Post-earthquake rescue path optimization LGNN algorithm coupling safety and efficiency

Abstract

The invention provides a safety and efficiency coupled post-earthquake rescue path optimization LGNN algorithm, and relates to the technical field of post-earthquake rescue path optimization, and the algorithm comprises the steps: obtaining multi-source data; constructing a road map through the multi-source data and the susceptibility result; according to an emergency rescue report in the multi-source data, determining a rescue center and a disaster site; constructing a lightweight graph neural network; inputting the road map, the rescue center and the disaster point into a lightweight map neural network; and taking the cumulative weight of each node from the rescue center to the road map as a node feature, iterating the node feature representation through the lightweight map neural network to obtain the minimum value of the cumulative weight of the disaster-affected point, and backtracking the cumulative weight of the disaster-affected point to obtain the optimal rescue path from the rescue center to the disaster-affected point. And efficient emergency rescue path optimization is realized on a wide-area complex road network in a wide-area strong earthquake area.

Description

Technical Field

[0001] The present application relates to the technical field of post-earthquake rescue path optimization, and in particular to a post-earthquake rescue path optimization LGNN algorithm that couples safety and efficiency. Background Art

[0002] At present, the main difficulty in optimizing rescue routes after a major earthquake is that the efficiency of path optimization in a wide-area complex road network is low, and the safety along the road (susceptibility to co-seismic disasters) is not considered, which cannot meet the needs of emergency rescue. After an earthquake, finding a fast route for emergency rescue vehicles between the rescue center and the disaster site is the shortest path problem in path optimization. The traditional global search path optimization algorithm achieves path optimization by traversing all edges and nodes to find the shortest path from the starting point to the target node. Among them, Dijkstra traverses all nodes to find the shortest path from all nodes to the starting point, and then queries to obtain the shortest path from the starting point to the end point. The A* algorithm uses the physical distance from the neighbor node to the current node as one of the bases for selecting the neighbor node. Due to the addition of this heuristic indicator, the A* algorithm can theoretically find the optimal path faster than Dijkstra. Because the traditional global search algorithm is a global search, it can generally find an optimal path from the starting point to the end point (if the path exists), which is relatively stable. However, due to its global traversal characteristics, the search efficiency is relatively low, which limits the application of traditional global search path optimization in wide-area strong earthquake areas.

[0003] Inspired by the group behavior of animals, the foraging behavior of microorganisms or natural phenomena, heuristic optimization algorithms have been applied to path optimization in recent years. They use local search and converge to the global optimum through gradual iteration. After defining the path according to the node ID or node attribute (path encoding), the node attributes are iterated multiple times and the top-k solutions are selected to gradually converge to the global optimum, and finally the shortest path is obtained according to the path definition. Genetic Algorithm (GA) imitates the law of gene inheritance (recombination and mutation produce new genes, and the fitness function eliminates poor genes) to optimize the solution. Unlike other problems such as the Traveling Salesman Problem (TSP), the number of nodes in a single solution of the metaheuristic algorithm used for path optimization is not fixed (the path length is uncertain), and the crossover and mutation of genes need to be redefined. Unlike the GA path optimization method, Particle Swarm Optimization (PSO) does not redefine the behavior of particles but converts the path solution (variable length solution) into node attributes (fixed length solution), that is, PSO is used to optimize the priority of nodes in the graph and select node IDs to form paths according to the priority. Pulse Coupled Neural Network (PCNN), Rapid Physarum Algorithm (RPA) and Ripple Spread Algorithm (RSA) methods simulate the information transmission and connection (pulse signal) of the primate visual system, the foraging process of Physarum, and the ripple diffusion (ripple relay race) of the liquid surface to optimize the path. Similar to PCNN, the Rosenance algorithm is also a path optimization method based on signal transmission. When the signal is transmitted to a node for the first time, the node is activated. The overlapping path is the shortest path through the superposition of the forward transmission of the starting signal and the reverse transmission of the end signal.

