nodes) are arranged according to their coordinate positions in the spatial coordinate system. For a job equipment node, a corresponding three-dimensional model representation is created in space according to its corresponding equipment attribute characteristics (such as the size, shape, etc. of the equipment). For example, if it is a maintenance basket equipment node, a three-dimensional basket model is created at the corresponding coordinate position according to the actual size and shape of the basket. For an environmental obstacle node, a corresponding three-dimensional obstacle model is created according to its geometric shape characteristics and material properties, etc.

[0124] Then, according to the feasible path coordinate sequence Sequence_feasible_path, the expected movement path of the job terminal equipment is marked in the three-dimensional scene. For each coordinate point in the path sequence, the corresponding position is found in the three-dimensional scene, and is connected by a line or other visualization method to form a representation of the movement path of the equipment. At the same time, the equipment motion parameter constraint condition Set_constraints is associated with the equipment model in the three-dimensional scene to provide a parameter basis for subsequent simulation operations. In this way, a three-dimensional dynamic scene reconstruction model Model_reconstruction is constructed that can reflect the job environment and the expected motion of the equipment, providing a visual and operable virtual environment for subsequent strategy verification.

[0125] Step S152: input the feasible path coordinate sequence and the equipment motion parameter constraint condition into the preloaded physical engine simulator, segment the simulation task based on the path curvature key points, perform virtual motion simulation operations in a segmented parallel computing mode, and generate simulation motion trajectory data and a virtual equipment state parameter set.

[0126] The extracted feasible path coordinate sequence Sequence_feasible_path and the equipment motion parameter constraint condition set Set_constraints are input into the preloaded physical engine simulator (set the physical engine simulator as Simulator). The physical engine simulator can simulate the motion of objects in a physical environment.

[0127] First, the feasible path coordinate sequence is analyzed to find the path curvature key points. Set the path curvature calculation function as Curvature1, and calculate the curvature value of each point on the path through this function. For points with large curvature value changes, they are determined as path curvature key points. Set the path curvature key point set as Set_key_points.

[0128] The simulation task is segmented based on the path curvature key points Set_key_points. The path is divided into multiple segments, denoted as Segment_1, Segment_2, …, Segment_n, and each segment is bounded by path curvature key points.

[0129] According to the segmented parallel computing mode, virtual motion simulation operations are performed on each path segment in the physical engine simulator. For each path segment Segment_i, the initial motion state and motion parameters of the device are set in the physical engine simulator according to the parameters in the device motion parameter constraint condition Set_constraints, such as speed, acceleration, steering angle, etc. For example, for the speed constraint V_min <= V <= V_max, the initial speed of the device at the starting point of the path segment is set to V_start, where V_min <= V_start <= V_max.

[0130] In the physical engine simulator, the motion process of the device on the path segment is simulated according to the physical laws and the set parameters. During the motion process, the position, velocity, acceleration, etc. of the device are recorded, and the simulation motion trajectory data Trajectory_data is generated. At the same time, other related state parameters of the device during the motion process are recorded, such as the attitude of the device, the force condition, etc., and the virtual device state parameter set Set_virtual_state is generated. Through parallel simulation operations on all path segments, the simulation motion trajectory data and the virtual device state parameter set of the entire feasible path are obtained.

[0131] Step S153: Perform spatial collision detection operations on the simulation motion trajectory data and the obstacle node coordinates in the work environment graph structure to generate a set of trajectory conflict point coordinates and a conflict timestamp sequence.

[0132] The simulation motion trajectory data Trajectory_data and the obstacle node coordinates in the work environment graph structure are subjected to spatial collision detection operations. For each point in the simulation motion trajectory data, denoted as Point_trajectory, its coordinate position in the spatial coordinate system is obtained. For each obstacle node in the work environment graph structure, denoted as Obstacle_node, its coordinate position and the size information of the obstacle (determined according to the obstacle geometric shape features and size information determined when the work environment graph structure was constructed previously) are obtained.

[0133] For example, a spatial collision detection algorithm (denoted as Collision_detect1) is used to determine whether the point collides with the obstacle according to the coordinate position of the point and the position and size information of the obstacle. For example, if the point is located within the spatial range occupied by the geometric shape of the obstacle, it is determined that a collision has occurred.

