Multi-agent cooperative offshore rescue task planning method and system

Abstract

The present application belongs to the technical field of task planning based on big data artificial intelligence, and particularly relates to a multi-agent cooperative offshore rescue task planning method and system. The method comprises S1: selecting the most suitable device combination for the current rescue task based on rescue device parameters, rescue target parameters and historical rescue device combination information through a particle swarm optimization algorithm, wherein the fitness score of the particle is determined by factors such as the navigation capability of the device, the environmental adaptation rescue cost of the target position and the sea conditions of the target area; S3: removing the area occupied by static obstacles in the gridded sea chart, extracting the passable grid and obtaining the global candidate route set; S4: updating the hydrological and meteorological dynamic data in the grid range of each route in the global candidate route set and solving the optimal smooth route. The purpose of synchronous configuration of rescue resources and navigation route, reuse of calculation results in the route planning process and reduction of the calculation overhead of the real-time obstacle avoidance algorithm is achieved.

Description

Technical Field

[0001] This invention belongs to the technical field of mission planning based on big data and artificial intelligence, and particularly relates to a method and system for planning maritime rescue missions in collaboration with multiple intelligent agents. Background Technology

[0002] Maritime rescue planning typically follows a three-step process: after the location of the rescue target is confirmed, the first step is to determine the optimal execution equipment based on the injury and the target sea conditions (determining "who to go"); the second step is to assess the feasibility of the rescue based on the static obstacles in the sea area between the departure point and the target location (determining "how to get there"); and the third step is to plan a specific route based on the dynamic obstacles between the preliminary routes (determining "how to get there specifically").

[0003] However, the methods used in the above three steps in the prior art (such as the multi-objective decision method in equipment selection, the A method in static path planning) (Or Dijkstra's algorithm, DWA in dynamic obstacle avoidance, or artificial potential field method) operate independently without forming an organic linkage. This leads to three problems: 1. The output results of equipment selection may not pass the subsequent path feasibility verification; 2. Static path planning schemes face frequent replanning in dynamic environments; 3. The calculation results of each step cannot be reused, resulting in high overall computational redundancy. Summary of the Invention

[0004] This application provides a multi-agent collaborative maritime rescue mission planning method and system, which aims to achieve the synchronous configuration of rescue resources and navigation routes, reuse of calculation results in the route planning process, and reduce the computational overhead of real-time obstacle avoidance algorithms.

[0005] The multi-agent cooperative maritime rescue mission planning method provided in this application includes: S0: Process the parameters of rescue equipment and rescue target to form standardized structured data; S1: Based on the parameters of rescue equipment, the parameters of rescue target, and the information of historical rescue equipment combinations, the particle swarm optimization algorithm is used to select the equipment combination that is most suitable for this rescue mission. The fitness score of the particles is determined by the equipment's navigation capability, the equipment's rescue capability, the historical similarity score, the environmental adaptation rescue cost of the target location, and the sea state of the target area. S2: Standardize the static obstacle data and dynamic meteorological and hydrological data of the sea area to generate raster nautical charts; S3: Remove the area occupied by static obstacles in the rasterized nautical chart, extract the passable grid, and obtain the global candidate route set. The global candidate route set contains the set of K shortest paths of the directed non-negative weighted graph from the starting point to the target point of the rescue equipment. S4: Update the hydrological and meteorological dynamic data within the grid range traversed by each route in the global candidate route set, and solve for the optimal smooth route.