[0004] At present, path optimization has made important progress and produced a large number of excellent results. However, there are still the following limitations: (1) The current path optimization algorithm mainly performs well on small-scale graphs (the number of nodes is less than or equal to 2000). However, earthquake rescue is often a large-scale complex road network. In real earthquake scenarios, the number of nodes may be greater than 10,000 or even 100,000. The current shortest path optimization algorithm has low computational efficiency on large-scale graphs due to the global traversal characteristics and unreasonable solution regularization methods, which limits its application in large-scale earthquake rescue road network graphs. (2) The current emergency rescue route optimization does not consider the path safety issues caused by the distribution of coseismic landslides, which is very detrimental to the smooth progress of emergency rescue and the safety of rescue workers. (3) At present, the evaluation of coseismic landslide susceptibility usually adopts all coseismic landslide training models and evaluates landslide susceptibility. However, there are many coseismic landslides and their distribution area is wide. The complete and accurate interpretation of these landslides usually takes weeks to months, which is difficult to support the 72-hour golden rescue and has limited practicality. Summary of the invention

[0005] The purpose of the present invention is to provide a post-earthquake rescue path optimization LGNN algorithm that couples safety and efficiency in order to solve the problem that the existing emergency rescue route optimization does not take into account the path safety caused by the co-seismic landslide distribution.

[0006] The above-mentioned purpose of the present application is achieved through the following technical solutions:

[0007] S1: Obtain multi-source data; predict the susceptibility of co-seismic landslides through multi-source data combined with CGBoost algorithm, obtain susceptibility results and verify them; the multi-source data include: terrain data, geological data, human engineering activity data, environmental data, rainfall data and earthquake data;

[0008] S2: Construct a road map through multi-source data and susceptibility results;

[0009] S3: Determine the rescue center and disaster-affected site based on the emergency rescue reports in multi-source data;

[0010] S4: Building lightweight graph neural networks;

[0011] S5: Input the road map, rescue centers, and disaster-affected sites into a lightweight graph neural network;

[0012] The accumulated weight of each node from the rescue center to the road graph is used as the node feature to characterize the node feature;

[0013] Through the lightweight graph neural network iterative node feature representation, the minimum cumulative weight of the disaster-stricken point is obtained. By backtracking the cumulative weight of the disaster-stricken point, the optimal rescue path from the rescue center to the disaster-stricken point is obtained.

[0014] An electronic device includes a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes a post-earthquake rescue path optimization LGNN algorithm that couples safety and efficiency.

[0015] A computer-readable storage medium stores instructions, which, when executed, execute an LGNN algorithm for optimizing a post-earthquake rescue path that couples safety and efficiency.

[0016] The beneficial effects of the technical solution provided by this application are:

[0017] 1. LGNN (Light Graph Neural Network) node parallel representation has high efficiency advantages when applied to large-scale complex road networks. Earthquake rescue is often a large-scale complex road network. In real earthquake scenarios, the number of nodes may be greater than 10,000, or even reach 100,000. The current shortest path optimization algorithm is not reasonable due to the global traversal characteristics and the solution regularization method in the heuristic optimization algorithm (it is difficult to find the optimal path). These methods have low computational efficiency on large-scale graphs. The LGNN algorithm represents the optimal rescue path as the problem of minimizing the cumulative road weight starting from the rescue center. For each node (intersection) in the road network, LGNN uses the cumulative weight of all roads from the rescue center (starting point) to the node (referred to as the node cumulative weight) as the node feature, and uses the node topology relationship and GNN message passing and feature aggregation mechanism to parallelly characterize the node features of each node, and backtrack to obtain the optimal path. Due to the high efficiency advantage of LGNN parallel representation, efficient emergency rescue path optimization can be achieved on a large-scale complex road network in a wide-area strong earthquake zone.

[0018] 2. The susceptibility of co-seismic landslides along the road is included as one of the road indicators to be considered in path optimization, which improves the rationality of path optimization. Emergency rescue route optimization is to find a safe route for rescue vehicles to quickly reach the rescue point. However, these high-risk areas of landslides along the road are dangerous areas for emergency rescue. The cracked rock and soil and loose deposits may cause emergency rescue personnel or vehicles to be buried or injured, seriously threatening the safety of rescue workers and rescued personnel. Therefore, the present invention takes into account the safety along the road, uses the susceptibility of co-seismic landslides along the road to measure the safety of the rescue road; quantifies road safety and travel efficiency (vehicle travel time) according to the landslide susceptibility level, road grade and road length, and couples them to carry out emergency rescue route optimization and improve the rationality of path optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present application will be further described below with reference to the accompanying drawings and embodiments, in which:

[0020] Figure 1 is a step diagram in an embodiment of the present application;

[0021] Figure 2 It is the susceptibility prediction map and statistical information map of the study area in the embodiment of the present application;

[0022] Figure 3 It is a path optimization result diagram of LGNN in five cases in the embodiment of the present application;

[0023] Figure 4 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to have a clearer understanding of the technical features, purposes and effects of the present application, the specific implementation methods of the present application are now described in detail with reference to the accompanying drawings.

[0025] The embodiments of the present application provide a post-earthquake rescue path optimization LGNN algorithm that couples safety and efficiency.