[0134] For the point of collision, record its coordinate position, generate the trajectory conflict point coordinate set Set_conflict_points. At the same time, record the time point of collision (according to the time information in the simulation motion trajectory data), generate the conflict timestamp sequence Sequence_conflict_timestamps. The trajectory conflict point coordinate set and the conflict timestamp sequence can accurately indicate the specific position and time of the conflict between the device motion trajectory and the obstacle in the simulation process.

[0135] Step S154: Perform a dynamic comparison operation between the virtual device state parameter set and the physical limit threshold of the device running parameter, and generate a device over-limit running parameter identifier and the corresponding time interval.

[0136] Perform a dynamic comparison operation between the parameters in the virtual device state parameter set Set_virtual_state and the physical limit threshold of the device running parameter. The physical limit threshold of the device running parameter is determined according to the design specifications and safety standards of the device, and the physical limit threshold of the speed is V_limit, the physical limit threshold of the acceleration is A_limit, the physical limit threshold of the steering angle is θ_limit, etc.

[0137] For the speed parameter V_virtual, the acceleration parameter A_virtual, the steering angle parameter θ_virtual, etc. in the virtual device state parameter set, compare them with the corresponding physical limit threshold respectively. If V_virtual>V_limit, mark the speed parameter over-limit, generate a device over-limit running parameter identifier (set as Flag_V_overlimit), and record the time interval of speed over-limit (according to the time information in the simulation motion trajectory data, determine the time points of the start and end of speed over-limit, set as Time_start_V_overlimit and Time_end_V_overlimit). Similarly, for the acceleration and steering angle parameters, perform similar comparison and recording operations.

[0138] Through such a dynamic comparison operation, the device over-limit running parameter identifier set Set_flags_overlimit and the time interval set Set_time_intervals corresponding to each identifier are generated. These identifiers and time intervals can clearly reflect whether the device running parameter exceeds the physical limit and the specific time range of over-limit in the simulation process.

[0139] Step S155: Fuse the trajectory conflict point coordinate set, the conflict timestamp sequence, and the device over-limit running parameter identifier to generate a strategy verification abnormal report, and extract the abnormal path segment coordinates and the abnormal device control instructions.

[0140] The trajectory conflict point coordinate set Set_conflict_points, the conflict timestamp sequence Sequence_conflict_timestamps, and the device over-limit running parameter identification set Set_flags_overlimit are fused to generate a strategy verification abnormality report Report_abnormality. In the report, the coordinate positions of the trajectory conflict points, the timestamps of the conflict occurrence, and the identification information of the device over-limit running parameters are recorded in detail.

[0141] According to the trajectory conflict point coordinate set and the conflict timestamp sequence, an abnormal path segment is determined. For example, if there is a collision risk between the path segments of two trajectory conflict points, the path segment is determined as an abnormal path segment. The coordinate information of the abnormal path segment is extracted, and the abnormal path segment coordinate set is Set_abnormal_path_coords.

[0142] At the same time, according to the device over-limit running parameter identification and the corresponding time interval, and the relationship between the device motion parameter constraint condition and the device control instruction set, an abnormal device control instruction that may cause the device to over-limit run is extracted. The abnormal device control instruction set is Set_abnormal_commands. The strategy verification abnormality report, the abnormal path segment coordinate set, and the abnormal device control instruction set provide specific problem information for subsequent adjustment and optimization of the dynamic emergency scheduling strategy.

[0143] Step S156: performing segmented smooth interpolation processing on the feasible path coordinate sequence according to the abnormal path segment coordinates to generate a replacement path coordinate sequence and recalculate the device motion parameter constraint condition.

[0144] According to the abnormal path segment coordinate set Set_abnormal_path_coords, the segmented smooth interpolation processing is performed on the feasible path coordinate sequence Sequence_feasible_path. For each abnormal path segment, the starting point coordinate is Coord_start, and the ending point coordinate is Coord_end.