[0006] In one implementation, S1 includes: S11: Initialize a particle swarm containing N one-dimensional device arrays of length D, and the initial position of any particle i. ,speed The optimal position found by particle i The optimal position found by the entire population Where D is the number of candidate devices, and the value of an element in the one-dimensional device array is 0 or 1, where 0 means that the device represented by the element is not selected, and 1 means that the device represented by the element is selected. S12: Update the velocity of particle i: , in, This represents the velocity of particle i in the next iteration, which is affected by the inertia of its current velocity. Inertia weight The optimal position found by particle i itself Individual learning factors The optimal position found through global search. and social learning factors Influence; S13: Adopt Update the position of particle i, where S represents the sigmoid function. Random(0,1) represents a random decimal number in the range (0,1); S14: Iterate through and update the velocity and position of particle i to obtain the position of particle i in D-dimensional space. ; S15: Evaluate the fitness of particle i at its current position. ; Where w is a weighting factor, navigation capability includes at least one of endurance, speed, and wind resistance level, rescue capability includes at least one of life-saving vehicle and passenger capacity, historical similarity score depends on historical rescue cases using this equipment, environmental adaptability includes at least one of ocean currents, wind speed, and water environment, and penalty items include at least one of rescue cost and sea state impact in the target area. S16: The fitness of particle i at its current position, calculated through S15, is compared with the fitness of particle i at its historical best position. If the fitness is better than the historical best, the optimal value is updated. Then, the fitness is compared with the fitness of the entire particle swarm at its historical best position. If the fitness is better than the historical best, the optimal value is updated. S17 Observe the particle convergence results and update. , , Factor, repeat S12 to S16 until a solution is found or a stopping condition is met after multiple iterations, and an optimal combination of devices is found.

[0007] In one implementation, S3 includes: S31: Construct a weighted directed graph with non-negative weights, including selecting the center point of a passable grid as a node, the direction from the starting point to the target point as a directed edge, and the distance of the edge as the edge weight. S32: Construct a shortest path tree, starting from the target point. Use Dijstra's algorithm to calculate the shortest edge from each node to the target point, forming a tree rooted at the target point. Any simple path from the starting point to the target point in this tree is the shortest path. The shortest distance is denoted as ; S33: Add deviation edges that are not part of the shortest path tree to the path calculation, including: assuming the starting point moves along a non-tree edge to a point that is not on the shortest path. From node M on the tree, walk along the tree edge to the target point, the distance is... , and the shortest distance The difference is ; The smaller the value, the less detour this deviation takes; S34: Repeat step S33 to calculate the values ​​generated by each node in the tree rooted at the target point. ,Will Sort by distance from the edge, forming an edge pile. ; S35: From the off-edge pile Select Replace short path Generate the k-th shortest path .

[0008] In one implementation, S3 further includes: S36: In areas with dense nodes, users remove some paths with high similarity.

[0009] In one implementation, S4 includes: S41: Update the hydrological and meteorological dynamics data within the raster range traversed by each route in the global candidate route set; S42: Set the starting point of the problem set for the ants. ,end Ant count and initial pheromone concentration matrix Heuristic functions volatile factors pheromone factors Heuristic function factor ; S43: Starting from time 0, every time period t, m ants start from the beginning of the problem set. Set off, in search of the destination path; S44: Let ant a be currently at position i. Calculate the probability that ant a chooses the next position j. , ;in The pheromone concentration along the path from position i to j within period t. For the heuristic function of the path from position i to j within period t; S45: If ant a has not yet reached the destination... If so, repeat step S44 to continue searching for the next position, but you cannot select a position that has already been visited as a new target; If ant a reaches the destination Update the pheromone concentration of all paths visited by ant a. The updated pheromone concentration is the path length from position i to position j. The reciprocal of the path length; It is obtained from the following 3 formulas:

[0010] Among them, wind speed and wind direction Wave height and wave direction Flow velocity and direction ; The smaller the value, the more stable the sea conditions at the next location; S46: After each time period t, update the pheromone concentration for the next period for all paths. :

[0011] in, The pheromone concentration left by ant a along path i to j in the previous period t, Δ: represents the increment, the pheromone increment of ant a along path i to j in period t; S47 The ant repeats S43 to S46 until, after multiple iterations, the pheromone concentration increases only on one or more paths, thus finding the optimal path as the optimal smooth route.

[0012] In one implementation, S0 includes: S01 inputs semantic device parameters and semantic target parameters into a pre-trained semantic model or a proprietary rescue equipment semantic model, extracts key information through the model, and outputs semi-structured data. Based on the standards for lifesaving and emergency response, S02 standardizes the semi-structured data output by S01 using big data tools, eliminates the dimensional influence between different parameters through a standardized dictionary set, and outputs unified structured data across the entire domain.

