Multi-unmanned aerial vehicle cooperative transportation dynamic decision-making method and system in emergency environment

Abstract

The invention provides a multi-unmanned aerial vehicle cooperative transportation dynamic decision-making method and system in an emergency environment, and the method comprises the steps: 1, obtaining the material state information reported by a disaster region node after an emergency occurs; step 2, constructing a hunter and prey local decision model taking the disaster-affected node as a main body, and forming a local priority and competition relationship between the nodes; 3, establishing a multi-target minimum cost maximum flow (MCMF) model, and solving to obtain a candidate transportation scheme; and step 4, inputting the candidate transportation scheme as an initial population into an improved non-dominated sorting genetic algorithm NSGA-II, performing non-dominated optimization on energy consumption, time, fairness and risk, and generating an optimal collaborative transportation strategy cluster approaching the Pareto frontier. According to the method, through a mode of combining local decision and global optimization, comprehensive improvement of multi-unmanned aerial vehicle collaborative transportation efficiency, fairness and safety is realized in a complex dynamic emergency environment.

Description

Technical Field

[0001] This invention relates to the field of emergency logistics scheduling and UAV collaborative control technology, specifically to a dynamic decision-making method and system for multi-UAV collaborative transportation in emergency environments. Background Technology

[0002] In emergencies such as earthquakes, explosions, and hazardous chemical spills, affected areas are often characterized by multiple locations, heterogeneous needs, severe road damage, and high timeliness requirements. Drones, with their high mobility, flexible deployment, and independence from ground transportation, have become an important means of transporting emergency supplies. However, existing drone emergency transport scheduling technologies still have the following shortcomings: Most existing methods employ rule-driven or single optimization models, making it difficult to simultaneously consider the individual urgency of disaster-stricken nodes and the overall efficiency at the system level, easily leading to local optima at the expense of global performance; most scheduling models assume that demand, resources, and network structure remain constant within the decision-making cycle, ignoring the objective facts of continuous changes in node states, real-time consumption of materials, and dynamic evolution of drone states during emergency events, resulting in insufficient model feasibility; existing minimum cost flow and shortest path methods typically perform simple weighted processing of multiple objectives, making it difficult to characterize the complex trade-offs between energy consumption, timeliness, fairness, and risk, and thus difficult to obtain high-quality multi-objective solution sets; in traditional flow models, materials can continuously flow through the network, but in emergency scenarios, materials are consumed or stored once they reach the demand node and should not continue to flow out, a scenario that existing models fail to adequately depict.

[0003] Therefore, there is an urgent need for a collaborative optimization method that can dynamically sense node status, integrate local game theory decision-making with global flow optimization, and perform fine-grained search through multi-objective evolutionary algorithms, in order to improve the overall decision-making performance and robustness of multi-UAV emergency transport systems in complex and uncertain environments. Summary of the Invention

[0004] Objective of the Invention: The technical problem to be solved by this invention is to address the shortcomings of existing technologies by providing a dynamic decision-making method and system for multi-UAV collaborative transportation in emergency environments. This method integrates a dynamic hunter-gatherer local decision-making model, a multi-objective minimum-cost maximum flow model, and an improved non-dominated sorting genetic algorithm, and is applicable to multi-UAV collaborative transportation scheduling in emergencies such as earthquakes, explosions, and hazardous chemical leaks.

[0005] This method includes the following steps:

[0006] Step 1: After an emergency occurs, obtain the material status information reported by the nodes in the disaster area. The material status information includes the node's material inventory, demand scale, urgency level, and risk level.

[0007] Step 2: Based on the material status information, construct a local decision-making model of hunters and prey with disaster-stricken nodes as the main body. Each node with material demand is a hunter node, and the suitability of potential material supply nodes as prey nodes is independently evaluated, forming a local priority and competition relationship between nodes.

[0008] Step 3: Based on the local decision-making results of the hunter and prey, establish a multi-objective minimum cost maximum flow (MCMF) model, dynamically adjust the cost parameters of the corresponding nodes and edges in the MCMF model, construct a transportation network that satisfies the capacity constraint and flow conservation constraint, and solve for candidate transportation schemes.