[0145] A segmented smooth interpolation algorithm (set as Interpolation1) is used to insert a series of new coordinate points between the starting point and the ending point of the abnormal path segment to smooth the path. For example, through a spline interpolation method, the coordinates of the inserted points are calculated according to the positions of the starting point and the ending point and the overall trend of the path. The inserted new coordinate point set is Set_inserted_points.

[0146] Combining these insertion points with the part of the original feasible path coordinate sequence except the abnormal path segment, a new path coordinate sequence Sequence_alternative_path is generated.

[0147] Then, the device motion parameter constraints are recalculated according to the alternative path coordinate sequence Sequence_alternative_path. The geometric characteristics of the alternative path, such as the length of the path, the change of curvature, etc., and the actual situation of the work environment, such as the distribution of obstacles, space limitations, etc., are considered.

[0148] For adjacent coordinate points on the path, denoted as Coord_i and Coord_i+1, the approximate range of speed is determined by calculating the distance d = Distance(Coord_i, Coord_i+1) between the two points and the expected passing time t. Assuming that the device needs to pass this path segment safely and reasonably within a certain time, according to the relationship between distance and time, the lower speed limit V_min and the upper speed limit V_max can be obtained, i.e. V_min <= V <= V_max, where V_min and V_max are determined based on the path length, device performance, safety requirements, etc.

[0149] For acceleration constraints, the curvature change of the path is analyzed. If the path curvature changes greatly, it means that the device needs a large acceleration to adjust the motion direction when passing this path segment. Denote the path curvature as Curvature, and through a curvature-related calculation method (denoted as Acceleration_compute), the lower limit A_min and the upper limit A_max of acceleration are calculated according to the curvature value and the turning ability of the device, i.e. A_min <= A <= A_max, to ensure that the device can meet the path requirements while not exceeding its physical limits.

[0150] The turning angle constraint is also determined according to the geometric shape of the path. For the change of path direction composed of adjacent coordinate points, the range of turning angle required by the device is calculated. Denote the path direction change angle from Coord_i to Coord_i+1 as Delta_theta, and according to the turning performance of the device and safety requirements, the lower limit θ_min and the upper limit θ_max of the turning angle are determined, i.e. θ_min <= θ <= θ_max.

[0151] The recalculated device motion parameter constraints such as speed, acceleration and turning angle are arranged into a new set, denoted as New_Set_constraints. This new set of device motion parameter constraints is generated based on the alternative path coordinate sequence and is more suitable for the adjusted path situation.

[0152] Step S157: input the alternative path coordinate sequence and the re-calculated device motion parameter constraint conditions into the physical engine simulator to perform a second verification operation, and if the trajectory conflict point coordinate set is empty and the device over-limit operation parameter identifier is not activated, mark it as a verification pass strategy.

[0153] The alternative path coordinate sequence Sequence_alternative_path and the re-calculated device motion parameter constraint condition set New_Set_constraints are input into the physical engine simulator Simulator again.

[0154] In the physical engine simulator, the operation is performed according to a similar process as the first verification. Based on the alternative path coordinate sequence, a new predicted movement path of the device in the three-dimensional scene is marked. According to the re-calculated device motion parameter constraint conditions, the initial motion state and motion parameters of the device are set. For example, at the starting point of the path, a suitable initial speed V_start is selected according to the lower limit of speed V_min and the upper limit of speed V_max, where V_min<=V_start<=V_max, and the initial values of acceleration and steering angle are set to satisfy the corresponding constraint conditions.

[0155] Then, the physical engine simulator simulates the movement process of the device along the alternative path according to the physical laws and the set parameters. During the movement process, the spatial collision detection operation is performed again to check whether the movement trajectory of the device collides with the obstacle nodes in the work environment graph structure. At the same time, the running parameters of the device are monitored to determine whether they exceed the physical limit threshold.

[0156] If the trajectory conflict point coordinate set is empty during the entire simulation process, that is, no point of collision between the movement trajectory of the device and the obstacle is detected, and the device over-limit operation parameter identifier is not activated, that is, the running parameters such as the speed, acceleration, and steering angle of the device do not exceed their physical limit thresholds, then the dynamic emergency scheduling strategy is marked as a verification pass strategy. This indicates that the adjusted dynamic emergency scheduling strategy can effectively guide the device to safely complete the emergency operation path planning considering the work environment and the physical limitations of the device, and has practical application feasibility.