[0013] In one implementation, S2 includes: S21 performs parsing, vector extraction, spatial interpolation, and field transformation on static obstacle data in the sea area, and outputs GeoJSON data containing attributes and influence range; S22 performs parsing, spatiotemporal alignment, and interpolation operations on meteorological and hydrological dynamic data, and outputs GeoJSON data containing spatiotemporal identifiers, core parameters, and spatial morphology. S23 performs object extraction, coordinate system transformation, symbolization, and drawing operations on the original nautical chart to generate a base map. The GeoJSON data vectors formed in S21 and S22 are then overlaid on the base map to form the rasterized nautical chart.

[0014] This application also provides a multi-agent cooperative maritime rescue mission planning system, including: The structured data generation unit processes and outputs standardized structured data composed of rescue equipment parameters and rescue target parameters. Equipment Combination Decision Unit: Obtains structured data output by the structured data generation unit, and selects the most suitable equipment combination for this rescue mission based on the rescue equipment parameters, rescue target parameters and historical rescue equipment combination information through particle swarm optimization algorithm. The fitness score of the particles is determined by the equipment's navigation capability, equipment's rescue capability, historical similarity score, environmental adaptation rescue cost of the target location and sea conditions of the target area. The rasterized nautical chart generation unit is used to standardize and process static obstacle data and dynamic meteorological and hydrological data of the sea area, and output rasterized nautical charts. The global candidate route set calculation unit obtains the output of the rasterized nautical chart generation unit, removes the area occupied by static obstacles in the rasterized nautical chart, extracts passable grids, and obtains the global candidate route set, which contains the set of K shortest paths of the directed non-negative weighted graph from the starting point to the target point of the rescue equipment. The route smoothing unit obtains the global candidate route set output by the global candidate route set calculation unit, and updates the hydrological and meteorological dynamic data within the raster range traversed by each route in the global candidate route set based on the data of the rasterized nautical chart output by the rasterized nautical chart generation unit, and solves for the optimal smoothed route.

[0015] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-described maritime rescue mission planning methods.

[0016] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described maritime rescue mission planning methods.

[0017] Beneficial effects: 1) This application links the adaptability of the equipment combination with the sea conditions at the target location, thereby realizing the feasibility coupling between the equipment combination and the subsequent route. This solves the problem of "equipment selection not considering the route" or "route planning ignoring equipment performance" in traditional phased planning. Therefore, the output results of equipment selection can be verified through the feasibility of the subsequent route, realizing the synchronous configuration of rescue resources and navigation routes.

[0018] 2) This application uses the paths in the set of K shortest paths of the directed non-negative weighted graph as spatial constraints for solving the optimal smooth route, reducing the complex four-dimensional dynamic path planning problem to a local smoothing problem, and reducing the computational cost of the real-time obstacle avoidance algorithm while maintaining the global path economy.

[0019] 3) Furthermore, this application overcomes the premature convergence problem of single-group intelligent algorithms in multimodal environments by separating static obstacles from dynamic sea conditions and processing them separately by solving the set of K shortest paths of the directed non-negative weighted graph and solving the optimal smooth route, thereby improving the stability and reliability of route planning results for rescue in highly dynamic sea areas. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the multi-agent collaborative maritime rescue mission planning method provided in Embodiment 1 of this application; Figure 2 This is a schematic diagram of the process for obtaining device combinations based on PSO in Embodiment 1 of this application; Figure 3 This is a schematic diagram of a rasterized nautical chart in Embodiment 1 of this application; Figure 4 A schematic diagram of the process for obtaining the global candidate route set based on Eppstein in Embodiment 1 of this application; Figure 5 This is a schematic diagram of a passable grid after removing the area occupied by static obstacles in a nautical chart grid in Embodiment 1 of this application; Figure 6 This is a schematic diagram of the global candidate route set after user intervention in Embodiment 1 of this application; Figure 7 This is a schematic diagram of the process of solving the optimal smooth path based on ACO in Embodiment 1 of this application; Figure 8 This is a schematic diagram of the optimal smooth path solved based on ACO in Embodiment 1 of this application. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. The principles and features of the present invention are described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention. Example 1

[0022] like Figure 1 As shown, the multi-agent collaborative maritime rescue mission planning method provided in this embodiment includes the following steps: S0: Rescue equipment parameters and rescue target parameters are processed through semantic models and big data to form standardized structured data.