[0009] Step 4: Using candidate transportation schemes as initial population input, the improved non-dominated sorting genetic algorithm NSGA-II is used to perform non-dominated optimization on energy consumption, time, fairness and risk, generating an optimal cooperative transportation strategy cluster that approximates the Pareto front.

[0010] Step 1 includes the following steps:

[0011] Step 1-1: Emergency Triggering and Dispatch System Initialization;

[0012] Following a public emergency, the dispatch system is automatically triggered, entering the emergency dispatch initialization state, determining the currently affected area, and generating a set of affected nodes. , where each node Each disaster-stricken area, temporary resettlement site, or emergency material support node serves as the basic decision-making unit for subsequent coordinated dispatch.

[0013] Step 1-2: Multi-node state awareness and asynchronous information reporting:

[0014] Each node At any moment Report the actual material status information vector to the scheduling system :

[0015] ,

[0016] in, Represents a node At any moment The total amount of emergency supplies currently available; Represents a node The size of the population currently served or covered; Represents a node The indicators of material supply recovery capacity or external supply accessibility; Represents a node The credibility or reliability weight of the information reported;

[0017] Steps 1-3: Information credibility weighting and anomaly suppression processing;

[0018] After receiving the state information vectors from each node, the scheduling system performs weighted processing on the data reported by the nodes to form a reliable state estimate:

[0019] ,

[0020] ;

[0021] in, Represents a node At any moment Weighted credible inventory of materials; Represents a node At any moment The weighted trusted service population size;

[0022] Steps 1-4: Calculate the supply and demand level indicators of materials at each node;

[0023] Based on the weighted state information, the scheduling system calculates the time of each node. per capita material quantity index :

[0024] ;

[0025] Steps 1-5: Global statistics update and dynamic threshold preparation;

[0026] The scheduling system calculates statistical reference values ​​at the system level based on the average material quantity per person across all nodes.

[0027] ,

[0028] ,

[0029] ,

[0030] in This represents the average amount of supplies per person across all nodes. Indicates at time The maximum value of the per capita material quantity index for all disaster-affected nodes; Indicates at time The minimum value of per capita material quantity for all affected nodes.

[0031] Step 2 includes:

[0032] Step 2-1: The system at time... Calculate the global dynamic threshold :

[0033] ,

[0034] in It is the standard deviation of per capita material quantity; These are adjustable adaptive weight parameters;

[0035] Step 2-2: Establish prey node determination rules;

[0036] If node satisfy:

[0037] ,

[0038] Then the node Nodes identified as prey are denoted as follows: ;in Indicates time The set of prey nodes;

[0039] For any node where there is a shortage of supplies When satisfied At that time, node Considered a hunter node, its goal is to acquire prey nodes. The assessment of potential support recipients is ongoing.

[0040] Steps 2-3: Modeling the competition constraints among hunters;

[0041] Introduce normalized weights:

[0042] ,

[0043] in Indicates at time Hunter node For prey nodes Normalized hunting preference weights;

[0044] Introducing a contention load function for prey nodes :

[0045] ,

[0046] in Represents the set of all hunter nodes;

[0047] Hunter Node For all prey nodes Calculate the local hunting utility value independently:

[0048] ,

[0049] in, The function representing the resource redundancy of a prey node; Represents the hunter node With prey node The flight distance or time cost between; This indicates the risk inherent in the prey node j itself; These are the weighting coefficients;

[0050] when When the load exceeds the prey node's capacity, the corresponding utility value will be dynamically penalized:

[0051] ,

[0052] in Indicates at time Hunter node From the prey node The initial local revenue value of acquiring resources; Indicates at time Hunter node From the prey node The actual profit value of acquiring resources, adjusted for competition penalties; It is a coefficient greater than 0, used to control the strength of the competition penalty. The larger the value, the stronger the system's ability to suppress competitive behavior; even slight competition will significantly reduce gains.

[0053] Simultaneously, a competition penalty item is introduced. :

[0054] ,

[0055] in Indicates at time The set of all nodes identified as hunters; Let be the decision variable, representing the hunter. Should you choose your prey? ,if A value of 1 indicates selection, otherwise it indicates no selection; For indicator functions;

[0056] Steps 2-4: Establish a local decision-making model for the hunter node and construct the reward functions for the hunter and prey;

[0057] Hunter Node Targeting the prey node Constructing a local reward function :

[0058] ,

[0059] in It is a node With prey node Spatial distance or flight time between them; It refers to the energy consumption or risk cost per unit of material transportation; These are weighting coefficients;

[0060] Each hunter node Independently solve for local optimal prey:

[0061] ;

[0062] in Indicates at time Hunter node The optimal prey node number is selected after independent evaluation based on the local reward function;

[0063] Employing an active prey transport mechanism:

[0064] The hunter node only expresses needs and evaluates preferences;

[0065] Actual transportation is determined by the prey node. Initiate a request to the hunter node.