[0026] Please refer to Figure 1 , Figure 1 It is a step diagram of a post-earthquake rescue path optimization LGNN algorithm for coupling safety and efficiency in an embodiment of the present application, including:

[0027] S1: Obtain multi-source data; predict the susceptibility of co-seismic landslides through multi-source data combined with CGBoost algorithm, obtain susceptibility results and verify them; the multi-source data include: terrain data, geological data, human engineering activity data, environmental data, rainfall data and earthquake data;

[0028] Specifically, CGBoost landslide susceptibility prediction under a small number of landslide samples. At present, the co-seismic landslide susceptibility evaluation usually adopts all co-seismic landslide training models and evaluates landslide susceptibility. However, there are a large number of co-seismic landslides and their distribution area is wide. The complete and accurate interpretation of these landslides usually takes weeks to months, which is difficult to support the 72-hour golden rescue and has limited practicality. The present invention uses 10% of the co-seismic landslide cataloging data to simulate a small amount of landslide data obtained in a short time after the earthquake for susceptibility modeling. Since the CGBoost algorithm has good susceptibility modeling capabilities under a small number of samples, the CGBoost co-seismic landslide susceptibility prediction using a small number of landslides can well support emergency rescue.

[0029] Specifically, a cross-graph attention network (Crossgat) and extreme gradient boosting trees (XGBoost) integration method, called the CGBoost algorithm, is applied, which comes from Chinese patent CN116894523A - a CGBoost algorithm, device and storage device for predicting the susceptibility of co-seismic landslides.

[0030] S2: Construct a road map through multi-source data and susceptibility results;

[0031] S3: Determine the rescue center and disaster-affected site based on the emergency rescue reports in multi-source data;

[0032] Specifically, according to the emergency rescue report, the help center and the disaster site are determined. According to the disaster situation and rescue force distribution reported in the emergency rescue response bulletin, the disaster site with serious casualties is regarded as the waiting rescue point (end point t), and the main rescue force deployed in the rescue report is regarded as the rescue center (starting point s).

[0033] S4: Building lightweight graph neural networks;

[0034] Specifically, mathematical modeling and analysis are performed on the problem of emergency rescue path optimization. The problem of route optimization for rescue vehicles in a large-scale complex road network is abstracted into a mathematical problem. The edge weights of the road graph are involved in the aggregation of neighbor node features, and the minimum aggregation is used to aggregate neighbor node features to construct a lightweight graph neural network (LGNN). The minimum aggregation of edge weights enables node features to converge, and the nodes of the minimum neighbor aggregation are represented in parallel, so that node features can quickly converge to the minimum, thereby improving the speed of path optimization.

[0035] S5: Input the road map, rescue centers, and disaster-affected sites into a lightweight graph neural network;

[0036] The accumulated weight of each node from the rescue center to the road graph is used as the node feature to characterize the node feature;

[0037] Through the lightweight graph neural network iterative node feature representation, the minimum cumulative weight of the disaster-stricken point is obtained. By backtracking the cumulative weight of the disaster-stricken point, the optimal rescue path from the rescue center to the disaster-stricken point is obtained.

[0038] Step S1 includes:

[0039] S11: Terrain data includes seamless digital elevation model data; geological data includes geological maps, pre-earthquake fault data and world digital soil maps; human engineering activity data includes land use data and pre-earthquake road network data; environmental data includes water network data; earthquake data includes vibration maps, post-earthquake 3D Google images, co-seismic landslide catalog data and real resettlement site data for earthquake emergency response;

[0040] Specifically, terrain factors are established through seamless digital elevation model data; engineering rock groups are constructed through geological maps; distances to faults are constructed through pre-seismic fault data; soil type factors are constructed through the world digital soil map; types and characteristics of human engineering activities are obtained through land use data; the distances of environmental factors to water systems are established through water network data; annual average rainfall, monthly average rainfall, monthly average rainfall change rate, and monthly average rainfall change acceleration factors are constructed through rainfall data; peak ground acceleration, peak seismic velocity, Mercalli earthquake intensity, and distance from the epicenter factor are extracted through vibration maps; and road accessibility is determined through pre-seismic road networks and post-seismic 3D Google images.

[0041] Specifically, the Digital soil map of the world. Rainfall data includes monthly rainfall from 1960 to 2018. ShakeMap is used to extract Peak Ground Acceleration (PGA), Peak Ground Velocity (PGV), Modified Mercalli Intensity (MMI) and distance factor from the epicenter.