[0023] In this embodiment, S0 includes: S01: Input the semantic device parameters and semantic target parameters into the pre-trained semantic model or the proprietary rescue equipment semantic model, extract key information through the model, and output semi-structured data; S02: Based on the International Lifesaving Equipment Code, Emergency Supplies Coding and Attribute Description and other lifesaving and emergency field standards, the semi-structured data output by S01 is standardized by big data tools. The influence of different dimensions between different parameters is eliminated by a standardized dictionary set, and the structured data with uniformity across the entire domain is output.

[0024] For example, semantic equipment parameters include: A. Lifeboat is 12 meters long, has a shallow draft to enter shoals and reefs, a maximum speed of 20 to 40 knots, a range of 1,000 to 3,000 nautical miles, a maximum capacity of 6 to 30 people, a lifting weight of 100 to 500 kilograms, is equipped with an automatic righting function, is equipped with a drone, can operate safely in sea states of 4 to 5 and waves of 1.5 to 2.5 meters, has a waterproof rating of IP67 (can withstand short-term immersion in water without damage), and is wind resistant to force 7 to 8. Through semantic model keyword extraction and data governance, structured data is formed: { "id": "boat_A", "type_onehot": [1,0,0], "L_m": 12.0, "draft_cm": 80, "turn_radius_m": 15, "v_min_kts": 20, "v_max_kts": 40, "range_nm": [100, 300], "crew": [6, 30], "lift_kg": [100, 500], "self_right": 1, "sea_state_max": 5, "wave_height_m": 2.5, "ip_rating": 67, "wind_bf_max": 8}.

[0025] S1: Based on the parameters of the rescue equipment, the parameters of the rescue target, and the historical information on the combination of rescue equipment, the Particle Swarm Optimization (PSO) algorithm is used to select the combination of equipment that is most suitable for this rescue mission. The fitness score of the particles is determined by the equipment's navigation capability, the equipment's rescue capability, the historical similarity score, the environmental adaptation rescue cost of the target location, and the sea state of the target area. The parameters of the rescue equipment include coordinates, range, speed, wind resistance rating, and life-saving attachments, while the parameters of the rescue target include coordinates, casualties, injuries, ocean currents, and water temperature.

[0026] Combination Figure 2 In this embodiment, S1 includes: S11: Initialize a particle swarm containing N one-dimensional device arrays of length D, and the initial position of any particle i. ,speed The optimal position found by particle i The optimal position found by the entire population Where D is the number of candidate devices, and the value of an element in the one-dimensional device array is 0 or 1, where 0 means that the device represented by the element is not selected, and 1 means that the device represented by the element is selected. S12: Update the velocity of particle i: , in, This represents the velocity of particle i in the next iteration, which is affected by the inertia of its current velocity. Inertia weight The optimal position found by particle i itself Individual learning factors The optimal position found through global search. and social learning factors Influence; In the continuous PSO problem, the position of particle i can be updated using... Formula. However, in this embodiment, in solving the discrete problem of device combination, a one-dimensional device array of length D (such as [1,0,1,1,0,...], where 1 indicates selecting the device and 0 indicates not selecting it) is needed to represent the position of particle i. Then: S13: Adopt Update the position of particle i, where S represents the sigmoid function. Random(0,1) represents a random decimal number in the range (0,1); S14: Iterate through and update the velocity and position of particle i to obtain the position of particle i in D-dimensional space. ; Since the fitness evaluation in this embodiment involves multi-dimensional qualitative rules (such as threshold determination, conditional bonus / subtraction) and discretized scoring, it is not suitable to directly fit using a single continuous function. It is more suitable to perform structured calculations through a rule engine or vertical domain model. Therefore: S15: Evaluate the fitness of particle i at its current position. ; Where w is a weighting factor, navigation capability includes at least one of endurance, speed, and wind resistance level, rescue capability includes at least one of life-saving vehicle and passenger capacity, historical similarity score depends on historical rescue cases using this equipment, environmental adaptability includes at least one of ocean currents, wind speed, and water environment, and penalty items include at least one of rescue cost and sea state impact in the target area. For example, if rescue boat A's range exceeds the distance to the rescue target location, it receives 0.5 points; if the number of rescue vehicles exceeds the number of personnel in the target area, it receives 1 point; if the equipment combination has performed similar sea area rescue missions, it receives 0.5 points; if equipped with drones to increase the search range and aerial delivery capability, it receives 1 point; if equipped with a crane, it receives 0.5 points; if the wind resistance level does not meet the wind speed of the target sea area, it loses 1 point; if W is set to 1, the fitness score is 2.5.