[0066] In step 3, when constructing the minimum cost maximum flow (MCMF) model:

[0067] Prey node With the Hunter Node The potential transport relationships between them are mapped as directed edges. Unit cost of the edge Defined as:

[0068] ,

[0069] in Basic transportation costs; This is the distance cost weighting coefficient; The weighting factor is used to calculate the partial returns.

[0070] In step 3, a multi-objective minimum cost maximum flow network is constructed. It includes the following structures:

[0071] Node set Including super source nodes Prey node set Hunter node set and Super Hub Node ;

[0072] in For at any time The first in the prey node set The last node, i.e., the last prey node; For at any time The first in the prey node set The last hunter node;

[0073] Directed edge set ,include:

[0074] Source to the edge of the prey ,capacity ,cost ;in Represents the prey node At any moment The amount of surplus materials that can be used for external allocation;

[0075] The prey approached the hunter. ,capacity ,cost Weighted sum of multiple factors:

[0076] ,

[0077] in, For preference weights and competition intensity The dynamic adjustment terms constituted; ; It is the cost of transportation time; It is a path risk factor; This is the distance cost weighting coefficient; This is the time cost weighting coefficient; This refers to the risk cost weighting coefficient. These are the weighting coefficients for the local decision adjustment term;

[0078] In the hunter local model, each hunter node Form a non-binary preference vector for all prey nodes. :

[0079] ,

[0080] in For at any time Hunter node For the prey node set The last (i.e., the last in numbered order) prey node Normalized preference weights; , Represents the prey node For the hunter node The degree of relative compatibility;

[0081] Hunters to the edge of the river ,capacity ,cost ;

[0082] The multi-objective minimum cost maximum flow (MCMF) model satisfies the following flow conservation constraints:

[0083] Source node constraints:

[0084] ,

[0085] That is, from the super source node The total amount of all outflowing resources cannot exceed the sum of the total amount of resources currently available for allocation across all hunting nodes;

[0086] Prey node constraints:

[0087] ,

[0088] That is, from any prey node The total amount of materials transported out cannot exceed the current available resources of that node.

[0089] Hunter node constraints (terminal consumption):

[0090] ,

[0091] That is, any hunter node The total amount of all materials received is equal to the amount that is ultimately consumed or retained, and this total amount cannot exceed the demand gap at that node.

[0092] Based on this, solve the following minimum-cost maximum flow problem to obtain candidate transportation solutions: ,

[0093] ,

[0094] in, This represents the maximum emergency response capacity achievable at the current moment.

[0095] In step 4, the improved non-dominated sorting genetic algorithm NSGA-II includes the following steps:

[0096] Step 4-1: Encode the candidate transportation schemes into population individuals. Each individual represents a complete UAV cooperative transportation decision scheme, including the following set of decision variables:

[0097] Actual resource flow from prey node to hunter node ;

[0098] Corresponding transportation route selection variables ;

[0099] Service priority or scheduling timing parameters of hunter nodes ;

[0100] Adjusting weights based on path risk or congestion ;

[0101] Step 4-2: To address the multi-dimensional decision-making requirements of the collaborative transportation of emergency supplies, the following multi-objective optimization function set is constructed:

[0102] ,

[0103] Among them, the energy consumption objective function is... for:

[0104] ,

[0105] This indicates the overall energy consumption cost of the drone during the transportation mission;

[0106] Time objective function for:

[0107]

[0108] This is used to minimize the arrival time of materials at the node with the greatest demand, reflecting the efficiency of emergency response;

[0109] Fairness objective function for:

[0110] ,

[0111] in To determine the actual amount of supplies obtained, This represents the demand, used to measure the degree of balanced protection across different disaster-stricken areas.

[0112] Risk objective function for:

[0113] ,

[0114] in The path risk coefficient reflects the uncertainty of the flight environment and safety risks.