[0042] S12: Construct disaster-pregnancy and disaster-causing factors through terrain data, geological data, human engineering activity data, environmental data, rainfall data and vibration map; disaster-pregnancy and disaster-causing factors are 20 landslide susceptibility prediction indicators;

[0043] S13: Based on 10% of the co-seismic landslide catalog data and disaster-pregnancy and disaster-causing factors, the CGBoost algorithm is used to predict the susceptibility of co-seismic landslides and obtain the susceptibility results;

[0044] S14: Validate the predicted susceptibility results using all coseismic landslide inventory data.

[0045] Specifically, the rationality of the path optimization results is verified through the real resettlement point data of earthquake emergency response.

[0046] Step S2 includes:

[0047] S21: Extract road grade and road length through pre-earthquake road network data in human engineering activity data;

[0048] S22: Calculate the road travel time according to the road grade and road length;

[0049] Specifically, for the road travel time, refer to the traffic speed limit regulations: the maximum speed on expressways is 100km / h, the maximum speed on intercity roads and main roads is 80km / h, the maximum speed on secondary roads and urban roads is 50km / h, and the maximum speed in residential areas is 30km / h. The maximum travel speed V of the road g is quantified according to the road grade and road type. g As shown in Table 1. Then Calculate the road travel time T g , where l g Indicates the length of the road g.

[0050] S23: quantifying the susceptibility results to determine the landslide susceptibility level; the landslide susceptibility level includes: very low susceptibility, low susceptibility, medium susceptibility, high susceptibility and very high susceptibility;

[0051] S24: Select the maximum value of the landslide susceptibility level in the road buffer as the landslide susceptibility level P of road g g , g∈{1,2,…,NS}, NS represents the number of roads in the road network;

[0052] Specifically, the susceptibility of landslides within 50 m on both sides of the road can well reflect the susceptibility of landslides on the road. Therefore, the maximum value of the landslide susceptibility level within the 50 m buffer zone of the road is selected as the landslide susceptibility level P of road g. g .

[0053] S25: Determination of road accessibility: Based on the pre-earthquake road network data and the post-earthquake 3D Google images, the inaccessible roads are interpreted to obtain the interpretation results; the inaccessible roads include: roads blocked by landslides and roads blocked by the accumulation of collapsed buildings;

[0054] S26: According to the interpretation results, the impassable roads are eliminated; the intersections of the impassable roads are used as nodes, the roads are used as edges, the road travel time and the landslide susceptibility level of the roads are used as edge weights, and a road map is constructed.

[0055] Step S4 includes:

[0056] The problem of optimizing rescue vehicle routes in a large-scale complex road network is abstracted into a mathematical problem and mathematical modeling is performed;

[0057] Define the node where the rescue center is located as the starting point s of the vehicle, and the disaster point with serious casualties as the end point t of the vehicle. Both s and t are nodes in the graph data. Assume that the number of vehicles at the starting point s is 1 and the vehicle has the ability to rescue only one disaster point.

[0058] The intersections of accessible roads are taken as nodes, roads are taken as edges, and the road travel time and the landslide susceptibility level of the road are taken as edge weights. The graph data is constructed as G(V,W), where N represents the number of road intersection nodes in the road network, and V = {0, 1, 2, ..., N-1} represents the node set. The adjacency matrix representing the edges of the graph data, is the weight matrix of the edge of the graph data;

[0059] For a road g with nodes i and j, the corresponding edges are (i, j) and (j, i); (i, j) represents the road g from node i to node j; (j, i) represents the road g from node j to node i, so A i,j =A j,i =1, if road g is an impassable road, then A i,j =A j,i =0;

[0060] Specifically, the purpose of the emergency rescue vehicle path optimization is to find a route that takes into account both safety and timeliness on the accessible road map. The path safety is the sum of the landslide susceptibility levels of all roads in the path. The landslide susceptibility level of road g is P g The lower it is, the safer it is for rescue vehicles to pass through the road. The path travel time is the accumulation of the road travel time in the path. The road travel time T g The fewer the number of g, the faster the search and rescue vehicles can pass through the road, which is more efficient and beneficial to the search and rescue of personnel. The landslide susceptibility level and the road travel time are equally important for rescue.

[0061] Landslide susceptibility level P of road g g and road travel time T g are equally important for emergency rescue operations. If there is a road g, the weight W of the edge (i, j) is i,j =P g +T g , W i,j =W j,i ; If there is no road g, then W i,j =W j,i =∞;

[0062] Define node s as the starting point, node t as the end point, a′,c′,d′,u′, as nodes different from s and t; s,a′,c′,d′,u′,t∈V;

[0063] Define the matrix P and matrix C to represent the path matrix and cumulative weight matrix respectively.