[0157] Step S158: generate a device control instruction code sequence based on the device motion parameter constraint conditions in the verification pass strategy, and synchronize and align the device control instruction code sequence with the alternative path coordinate sequence by time stamp to generate an emergency scheduling instruction set.

[0158] For the set of device motion parameter constraints New Set constraints in the verification pass strategy, translate them into a sequence of device control instruction codes. For the velocity constraint V_min <= V <= V_max, according to the control protocol and instruction coding rules of the device, generate the corresponding velocity control instruction code. For example, if the velocity control instruction code of the device adopts a certain pre-defined binary coding method, determine the code value representing the speed range under this coding method according to V_min and V_max. Similarly, for the acceleration constraint A_min <= A <= A_max and the steering angle constraint θ_min <= θ <= θ_max, generate the corresponding acceleration control instruction code and steering angle control instruction code respectively. Arrange these codes in a set order to form the device control instruction code sequence Set command codes.

[0159] At the same time, for the alternative path coordinate sequence Sequence alternative path, add timestamp information to each coordinate point. According to the expected motion speed of the device on the path and the length of the path segment and other factors, calculate the expected time for the device to reach each coordinate point. Let the average speed of the device on the path be V avg, for the i-th coordinate point Coord i on the path, calculate the distance d i from the starting point of the path to Coord i, and the expected passing time t i = d i / V avg according to V avg, to determine the timestamp T i of Coord i.

[0160] Synchronize and align the device control instruction code sequence Set command codes and the alternative path coordinate sequence Sequence alternative path according to the timestamps. For each instruction code in the device control instruction code sequence, find the path coordinate point within the corresponding timestamp range. For example, a certain velocity control instruction code corresponds to the speed control of the device within a certain time range, find the set of coordinate points on the motion path of the device within this time range. Combine these corresponding instruction codes and path coordinate points to generate the emergency scheduling instruction set Set emergency commands. This emergency scheduling instruction set not only contains the control instructions that the device should execute at different positions, but also ensures the consistency of the instructions and the path coordinates in time.

[0161] Step S159: Perform a packaging operation on the emergency scheduling instruction set, add job terminal device address identification and instruction priority labels, generate target transmission data packets and send them to the job terminal device.

[0162] The encapsulation operation is performed on the emergency dispatch instruction set Set_emergency_commands. First, a job terminal device address identifier is added to each instruction in the set. The job terminal device address identifier is information used to uniquely identify the job terminal device receiving the instruction, such as the MAC address of the device or a specific device number, etc. Let the job terminal device address identifier be Device_address. For each instruction Command_i in the emergency dispatch instruction set, Device_address is associated with Command_i, such as adding a field in the data structure of the instruction to store Device_address.

[0163] Then, an instruction priority label is added to each instruction. The instruction priority label is determined according to the urgency and importance of the instruction. For example, the priority of an instruction involving emergency braking of the device or avoiding a major hazard is high, while the priority of some routine adjustment instructions is relatively low. Let the instruction priority be divided into multiple levels, such as high, medium, and low. Through a priority determination algorithm (Priority_determine), the priority of each instruction is determined according to the type of instruction, the degree of influence on device safety and emergency operation, and other factors, and the corresponding priority label Priority_tag_i is added to the instruction.

[0164] The emergency dispatch instruction set with the added job terminal device address identifier and instruction priority label is packaged to generate the target transmission data packet Packet_target. The format of the data packet is defined according to the communication protocol, which may include a packet header, an instruction data part, a check code, etc. The packet header part may include the length of the data packet, the version number, etc. The instruction data part stores the emergency dispatch instruction after adding the identifier and label, and the check code is used to verify the integrity of the data packet during transmission.

[0165] Finally, the target transmission data packet Packet_target is sent to the job terminal device through a communication method such as a wireless network or a wired network. During transmission, the corresponding communication protocol is followed to ensure that the data packet is accurately transmitted to the specified job terminal device, so that the job terminal device can correctly receive and execute the emergency dispatch instruction.