[0115] Step 4-3: Use the fast non-dominated sorting method to divide the population into multiple non-dominated layers:

[0116] ;

[0117] Any two individuals and If the following conditions are met:

[0118] ,

[0119] Then it is called Dominate ;

[0120] Prioritize retaining low-level (high-priority) non-dominated solutions to ensure that the solution set approximates the Pareto front;

[0121] Step 4-4: Execute genetic operators to generate offspring population:

[0122] Adaptive crossover operation: The crossover probability is dynamically adjusted with the number of iterations.

[0123] ,

[0124] in This represents the current iteration number. This represents the maximum number of iterations. Indicates the first In the next iteration, the algorithm uses an adaptive crossover probability; This is the preset maximum crossover probability; This is the preset minimum crossover probability;

[0125] Risk-guided mutation manipulation: For high-risk individuals, enhance mutation probability.

[0126] ,

[0127] in This represents the current average risk value of the population. This is the adjustment coefficient; For individuals Risk-adjusted mutation probability; The baseline probability of variation; Represents an individual The risk objective function value;

[0128] Steps 4-5: Merge parent and offspring populations using an elite retention strategy, and select individuals to form a new generation population based on non-dominated hierarchy and crowding distance; use rapid non-dominated sorting to divide the population into two or more non-dominated hierarchies, and calculate the improved crowding distance for individuals within the same non-dominated hierarchy. ;

[0129] Steps 4-6: Terminate the algorithm when the maximum number of iterations is reached or the non-dominated front converges for multiple consecutive generations, and output the final non-dominated solution set as the optimal cooperative transportation strategy cluster.

[0130] In steps 4-5, the improved congestion distance is calculated using the following formula. :

[0131] ,

[0132] in is the weighting coefficient; M is the total number of optimization objectives; This indicates that in the current non-dominated layer, the first... The maximum value of each objective function; This indicates that in the current non-dominated layer, the first... The minimum value of an objective function; Indicates that in the order of the first After sorting the objective function values ​​in ascending order, the individual The next adjacent individual's first One objective function value; Indicates that in the order of the first After sorting the objective function values ​​in ascending order, the individual The next adjacent individual's first The objective function value.

[0133] This invention also provides a dynamic decision-making system for multi-UAV collaborative transportation in an emergency environment to implement the method, comprising: an information perception module for acquiring and processing material status information reported by each node; a local decision-making module for executing a hunter-prey local decision-making model and outputting node preferences and competitive relationships; a global optimization module for constructing and solving a multi-objective minimum cost maximum flow model to generate candidate transportation schemes; and a multi-objective evolution module for executing an improved non-dominated sorting genetic algorithm to perform multi-objective optimization on the candidate schemes and output a Pareto optimal policy cluster.

[0134] The present invention also provides an electronic device, including a processor and a memory, wherein the memory stores program code that, when executed by the processor, implements the method described herein.

[0135] The present invention also provides a storage medium storing a computer program or instructions that, when the computer program or instructions are run on a computer, execute the steps of the method described.

[0136] Beneficial Effects: Compared with existing multi-UAV emergency dispatch methods, this invention has significant advantages in modeling level and decision-making mechanism. First, by introducing a dynamic hunter-prey model, each demand node can autonomously report and make locally rational decisions based on its own real state, thereby effectively characterizing the asymmetric game relationship between nodes in an emergency environment with incomplete information and highly heterogeneous demands. This model avoids the strong dependence on globally accurate information in traditional centralized methods and improves the system's adaptability to sudden disturbances.

[0137] Secondly, this invention does not directly use the hunter model as the execution strategy, but innovatively uses it as a reference mechanism for global optimization. By dynamically adjusting the edge costs in the minimum-cost maximum flow model, local preferences, competitive relationships, and risk perception are embedded in a unified flow network structure. This design not only avoids the additional energy consumption caused by drones returning empty, but also significantly improves the overall efficiency and fairness of resource allocation.