[0064] definition The element in the sth row and tth column of the matrix P represents the path from node s to node t;

[0065] definition represents the element in the sth row and tth column of C, which represents the cumulative weight from node s to node t;

[0066] If there are edges (s,a′),(a′,c′),(c′,d′),…,(u′,t), their corresponding paths are g1,g2,g3,…,g l ,g1,g2,g3,…,g l ∈{1,2,…,NS}, then the path from s to t is The corresponding cumulative weight from s to t is

[0067] Based on the above definition, the path from the starting point s to the end point t on the graph data G(V,W) that considers security and timeliness is preferably described as a minimization problem on G(V,W), as follows:

[0068]

[0069] Where min ' It represents the function of finding the minimum value of a variable;

[0070] definition The minimum value of represents the minimum cumulative weight from s to t; The corresponding path from s to t is The minimum value of is the shortest path;

[0071] Through the triangle inequality property of the shortest path, we get the following proposition and inference:

[0072] Proposition 1: For represents the set of neighbor nodes of t and

[0073] Known W u,t , W u,t represents the weight of the edge (u, t), where u represents a neighbor node of t; if given but min(·) is a function that finds the minimum value in a set. The corresponding u is u′, then

[0074] Corollary 1: Definition achieve , the corresponding neighbor node of t is u′, then among the neighbor nodes u of node t, only u=u′ satisfies That is, the predecessor node of t on the shortest path from s to t And at this time achieve For any other u=u e ,regardless Whether it is achieved have u e represents the set of neighbor nodes of node t that does not include u′;

[0075] Specifically, in order to speed up The calculation will The calculation is converted into a parallel representation of the GNN nodes of the graph neural network, and the speed of path optimization for large graphs is further improved with the help of the parallelism of the GNN node representation.

[0076] Will The solution is defined as of Minimization problem, i∈V, and by Solution

[0077] The cumulative weight from node s to node i is used as the feature of node i, and Modeled as the representation process of node i;

[0078] The node features of each node i are parallelly represented by using the node topology relationship and the message passing and feature aggregation mechanism of the graph neural network GNN.

[0079] initial i≠s, due to the minimum feature aggregation mechanism, after iterative characterization Gradually converge to For i = s, we have exist After convergence, we iterate back to the predecessor node through Corollary 1 and get

[0080] Step S4 also includes:

[0081] Using minimum value aggregation, the edge weights are involved in the aggregation of neighbor node features. The initial node representation of the lightweight graph neural network is as follows:

[0082]

[0083] in represents the kth iteration represents the cumulative edge weight from node s to node i; represents the k-1th iteration represents the cumulative edge weight from node s to node j;

[0084] For k = 0, i = s, we have

[0085] For i = other∈V-{s}, we have

[0086] For node t, h t express That is, the cumulative edge weight from node s to node t; when h t When k drops from k = 0 to k = k', and no longer changes from k = k' to k = k' + τ, that is, Reach the minimum value and find satisfaction through inference 1 The predecessor node u′ of t is obtained by iterative backtracking. τ represents the maximum number of iterations that the node features on the graph remain unchanged after they are not ∞;

[0087] Specifically, the LGNN (Light Graph Neural Network) bidirectional path optimization algorithm can further improve the speed of rescue path optimization.

[0088] In the positive weighted undirected graph G(V,W), s is the starting point and t is the end point. The bidirectional algorithm calculates the value of each node i from the starting point s and the end point t respectively. and If the node v and Both reach the minimum value. Since G(V,W) is an undirected graph, Corresponding Reversed(·) indicates a reversal function;

[0089] In the LGNN bidirectional path optimization algorithm, and Represented as node features, let Therefore, for each node i, LGNN represents the node features from s to i in parallel and the node features from t to s D i ;

[0090] The bidirectional node characterization process of the LGNN bidirectional path optimization algorithm is as follows:

[0091]

[0092] in, Represents the characteristics of node i at the kth iteration and When i = s When i=other When i = t When i=other

[0093] For any node i, Therefore If there exists a node v such that When the node v corresponds to First converge to hour, has converged, and at other nodes i e ∈V-{v} corresponds to During the convergence process, has ceased to change, so when It can be judged when it drops to no longer change Convergence, without having to wait until any node i corresponds to No longer changes.