[0166] Step S1510: updating the feasible path coordinate sequence in the dynamic emergency dispatch strategy according to the instruction confirmation signal and real-time position feedback data returned by the job terminal device, and triggering a periodic strategy validity verification operation.

[0167] The work terminal device returns an instruction confirmation signal to the system after receiving the target transmission data packet Packet_target and successfully parsing and executing the emergency dispatch instruction. At the same time, the work terminal device acquires real-time position information of itself through its own positioning system or other position detection means, and sends the real-time position feedback data back to the system.

[0168] After receiving the instruction confirmation signal and real-time position feedback data returned by the work terminal device, the feasible path coordinate sequence in the dynamic emergency dispatch strategy is first updated according to the real-time position feedback data. Assuming that the position fed back by the work terminal device in real time is Current_position, the Current_position is compared with the feasible path coordinate sequence in the dynamic emergency dispatch strategy. If the deviation between the Current_position and a coordinate point in the path sequence exceeds a set threshold (set as Deviation_threshold), it indicates that the actual position of the work terminal device deviates from the expected path.

[0169] Through a path adjustment algorithm (set as Path_adjust), the feasible path coordinate sequence is adjusted according to the Current_position and the actual situation of the work environment (such as the position of obstacles, the state of other devices, etc.). For example, if the work terminal device deviates from the expected path because it encounters a newly appeared obstacle, the path adjustment algorithm will re-plan the path to bypass the obstacle, and add a new coordinate point in the feasible path coordinate sequence to guide the work terminal device back to the reasonable emergency operation path.

[0170] After updating the feasible path coordinate sequence, a periodic strategy validity verification operation is triggered. According to the previous strategy validity verification process, that is, starting from constructing a three-dimensional dynamic scene reconstruction model, the operations of virtual motion simulation, collision detection, parameter comparison, etc. are sequentially performed to verify the validity of the dynamic emergency dispatch strategy again. In this way, it can be ensured that during the actual execution of the emergency dispatch strategy by the work terminal device, even if there is a position deviation, the strategy can be adjusted in time and its validity can be verified, so as to ensure that the emergency dispatch strategy always adapts to the actual situation of the work site, and to ensure the safe performance of the aerial work.

[0171] And, in the above scenario, high-altitude work personal safety emergency dispatching and work guidance functions can be provided, such as a configurable intelligent safety helmet, a vital sign monitoring bracelet, and a management end system. The vital sign monitoring bracelet is used to collect heart rate, blood pressure, and blood oxygen data of the wearer of the intelligent safety helmet. The intelligent safety helmet includes a video, voice intercom, SOS alarm, fall early warning, and proximity electric detection module. Before high-altitude work, the data collected by the vital sign monitoring bracelet and the relevant detection modules of the intelligent safety helmet are used to determine whether the worker is suitable for high-altitude work. For example, if the heart rate collected by the vital sign monitoring bracelet exceeds the normal range, or the proximity electric detection module of the intelligent safety helmet detects a strong electric field nearby, it is determined that the worker is not suitable for the current high-altitude work, and the management end system background is directly warned, and the voice and video functions of the intelligent safety helmet are used for emergency handling, such as notifying the worker to stop working and guiding him to return to the ground safely. During high-altitude work, on the one hand, the state of the worker is continuously monitored, and on the other hand, the video collection function of the intelligent safety helmet provides convenience for remote guidance of work, and the collected video data is part of a real-time video data set, which is used for subsequent processing together with data collected by other cameras.

[0172] Figure 2 A schematic diagram of exemplary hardware and software components of a high-altitude work emergency dispatching system 100 that can implement the idea of the present application is shown. For example, a processor 120 can be used on the high-altitude work emergency dispatching system 100 and used to perform the functions in the present application.