[0027] S16: The fitness of particle i at its current position, calculated through S15, is compared with the fitness of particle i at its historical best position. If the fitness is better than the historical best, the optimal value is updated. Then, the fitness is compared with the fitness of the entire particle swarm at its historical best position. If the fitness is better than the historical best, the optimal value is updated.

[0028] S17 Observe the particle convergence results and update. , , Factor, repeat S12 to S16 until a solution is found or a stopping condition is met after multiple iterations, and an optimal combination of devices is found.

[0029] The stopping conditions include any of the following: 1) The training rounds are reached, the particle swarm does not aggregate, and no matching device combination is found; 2) The training rounds are reached, the particle swarm aggregates at one or more locations, and one or more matching device combinations are found.

[0030] S2 standardizes multi-source, heterogeneous static obstacle data and dynamic meteorological and hydrological data of the sea area to generate rasterized nautical charts, such as... Figure 3 As shown in the image, the map displays the areas where hydrological and meteorological obstacles are located, the areas where marine obstacles are located, restricted areas, and the locations of rescue targets.

[0031] In this embodiment, S0 and S2 can be performed simultaneously or completed in advance before task planning. The static obstacle data and dynamic meteorological and hydrological data of the sea area should be kept up-to-date or remain largely unchanged.

[0032] In this embodiment, S2 includes: S21: Analyze, extract vectors, perform spatial interpolation and field transformation on static obstacle data (reefs, shipwrecks, etc.) and restricted areas in the sea area, and output GeoJSON data containing attributes and influence range; S22: Parse, spatiotemporally align, and interpolate dynamic meteorological and hydrological data (wind, waves, currents, etc.) to output GeoJSON data containing spatiotemporal identifiers, core parameters, and spatial morphology; S23: Extract landmarks, transform coordinate systems, symbolize and draw symbols from the original nautical chart to generate a base map. Overlay the GeoJSON data vectors formed in S21 and S22 onto the base map to form a coarse-grained rasterized nautical chart with a resolution of several kilometers to tens of kilometers.

[0033] S3: Remove the area occupied by static obstacles from the nautical chart grid and extract the passable grid, such as... Figure 5 The green grid represents the feasible path from the starting point to the target for the rescue equipment. This can be viewed as a K-Shortest Paths (KSP) problem in a directed non-negative weighted graph. In this embodiment, the KSP problem is solved using the Eppstein algorithm.

[0034] Combination Figure 4 In this embodiment, S3 includes: S31 Construct a weighted directed graph G with non-negative weights, including selecting the center point of a passable grid as a node, the direction from the starting point U to the target point T as a directed edge, and the distance of the edge as the edge weight. S32 Construct a shortest path tree, starting from the target point T. Using Dijstra's algorithm, calculate the shortest edge from each node X to the target point T (there may be multiple short edges of equal distance from X to T; choose one). This forms a tree rooted at the target point T. Any simple path from the starting point U to the target point T in this tree is the shortest path. The shortest distance is denoted as ; S33 adds deviation edges that are not part of the shortest path tree to the path calculation, including assuming the starting point U moves along a non-tree edge to a point that is not on the shortest path. From node M on the tree, follow the tree edge to the target point T, the distance is... , and the shortest distance The difference is , The smaller the value, the less detour this deviation takes; S34 Repeat step S33 to calculate the nodes generated by each node in the tree rooted at the target point T. ,Will Sort by distance from the edge to form an edge pile. ; S35 from the off-side pile Select replace (That is, the current node does not take the shortest path, but takes a detour) Generate the k-th shortest path. ; S36 is located in a densely populated area because of the deviation between edges. The small differences in similarity mean that the calculated rankings of adjacent paths are not significantly different from actual rescue routes. Therefore, it's advisable to selectively remove some highly similar paths and manually intervene in the global candidate route set, such as... Figure 6 As shown, multiple short routes may be calculated from the left direction P1, but they are only slightly different from P1. For ease of explanation, the left-side route P1 and the right-side route P2 are manually selected as global candidate routes.