[0094] Step S4 also includes:

[0095] In the LGNN bidirectional path optimization algorithm, C s,k ,C t,k =D k [:,0],D k [:,1],

[0096]

[0097] C s,k represents the cumulative weight vector C from node s to all nodes represented by the kth iteration S ;

[0098] C t,k Represents the cumulative weight vector C from node t to all nodes represented by the kth iteration t ;

[0099] make When it decreases until it stops changing from k = k' to k = k' + τ, it is determined that has converged;

[0100] Mark min(C s,k +C t,k) corresponds to node v, then node v is the intersection node of bidirectional optimization, that is, the node on the shortest path;

[0101] By formula get and Based on node v as the starting point and The optimal rescue path obtained by backtracking.

[0102] Table 1. Quantitative rules for maximum speed

[0103]

[0104] The present invention adopts 10% random co-seismic landslides after earthquake to simulate the small amount of co-seismic landslide data that can be obtained after earthquake, and adopts CGBoost algorithm to realize more accurate prediction of co-seismic landslide susceptibility in wide-area strong earthquake area, so as to better support emergency rescue; according to the 10% co-seismic landslides obtained randomly, CGBoost is used to predict the susceptibility of co-seismic landslides in M w The landslide susceptibility prediction under the 6.9Lombok earthquake was carried out, and the overall accuracy, precision, recall, F1 index and Kappa coefficient of susceptibility prediction were 0.9233, 0.9387, 0.9079, 0.9231 and 0.8467 respectively. Figure 2 The susceptibility map and statistical information of the study area are shown. NPLHVH represents the percentage of landslides falling in high susceptibility areas and very high susceptibility areas. Practicality refers to the practicality of susceptibility prediction represented by the percentage of the area of ​​high and very high susceptibility areas. In the figure, 92.25% of the landslides in the study area fall in high and very high susceptibility areas, and the area of ​​high and very high susceptibility areas accounts for 15.73%. Determining the vast low and very low susceptibility areas can strongly support effective and efficient emergency rescue operations.

[0105] Figure 3 The figure shows the path optimization results of LGNN in five cases. s1-s5, t1-t5 represent the rescue center and the disaster site. Figure 3 (a)-(e) represent the path optimization results of Situation 1-5 with and without considering the susceptibility, respectively. ORE represents the optimal path obtained by considering only the traffic efficiency, and ORSE represents the optimal path obtained by considering both the traffic efficiency and the safety along the road. Figure 3 (f) Statistical information of the preferred paths in Situations 1-5. In Situations 2-5, the paths without considering susceptibility pass through more sections with high and extremely high landslide susceptibility. Figure 3(f) It can be seen that the sum of the landslide susceptibility levels of the paths selected by Situations 1-5 considering susceptibility is smaller than the sum of the landslide susceptibility levels of the paths selected without considering susceptibility, indicating that the paths selected after considering susceptibility have lower landslide susceptibility and are safer to a certain extent.

[0106] Table 2 shows the quantitative evaluation results of path optimization of various methods, where CW represents the cumulative edge weight of the optimal path. ET represents the algorithm running time. SIT represents the single iteration time of the algorithm. It can be seen from the table that compared with the mainstream path optimization algorithms Dijkstra, A*, RPA, Jaya and RSA, LGNN has the best path optimization performance. It not only finds the optimal path (the smallest CW value), but also has the highest path optimization efficiency (the smallest ET value), which can better support safe, effective and efficient post-earthquake emergency rescue work.

[0107] Table 2. Quantitative evaluation results of various method path optimization

[0108]

[0109] The present application also discloses an electronic device. Figure 4 , Figure 4 The electronic device 500 may include: at least one processor 501 , at least one network interface 504 , a user interface 503 , a memory 505 , and at least one communication bus 502 .

[0110] The communication bus 502 is used to realize the connection and communication between these components.

[0111] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.

[0112] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0113] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for loading by a processor to execute the above-mentioned LGNN algorithm for optimizing post-earthquake rescue paths that couples safety and efficiency.

[0114] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0115] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A post-earthquake rescue path optimization LGNN algorithm that couples safety and efficiency, characterized by: include: S1: Acquire multi-source data; Through multi-source data, combined with CGBoost algorithm, the susceptibility of co-seismic landslides is predicted, and the susceptibility results are obtained and verified; the multi-source data include: terrain data, geological data, human engineering activity data, environmental data, rainfall data and earthquake data; S2: Construct a road map through multi-source data and susceptibility results; S3: Determine the rescue center and disaster-affected site based on the emergency rescue reports in multi-source data; S4: Building lightweight graph neural networks; S5: Input the road map, rescue centers, and disaster-affected sites into a lightweight graph neural network; The accumulated weight of each node from the rescue center to the road graph is used as the node feature to characterize the node feature; Through the lightweight graph neural network iterative node feature representation, the minimum cumulative weight of the disaster-stricken point is obtained. By backtracking the cumulative weight of the disaster-stricken point, the optimal rescue path from the rescue center to the disaster-stricken point is obtained.