[0173] The high-altitude work emergency dispatching system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the high-altitude work emergency dispatching method of the present application. Although only one server is shown in the present application, for the sake of convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0174] For example, the high-altitude work emergency dispatching system 100 can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, a ROM, or a RAM, or any combination thereof. Exemplarily, the high-altitude work emergency dispatching system 100 can also include program instructions stored in a ROM, a RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The high-altitude work emergency dispatching system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0175] For the convenience of description, only one processor is described in the aerial work emergency dispatching system 100. However, it should be noted that the aerial work emergency dispatching system 100 in the present application can also include multiple processors, and thus the steps performed by one processor described in the present application can also be jointly performed or separately performed by multiple processors. For example, if the processor of the aerial work emergency dispatching system 100 performs steps A and B, it should be understood that steps A and B can also be jointly performed by two different processors or separately performed in one processor. For example, a first processor performs step A, a second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0176] In addition, the embodiment of the present application further provides a readable storage medium, wherein computer executable instructions are preset in the readable storage medium, and when a processor executes the computer executable instructions, the aerial work emergency dispatching method is realized.

[0177] It should be noted that, in order to simplify the description of the present application and to help the understanding of one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A method for emergency dispatching of high-altitude operations, characterized in that, The method includes: Acquire a set of real-time video data from the high-altitude work area, and perform preprocessing operations on the real-time video data set to generate a standardized video data stream; The video analysis model is invoked to perform feature extraction on the standardized video data stream, generating a set of work environment features and a set of worker behavior features; Based on the set of operational environment features, an operational environment graph structure is constructed. A graph convolutional network is then invoked to perform multi-level feature aggregation operations on the operational environment graph structure to generate an environmental state feature vector. A dynamic emergency dispatch strategy is generated based on the environmental state feature vector and the set of operator behavior features. The effectiveness verification operation of the dynamic emergency dispatch strategy is performed to obtain the strategy verification result, and an emergency dispatch instruction set is generated based on the strategy verification result. The emergency dispatch instruction set is then transmitted to the operation terminal equipment to trigger the emergency response operation.

2. The emergency dispatch method for high-altitude operations according to claim 1, characterized in that, The step of performing preprocessing operations on the real-time video data set to generate a standardized video data stream includes: A resolution equalization process is performed on the real-time video data set to generate initial video data frames with a uniform resolution format; The moving target detection algorithm is invoked to perform noise filtering on the initial video data frame to remove interfering pixel regions that are not working targets, thus obtaining the filtered video data frame; Perform time-series alignment on the filtered video data frames to generate a standardized video data stream with consecutive timestamps; A color gamut conversion operation is performed on the standardized video data stream to map the original color gamut space to a preset standardized color gamut space. An illumination compensation operation is then performed on the converted video data stream to eliminate shadow areas and overexposed areas in the initial video data frames, resulting in a compensated video data stream. Perform frame rate stability testing on the compensated video data stream to eliminate video jitter and generate a time-coherent standardized video data stream.

3. The emergency dispatch method for high-altitude operations according to claim 1, characterized in that, The step of calling the video analysis model to perform feature extraction on the standardized video data stream, generating a set of work environment features and a set of worker behavior features, includes: The standardized video data stream is divided into multiple video segments of equal length, and a three-dimensional convolutional feature extraction operation is performed on each video segment to generate a spatiotemporal correlation feature matrix; Attention weight allocation is performed on the spatiotemporal correlation feature matrix to generate a visual focus distribution map with regional importance annotations. Based on the visual focus distribution map, edge tracking is performed on the outline of the working equipment to extract the equipment operating status features and spatial displacement trajectory features. Perform skeletal key point detection on the limb movements of workers to generate sets of movement amplitude features and movement speed features; The set of behavioral characteristics of the operator is generated by aggregating the equipment operating status characteristics, spatial displacement trajectory characteristics, motion amplitude characteristics, and motion speed characteristics. Perform texture analysis on the fixed obstacles in the working environment to extract surface roughness features and geometric shape features; By fusing the surface roughness features, geometric features, and equipment operating status features through a graph neural network, a dynamic interaction model of material properties and equipment kinematic parameters is constructed, generating a set of environmental interaction risk features.