[0035] S4: Update the hydrological and meteorological dynamic data within the raster range traversed by each route in the global candidate route set, and solve for the optimal smoothed route. (The last part, "Globally optimal route," appears to be an error and doesn't need a direct translation.) The grid range is several kilometers, and the available routes for rescue equipment are affected by the dynamic changes in hydrology and meteorology within this range. In this embodiment, the optimal smooth route is obtained by dynamically calculating the impact of local obstacles based on the Ant Colony Algorithm (ACO).

[0036] Combination Figure 7 In this embodiment, S4 includes: S41: Using the hydrological and meteorological dynamic data obtained in S22, update the hydrological and meteorological dynamic data within the grid range traversed by each route in the global candidate route set; S42: Set the starting point of the problem set for the ants. ,end Ant count and initial pheromone concentration matrix Heuristic functions volatile factors pheromone factors Heuristic function factor ; S43: Starting from time 0, every time period t, m ants start from the beginning of the problem set. Set off, in search of the destination path; S44: Each ant can independently choose its path to the destination. The ant follows a route, making decisions based on the pheromone concentration along the path at the current time and heuristic information when searching for its next location. Let ant a be currently at location i, and calculate the probability that ant a chooses the next location j. , ;in The pheromone concentration along the path from position i to j within period t. For the heuristic function of the path from position i to j within period t; S45: If ant a has not yet reached the destination... If so, repeat step S44 to continue searching for the next position, but you cannot select a position that has already been visited as a new target; If ant a reaches the destination Update the pheromone concentration of all paths visited by ant a. The updated pheromone concentration is the path length from position i to position j. The reciprocal of the path length; It is obtained from the following 3 formulas:

[0037] Among them, wind speed and wind direction Vd, wave height and wave direction Flow velocity and direction ; The smaller the value, the more stable the sea conditions at the next location.

[0038] S46: After each time period t, update the pheromone concentration for the next period for all paths. This is to avoid local rapid convergence of the solution space, which could prevent the global optimal solution from being found.

[0039] in, The pheromone concentration left by ant a along path i to j in the previous period t; Δ: represents the increment, the pheromone increment of ant a along path i to j in period t.

[0040] S47. Ants repeat steps S43 to S46 until, after multiple iterations, the pheromone concentration increases only along one or more fixed paths, thus finding the optimal path as the optimal smooth route. Figure 8 The blue dashed lines represent the optimal routes for the red paths P1 and P2 after ACO smoothing.

[0041] This embodiment also provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement any of the above-described maritime rescue mission planning methods.

[0042] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described maritime rescue mission planning methods. Example 2

[0043] This embodiment provides a multi-agent collaborative maritime rescue mission planning system for performing the maritime rescue mission planning method provided in Embodiment 1. The system includes: The structured data generation unit processes and outputs standardized structured data based on the parameters of rescue equipment and the parameters of rescue targets.

[0044] Equipment Combination Decision Unit: Obtains structured data output by the structured data generation unit, and selects the most suitable equipment combination for this rescue mission based on the rescue equipment parameters, rescue target parameters and historical rescue equipment combination information through particle swarm optimization algorithm. The fitness score of the particles is determined by the equipment's navigation capability, equipment's rescue capability, historical similarity score, environmental adaptation rescue cost of the target location and sea conditions of the target area. The rasterized nautical chart generation unit is used to standardize and process static obstacle data and dynamic meteorological and hydrological data of the sea area, and output rasterized nautical charts. The global candidate route set calculation unit obtains the output of the rasterized nautical chart generation unit, removes the area occupied by static obstacles in the rasterized nautical chart, extracts passable grids, and obtains the global candidate route set, which contains the set of K shortest paths of the directed non-negative weighted graph from the starting point to the target point of the rescue equipment. The route smoothing unit obtains the global candidate route set output by the global candidate route set calculation unit, and updates the hydrological and meteorological dynamic data within the raster range traversed by each route in the global candidate route set based on the data of the rasterized nautical chart output by the rasterized nautical chart generation unit, and solves for the optimal smoothed route.