2. The LGNN algorithm for post-earthquake rescue path optimization that couples safety and efficiency as claimed in claim 1, characterized in that: Step S1 includes: S11: Terrain data includes seamless digital elevation model data; geological data includes geological maps, pre-earthquake fault data and world digital soil maps; human engineering activity data includes land use data and pre-earthquake road network data; environmental data includes water network data; earthquake data includes vibration maps, post-earthquake 3D Google images, co-seismic landslide catalog data and real resettlement site data for earthquake emergency response; S12: Construct disaster-pregnancy and disaster-causing factors through terrain data, geological data, human engineering activity data, environmental data, rainfall data and vibration map; disaster-pregnancy and disaster-causing factors are 20 landslide susceptibility prediction indicators; S13: Based on 10% of the co-seismic landslide catalog data and disaster-pregnancy and disaster-causing factors, the CGBoost algorithm is used to predict the susceptibility of co-seismic landslides and obtain the susceptibility results; S14: Validate the predicted susceptibility results using all coseismic landslide inventory data.

3. The LGNN algorithm for post-earthquake rescue path optimization that couples safety and efficiency as claimed in claim 1, characterized in that: Step S2 includes: S21: Extract road grade and road length through pre-earthquake road network data in human engineering activity data; S22: Calculate the road travel time according to the road grade and road length; S23: quantifying the susceptibility results to determine the landslide susceptibility level; the landslide susceptibility level includes: very low susceptibility, low susceptibility, medium susceptibility, high susceptibility and very high susceptibility; S24: Select the maximum value of the landslide susceptibility level in the road buffer as the landslide susceptibility level P of road g g , g∈{1,2,…,NS}, NS represents the number of roads in the road network; S25: Determination of road accessibility: Based on the pre-earthquake road network data and the post-earthquake 3D Google images, the inaccessible roads are interpreted to obtain the interpretation results; the inaccessible roads include: roads blocked by landslides and roads blocked by the accumulation of collapsed buildings; S26: According to the interpretation results, the impassable roads are eliminated; the intersections of the impassable roads are used as nodes, the roads are used as edges, the road travel time and the landslide susceptibility level of the roads are used as edge weights, and a road map is constructed.