[0138] Furthermore, by using the multi-objective minimum cost maximum flow solution as the initial population for the improved NSGA-II, this invention effectively reduces the search space and accelerates the convergence speed of the multi-objective evolutionary process. After introducing risk-guided mutation, adaptive crossover, and an improved congestion ranking mechanism, the resulting non-dominated solution set exhibits superior equilibrium characteristics in terms of energy consumption, time, fairness, and security, thus providing emergency management departments with a set of dynamically switchable and robust optimal collaborative transportation strategies. Attached Figure Description

[0139] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0140] Figure 1 This is a flowchart of the method disclosed in this invention.

[0141] Figure 2 It is a hunter-prey preference weight matrix diagram.

[0142] Figure 3 This is a flowchart of the MCMF emergency supplies dispatch network.

[0143] Figure 4 It is a three-dimensional Pareto frontier diagram of emergency material dispatch.

[0144] Figure 5 This is a comparison chart of the convergence curves of the MCMF initial solution versus the randomly initialized NSGA-II.

[0145] Figure 6 This is a comparison of the initial solution of MCMF versus the three-dimensional Pareto front of NSGA-II with random initialization.

[0146] Figure 7 This is a comparison chart of the convergence speeds of the MCMF initial solution versus the randomly initialized NSGA-II solution.

[0147] Figure 8 This is a comparison (two-dimensional projection) of the distribution of the initial solution of MCMF versus the solution of randomly initialized NSGA-II. Detailed Implementation

[0148] This invention provides a dynamic decision-making method for multi-UAV cooperative transportation in emergency environments, comprising the following steps:

[0149] Step 1: Acquisition and processing of material status information at disaster-stricken nodes;

[0150] A hazardous chemical leak occurred in a city, causing the contaminated area to spread in a strip-like pattern. The emergency response entered a continuous material support phase. Within 30 minutes of the incident, stable supply links were established with external emergency warehouses, a rear airport, and multiple drone routes, allowing emergency supplies to be delivered to the affected area in a quasi-continuous flow. Following the emergency, the dispatch system was triggered and entered emergency operation mode, operating within a pre-set dispatch cycle. Within the disaster area, information on the status of supplies at multiple nodes is obtained.

[0151] Step 1-1: Determine the set of disaster-affected nodes;

[0152] In this embodiment, the system determines the set of key nodes within the accident-affected area as follows: This includes: external material supply access nodes. Regional relay node Key demand nodes .

[0153] Steps 1-2: Report the status information of node materials;

[0154] At the start of the current scheduling cycle, each node reports its material status information to the scheduling system. The material status information includes at least: the node's real-time material demand. ; Population scale covered by nodes Node risk level The maximum material retention ratio of the node in the current cycle. The material requirements and risk levels for relay nodes and key demand nodes are shown in the implementation plan below. External supply access nodes report their maximum injection capacity within the current cycle. Some node parameters are shown in Tables 1 and 2 below.

[0155] Table 1 Relay Node Parameter Table

[0156]

[0157] Table 2 Terminal Node Parameter Table

[0158]

[0159] Steps 1-3: Status information organization and output;

[0160] The scheduling system aggregates the material status information reported by each node to form a set of node status vectors for the current scheduling cycle. This serves as the input data basis for subsequent hunter-prey local decision-making models and global optimization models. Relay nodes may either intercept materials due to their own needs or continue to transfer remaining materials.

[0161] Step 2: Construction of a local decision-making model for hunters and prey based on node states

[0162] After acquiring the status information of node resources, the scheduling system constructs a local decision-making model of hunters and prey based on the information, with the disaster-stricken node as the main body, to characterize the priority relationship and competitive behavior between nodes.

[0163] Step 2-1, Determining the hunter node and the prey node;

[0164] In this embodiment, a node that meets the following conditions is considered a hunter node: That is, all nodes with unmet resource needs participate in the local decision-making process as hunter nodes.

[0165] Prey nodes are not a fixed set of nodes, but rather refer to nodes that, within the current scheduling cycle, have the ability to access external resources or have surplus transshipment capacity in the network.

[0166] Step 2-2, Calculation of hunter priority index;

[0167] To characterize the urgency of resource acquisition for different hunter nodes, the system calculates the hunting priority index of each node based on the information reported by the nodes: ;

[0168] in:

[0169] This indicates the per capita shortage of supplies at each node;

[0170] Indicates the risk level of the environment in which the node is located;

[0171] This represents the competition modifier derived from the competition between hunters and prey.

[0172] This priority index is used to reflect the relative importance of nodes in the resource competition process. Some calculation results are shown in Table 3.