[0045] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0046] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A multi-agent collaborative maritime rescue mission planning method, characterized in that, include: S0: Process the parameters of rescue equipment and rescue target to form standardized structured data; S1: Based on the parameters of rescue equipment, the parameters of rescue target, and the information of historical rescue equipment combinations, the particle swarm optimization algorithm is used to select the equipment combination that is most suitable for this rescue mission. The fitness score of the particles is determined by the equipment's navigation capability, the equipment's rescue capability, the historical similarity score, the environmental adaptation rescue cost of the target location, and the sea state of the target area. S2: Standardize the static obstacle data and dynamic meteorological and hydrological data of the sea area to generate raster nautical charts; S3: Remove the area occupied by static obstacles in the rasterized nautical chart, extract the passable grid, and obtain the global candidate route set. The global candidate route set contains the set of K shortest paths of the directed non-negative weighted graph from the starting point to the target point of the rescue equipment. S4: Update the hydrological and meteorological dynamic data within the grid range traversed by each route in the global candidate route set, and solve for the optimal smooth route.

2. The maritime rescue mission planning method according to claim 1, characterized in that, S1 includes: S11: Initialize a particle swarm containing N one-dimensional device arrays of length D, and the initial position of any particle i. ,speed The optimal position found by particle i The optimal position found by the entire population Where D is the number of candidate devices, and the value of an element in the one-dimensional device array is 0 or 1, where 0 means that the device represented by the element is not selected, and 1 means that the device represented by the element is selected. S12: Update the velocity of particle i: ， in, This represents the velocity of particle i in the next iteration, which is affected by the inertia of its current velocity. Inertia weight The optimal position found by particle i itself Individual learning factors The optimal position found through global search. and social learning factors Influence; S13: Adopt Update the position of particle i, where S represents the Sigmoid function. Random(0,1) represents a random decimal number in (0,1); S14: Iterate through and update the velocity and position of particle i to obtain the position of particle i in D-dimensional space. ; S15: Evaluate the fitness of particle i at its current position. ; Where w is a weighting factor, navigation capability includes at least one of endurance, speed, and wind resistance level, rescue capability includes at least one of life-saving vehicle and passenger capacity, historical similarity score depends on historical rescue cases using this equipment, environmental adaptability includes at least one of ocean currents, wind speed, and water environment, and penalty items include at least one of rescue cost and sea state impact in the target area. S16: The fitness of particle i at its current position, calculated through S15, is compared with the fitness of particle i at its historical best position. If the fitness is better than the historical best, the optimal value is updated. Then, the fitness is compared with the fitness of the entire particle swarm at its historical best position. If the fitness is better than the historical best, the optimal value is updated. S17 Observe the particle convergence results and update. , , Factor, repeat S12 to S16 until a solution is found or a stopping condition is met after multiple iterations, and an optimal combination of devices is found.

3. The maritime rescue mission planning method according to claim 1, characterized in that, S3 include: S31: Construct a weighted directed graph with non-negative weights, including selecting the center point of a passable grid as a node, the direction from the starting point to the target point as a directed edge, and the distance of the edge as the edge weight. S32: Construct a shortest path tree, starting from the target point. Use Dijstra's algorithm to calculate the shortest edge from each node to the target point, forming a tree rooted at the target point. Any simple path from the starting point to the target point in this tree is the shortest path. The shortest distance is denoted as ; S33: Add deviation edges that are not part of the shortest path tree to the path calculation, including: assuming the starting point moves along a non-tree edge to a point that is not on the shortest path. From node M on the tree, walk along the tree edge to the target point, the distance is... , and the shortest distance The difference is ; The smaller the value, the less detour this deviation takes; S34: Repeat step S33 to calculate the values ​​generated by each node in the tree rooted at the target point. ,Will Sort by distance from the offset edge, forming an offset edge heap. ; S35: From the off-edge pile Select Replace short path Generate the k-th shortest path .

4. The maritime rescue mission planning method according to claim 3, characterized in that, S3 also includes: S36: In areas with dense nodes, users remove some paths with high similarity.