4. The emergency dispatch method for high-altitude operations according to claim 1, characterized in that, The construction of the work environment graph structure based on the set of work environment features includes: Set the work equipment nodes, environmental obstacle nodes, and personnel location nodes as the basic node set of the graph structure; The first type of edge connection relationship is constructed based on the actual distance between the device node and the obstacle node, and the second type of edge connection relationship is constructed based on the operational correlation between the personnel location node and the device node. A node attribute feature set is constructed based on environmental monitoring sensor data. The node attribute feature set includes equipment operating parameters, obstacle material properties, and personnel status parameters. The personnel status parameters include personnel protective equipment status parameters and personnel vital sign status parameters. The basic node set, edge connection relationship and node attribute feature set are combined to generate the working environment graph structure; Perform a topology integrity check on the aforementioned work environment graph structure, supplement missing edge connections and delete redundant nodes; The corresponding parameter values ​​in the node attribute feature set are updated according to the dynamic changes of the equipment operating parameters, and the actual distance between the equipment node and the obstacle node is recalculated based on the real-time position of the equipment node. The first type of edge connection relationship is dynamically adjusted to generate a dynamically updated working environment graph structure.

5. The emergency dispatch method for high-altitude operations according to claim 4, characterized in that, The step of combining the basic node set, edge connection relationships, and node attribute feature set to generate the job environment graph structure includes: Perform vectorization encoding on the device operating parameters in the node attribute feature set to generate device attribute vectors; Perform texture feature extraction on the obstacle material attributes in the node attribute feature set to generate an obstacle attribute vector; Perform vectorization encoding on the personnel status parameters in the node attribute feature set to generate personnel attribute vectors; The device attribute vector, obstacle attribute vector, and personnel attribute vector are mapped to a unified feature dimension space through a fully connected layer, and then concatenated according to the channel dimension to generate a node feature matrix. Calculate the spatial correlation between the device node and the obstacle node based on the actual distance value of the first type of edge connection relationship, and generate the first type of edge weight matrix; Calculate the interaction strength value between the personnel location node and the equipment node based on the operational correlation parameter of the second type of edge connection relationship, and generate the second type of edge weight matrix; The node feature matrix is ​​mapped to the first type of edge weight matrix and the second type of edge weight matrix to generate initial graph structure data. Perform reachability verification between nodes on the initial graph structure data, delete edge connections that do not meet the preset connectivity threshold, supplement missing key edge connections, and generate a fully connected working environment graph structure; Based on the real-time update frequency of the equipment operating parameters, the fully connected operation environment graph structure is dynamically incrementally updated to generate a dynamic operation environment graph structure with a timestamp version identifier.

6. The emergency dispatch method for high-altitude operations according to claim 1, characterized in that, The invoked graph convolutional network performs multi-level feature aggregation operations on the job environment graph structure to generate an environment state feature vector, including: In the initial convolutional layer of the graph convolutional network, feature aggregation operation is performed on the direct neighboring nodes of each node in the working environment graph structure to generate a primary aggregated feature vector; In the intermediate convolutional layer of the graph convolutional network, a cross-level feature propagation operation is performed on the indirectly adjacent nodes with a two-edge connection relationship to generate an intermediate aggregated feature vector. The primary aggregated feature vector and the intermediate aggregated feature vector are fused in the final convolutional layer of the graph convolutional network to generate an environment state feature vector containing global dependencies. The spatial coordinate encoding information embedded in the environmental state feature vector is subjected to dimensionality reduction and visualization processing to generate an environmental risk heat map and mark the coordinates of high-risk areas; Spatial clustering analysis is performed on the coordinate data of the environmental risk heatmap to identify risk clustering areas and generate area boundary labeling information; The node attribute feature set of the operational environment graph structure is updated based on the region boundary annotation information to generate an optimized environmental state feature vector.