4. The LGNN algorithm for post-earthquake rescue path optimization that couples safety and efficiency as claimed in claim 3, characterized in that: Step S4 includes: The problem of optimizing rescue vehicle routes in a large-scale complex road network is abstracted into a mathematical problem and mathematical modeling is performed; Define the node where the rescue center is located as the starting point s of the vehicle, and the disaster point with serious casualties as the end point t of the vehicle. Both s and t are nodes in the graph data. Assume that the number of vehicles at the starting point s is 1 and the vehicle has the ability to rescue only one disaster point. The intersections of accessible roads are taken as nodes, roads are taken as edges, and the road travel time and the landslide susceptibility level of the road are taken as edge weights. The graph data is constructed as G(V,W), where N represents the number of road intersection nodes in the road network, and V = {0, 1, 2, ..., N-1} represents the node set. The adjacency matrix representing the edges of the graph data, is the weight matrix of the edge of the graph data; For a road g with nodes i and j, the corresponding edges are (i, j) and (j, i); (i, j) represents the road g from node i to node j; (j, i) represents the road g from node j to node i, so A i,j =A j,i =1, if road g is an impassable road, then A i,j =A j,i =0; Landslide susceptibility level P of road g g and road travel time T g are equally important for emergency rescue operations. If there is a road g, the weight W of the edge (i, j) is i,j =P g +T g , W i,j =W j,i ; If there is no road g, then W i,j =W j,i =∞; Define node s as the starting point, node t as the end point, a ′ ,c ′ ,d ′ ,u ′ , is a node different from s,t; s,a ′ ,c ′ ,d ′ ,u ′ ,t∈V; Define the matrix P and matrix C to represent the path matrix and cumulative weight matrix respectively. definition The element in the sth row and tth column of the matrix P represents the path from node s to node t; definition represents the element in the sth row and tth column of C, which represents the cumulative weight from node s to node t; If there are edges (s,a′),(a′,c′),(c′,d′),…,(u′,t), their corresponding paths are g1,g2,g3,…,g l ,g1,g2,g3,…,g l ∈{1,2,…,NS}, then the path from s to t is The corresponding cumulative weight from s to t is Based on the above definition, the path from the starting point s to the end point t on the graph data G(V,W) that considers security and timeliness is preferably described as a minimization problem on G(V,W), as follows: Where min ′ It represents the function of finding the minimum value of a variable; definition The minimum value of represents the minimum cumulative weight from s to t; The corresponding path from s to t is The minimum value of is the shortest path; Through the triangle inequality property of the shortest path, we get the following proposition and inference: Proposition 1: For represents the set of neighbor nodes of t and Known W u,t , W u,t represents the weight of the edge (u, t), where u represents a neighbor node of t; if given but min(·) is a function that finds the minimum value in a set. The corresponding u is u ′ ,but Corollary 1: Definition achieve When , the corresponding neighbor node of t is u ′ , then among the neighboring nodes u of node t, only u=u ′ satisfy That is, the predecessor node of t on the shortest path from s to t And at this time achieve For any other u=u e ,regardless Whether it is achieved have u e Indicates that u is not included ′ The set of neighbor nodes of node t; Will The solution is defined as of Minimization problem, i∈V, and by Solution The cumulative weight from node s to node i is used as the feature of node i, and Modeled as the representation process of node i; The node features of each node i are parallelly represented by using the node topology relationship and the message passing and feature aggregation mechanism of the graph neural network GNN. initial i≠s, due to the minimum feature aggregation mechanism, after iterative characterization Gradually converge to For i = s, we have exist After convergence, we can iterate back to the predecessor node through Corollary 1 and obtain 5. The LGNN algorithm for post-earthquake rescue path optimization that couples safety and efficiency as claimed in claim 4, characterized in that: Step S4 also includes: Using minimum value aggregation, the edge weights are involved in the aggregation of neighbor node features. The initial node representation of the lightweight graph neural network is as follows: in represents the kth iteration represents the cumulative edge weight from node s to node i; represents the k-1th iteration represents the cumulative edge weight from node s to node j; For k=0,i=s, we have For i = other∈V-{s}, we have For node t, h t express That is, the cumulative edge weight from node s to node t; when h t From k = 0 to k = k ′ , and from k = k ′ to k = k ' +τ no longer changes, that is, Reach the minimum value and find satisfaction through inference 1 The predecessor node u of t ′ , and then iterate back to get τ represents the maximum number of iterations that the node features on the graph remain unchanged after they are not ∞; In the positive weighted undirected graph G(V,W), s is the starting point and t is the end point. The bidirectional algorithm calculates the value of each node i from the starting point s and the end point t respectively. and If the node v and Both reach the minimum value. Since G(V,W) is an undirected graph, Corresponding Reversed(·) indicates a reversal function; In the LGNN bidirectional path optimization algorithm, and Represented as node features, let Therefore, for each node i, LGNN represents the node features from s to i in parallel and the node features from t to s D i ; The bidirectional node characterization process of the LGNN bidirectional path optimization algorithm is as follows: in, Represents the characteristics of node i at the kth iteration and When i = s When i=other When i = t When i=other For any node i, Therefore If there exists a node v such that When the node v corresponds to First converge to hour, has converged, and at other nodes i e ∈V-{v} corresponds to During the convergence process, has ceased to change, so when It can be judged when it drops to no longer change Convergence, without having to wait until any node i corresponds to No longer changes.

6. The LGNN algorithm for post-earthquake rescue path optimization that couples safety and efficiency as claimed in claim 5, characterized in that: Step S4 also includes: In the LGNN bidirectional path optimization algorithm, C s,k ,C t,k =D k [:,0],D k [:,1], C s,k represents the cumulative weight vector C from node s to all nodes represented by the kth iteration s ; C t,k Represents the cumulative weight vector C from node t to all nodes represented by the kth iteration t ; make Decrease until k=k ′ to k = k ′ When +τ no longer changes, it is considered has converged; Mark min(C s,k +C t,k ) corresponds to node v, then node v is the intersection node of bidirectional optimization, that is, the node on the shortest path; By formula get and Based on node v as the starting point and The optimal rescue path obtained by backtracking.

7. An electronic device, characterized in that: The electronic device comprises a processor (501), a memory (505), a user interface (503) and a network interface (504), wherein the memory (505) is used to store instructions, the user interface (503) and the network interface (504) are used to communicate with other devices, and the processor (501) is used to execute the instructions stored in the memory (505) so that the electronic device executes the post-earthquake rescue path optimization LGNN algorithm as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the post-earthquake rescue path optimization LGNN algorithm as described in any one of claims 1-6 is executed.

Images