5. The maritime rescue mission planning method according to claim 3, characterized in that, S4 include: S41: Update the hydrological and meteorological dynamics data within the raster range traversed by each route in the global candidate route set; S42: Set the starting point of the problem set for the ants. ,end Ant count and initial pheromone concentration matrix Heuristic functions volatile factors pheromone factors Heuristic function factor ; S43: Starting from time 0, every time period t, m ants start from the beginning of the problem set. Set off, in search of the destination path; S44: Let ant a be currently at position i. Calculate the probability that ant a chooses the next position j. , ;in The pheromone concentration along the path from position i to j within period t. For the heuristic function of the path from position i to j within period t; S45: If ant a has not yet reached the destination... If so, repeat step S44 to continue searching for the next position, but you cannot select a position that has already been visited as a new target; If ant a reaches the destination Update the pheromone concentration of all paths visited by ant a. The updated pheromone concentration is the path length from position i to position j. The reciprocal of the path length; It is obtained from the following 3 formulas: Among them, wind speed and wind direction Wave height and wave direction Flow velocity and direction ; The smaller the value, the more stable the sea conditions at the next location; S46: After each time period t, update the pheromone concentration for the next period for all paths. : in, The pheromone concentration left by ant a along path i to j in the previous period t, Δ: represents the increment, the pheromone increment of ant a along path i to j in period t; S47 The ant repeats S43 to S46 until, after multiple iterations, the pheromone concentration increases only on one or more paths, thus finding the optimal path as the optimal smooth route.

6. The maritime rescue mission planning method according to claim 1, characterized in that, S0 includes: S01 inputs semantic device parameters and semantic target parameters into a pre-trained semantic model or a proprietary rescue equipment semantic model, extracts key information through the model, and outputs semi-structured data. Based on the standards for lifesaving and emergency response, S02 standardizes the semi-structured data output by S01 using big data tools, eliminates the dimensional influence between different parameters through a standardized dictionary set, and outputs unified structured data across the entire domain.

7. The maritime rescue mission planning method according to claim 1, characterized in that, S2 include: S21 performs parsing, vector extraction, spatial interpolation, and field transformation on static obstacle data in the sea area, and outputs GeoJSON data containing attributes and influence range; S22 performs parsing, spatiotemporal alignment, and interpolation operations on meteorological and hydrological dynamic data, and outputs GeoJSON data containing spatiotemporal identifiers, core parameters, and spatial morphology. S23 performs object extraction, coordinate system transformation, symbolization, and drawing operations on the original nautical chart to generate a base map. The GeoJSON data vectors formed in S21 and S22 are then overlaid on the base map to form the rasterized nautical chart.

8. A multi-agent collaborative maritime rescue mission planning system, characterized in that, include: The structured data generation unit processes and outputs standardized structured data composed of rescue equipment parameters and rescue target parameters. Equipment Combination Decision Unit: Obtains structured data output by the structured data generation unit, and selects the most suitable equipment combination for this rescue mission based on the rescue equipment parameters, rescue target parameters and historical rescue equipment combination information through particle swarm optimization algorithm. The fitness score of the particles is determined by the equipment's navigation capability, equipment's rescue capability, historical similarity score, environmental adaptation rescue cost of the target location and sea conditions of the target area. The rasterized nautical chart generation unit is used to standardize and process static obstacle data and dynamic meteorological and hydrological data of the sea area, and output rasterized nautical charts. The global candidate route set calculation unit obtains the output of the rasterized nautical chart generation unit, removes the area occupied by static obstacles in the rasterized nautical chart, extracts passable grids, and obtains the global candidate route set, which contains the set of K shortest paths of the directed non-negative weighted graph from the starting point to the target point of the rescue equipment. The route smoothing unit obtains the global candidate route set output by the global candidate route set calculation unit, and updates the hydrological and meteorological dynamic data within the raster range traversed by each route in the global candidate route set based on the data of the rasterized nautical chart output by the rasterized nautical chart generation unit, and solves for the optimal smoothed route.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the maritime rescue mission planning method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the maritime rescue mission planning method according to any one of claims 1 to 8.

Images