7. The emergency dispatch method for high-altitude operations according to claim 1, characterized in that, The step of generating a dynamic emergency dispatch strategy based on the environmental state feature vector and the set of operator behavior features includes: After aligning the environmental state feature vector and the set of worker behavior features to the same feature dimension by inputting them into a linear projection layer, they are then input into a pre-trained spatiotemporal correlation model. Through a cross-attention mechanism, feature interaction verification is performed on the node attribute features in the environmental state feature vector and the dynamic parameters in the set of worker behavior features, generating an interaction matrix. Based on the association weight value between each node and behavioral feature in the interaction matrix, the risk correlation parameter between high-risk nodes and abnormal personnel behavior features in the work environment graph structure is extracted. Based on the risk correlation parameter, an adaptive scaling operation is performed on the node spatial coordinate features in the environmental state feature vector to generate a risk-weighted spatial distribution feature vector. The risk-weighted spatial distribution feature vector and the edge connection relationship of the work environment graph structure are input into the graph attention network, and a topology-optimized feature graph containing emergency obstacle avoidance paths between nodes is generated through the edge weight dynamic adjustment algorithm. Based on the path connectivity parameters in the topology optimization feature graph, the Dijkstra algorithm is called in combination with the edge weights of the graph attention network to calculate the shortest safe movement path sequence of the work terminal equipment in the work environment graph structure, and the equipment motion parameter constraints corresponding to the shortest safe movement path sequence are generated. A dynamic compatibility check is performed on the constraints of the device motion parameters and the real-time device operating parameters. Based on the check results, path segments that exceed the physical limits of the device are eliminated, and a dynamic emergency scheduling strategy containing feasible path coordinates and device control command set is generated.

8. The emergency dispatch method for high-altitude operations according to claim 1, characterized in that, The step of performing an effectiveness verification operation on the dynamic emergency dispatch strategy to obtain a strategy verification result includes: Extract the feasible path coordinate sequence and equipment motion parameter constraints from the dynamic emergency dispatch strategy, and construct a three-dimensional dynamic scene reconstruction model based on the spatial coordinate system of the work environment map structure; The feasible path coordinate sequence and device motion parameter constraints are input into the preloaded physics engine simulator. The simulation task is segmented based on the path curvature key points. The virtual motion simulation operation is executed in a segmented parallel computing mode to generate simulation motion trajectory data and a set of virtual device state parameters. A spatial collision detection operation is performed on the simulated motion trajectory data and the coordinates of obstacle nodes in the operational environment diagram structure to generate a set of trajectory conflict point coordinates and a conflict timestamp sequence. The set of virtual device status parameters is dynamically compared with the physical limit threshold of the device operating parameters to generate an identifier of the device's out-of-limit operating parameters and the corresponding time interval. The strategy verification anomaly report is generated by integrating the set of coordinates of the trajectory conflict points, the conflict timestamp sequence, and the equipment over-limit operation parameter identifiers, and the coordinates of the abnormal path segments and the abnormal equipment control commands are extracted. Based on the coordinates of the abnormal path segments, perform piecewise smooth interpolation on the feasible path coordinate sequence to generate an alternative path coordinate sequence and recalculate the equipment motion parameter constraints. The alternative path coordinate sequence and the recalculated device motion parameter constraints are input into the physics engine simulator to perform a secondary verification operation. If the set of trajectory conflict point coordinates is empty and the device over-limit operation parameter identifier is not activated, it is marked as a verification pass strategy.

9. The emergency dispatch method for high-altitude operations according to claim 8, characterized in that, The step of generating an emergency dispatch instruction set based on the strategy verification result and transmitting the emergency dispatch instruction set to the work terminal equipment to trigger emergency response operations includes: Based on the equipment motion parameter constraints in the verification strategy, a sequence of equipment control instructions is generated, and the sequence of equipment control instructions is synchronized with the alternative path coordinate sequence by timestamp synchronization to generate an emergency dispatch instruction set. The emergency dispatch instruction set is encapsulated by adding the work terminal device address identifier and instruction priority tag, generating a target transmission data packet and sending it to the work terminal device. The feasible path coordinate sequence in the dynamic emergency dispatch strategy is updated based on the instruction confirmation signal returned by the operation terminal equipment and the real-time location feedback data, triggering a periodic strategy effectiveness verification operation.

10. A high-altitude operation emergency dispatch system, characterized in that, The device includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the high-altitude operation emergency dispatch method according to any one of claims 1-9.

Images