[0173] Table 3 Calculation Results

[0174]

[0175] This priority does not directly generate transportation decisions, but is used to adjust the "withholding costs" in subsequent MCMFs.

[0176] Steps 2-3: Output the local decision-making results for both the hunter and the prey;

[0177] It should be noted that in this embodiment, the hunter and prey models do not directly generate the final transportation route or scheduling scheme; their output is the hunting priority index of each node. The relative competitive relationship and preemption intensity between nodes.

[0178] The above results are used as input parameters to adjust the cost and constraint settings in the subsequent multi-objective minimum-cost maximum flow model. The resulting hunter-prey preference weight matrix is ​​as follows: Figure 2 As shown.

[0179] Step 3: Solve the multi-objective minimum cost maximum flow problem based on the hunter and prey results;

[0180] After obtaining the local decision results of the hunter and the prey, the scheduling system constructs and solves a multi-objective minimum cost maximum flow model to obtain candidate transportation schemes that satisfy capacity constraints and flow conservation constraints.

[0181] Step 3-1: Construction of the interceptable flow network structure;

[0182] To depict the actual behavior of nodes in partially retaining and partially transferring resources after obtaining them, each node... Perform node-by-node modeling to construct the following structure:

[0183] ;

[0184] in:

[0185] Edge capacity is set as follows:

[0186] ;

[0187] Edge costs are set as follows:

[0188] ;

[0189] This edge represents the behavior of a node intercepting the flow of resources passing through it, and its cost is adjusted by the hunting priority index.

[0190] Step 3-2, setting flow conservation and interception constraints;

[0191] In the minimum cost maximum flow model, the following constraints are imposed on each node:

[0192] ;

[0193] And it satisfies:

[0194] ;

[0195] in:

[0196] This indicates the actual flow of materials intercepted by the node;

[0197] This indicates the flow of goods that continue to be transferred downstream.

[0198] By applying the above constraints, we can achieve the modeling goal of allowing nodes to be retained without violating the overall flow conservation.

[0199] Step 3-3: Solve for candidate transportation schemes;

[0200] Under the above network structure and parameter settings, the scheduling system aims to minimize the overall system cost and maximize the guaranteed flow, and solves the multi-objective minimum cost maximum flow problem to obtain candidate transportation schemes within the current scheduling cycle.

[0201] In this embodiment, the solution results include: actual external injection flow; material retention at each node; and material transfer flow distribution between nodes. This candidate transportation scheme serves as an execution reference for the current scheduling cycle and can be used as an initial solution for subsequent multi-objective optimization or rolling replanning.

[0202] Within the current scheduling period, the MCMF solution results are as follows: The actual node interception and typical node guarantee rates are shown in Tables 4 and 5.

[0203] External actual injection volume ;

[0204] Table 4 Actual Retention Quantity at Each Node

[0205]

[0206] Table 5 Typical Node Guarantee Rate Table

[0207]

[0208] Scheduling cycle: ΔT = 20 min; MCMF emergency material dispatching network flow diagram as follows: Figure 3 As shown.

[0209] Step 4: Multi-objective optimization based on candidate transportation schemes;

[0210] After solving the multi-objective minimum cost maximum flow model described in step 3, the scheduling system obtains a set of candidate transportation schemes that satisfy capacity constraints and flow conservation constraints. To further optimize the balance between multiple objectives such as energy consumption, response time, fairness, and risk, this embodiment introduces a multi-objective optimization process based on the non-dominated sorting genetic algorithm NSGA-II to perform secondary optimization on the candidate transportation schemes.

[0211] Step 4-1: Candidate scheme encoding and initial population construction;

[0212] The scheduling system uniformly encodes the candidate transportation schemes obtained in step 3. Each encoded individual corresponds to a complete UAV cooperative transportation decision scheme, which includes at least the following decision variables:

[0213] The distribution relationship of material transfer flow between each node;

[0214] The actual amount of materials retained at each node;

[0215] The execution path and task allocation sequence of the drones at each transport edge.

[0216] In this embodiment, the feasible solution set of multi-objective minimum cost maximum flow obtained in step (3) is directly used as the initial population of the NSGA-II algorithm to avoid the problem of a large number of infeasible solutions caused by random initialization, thereby improving the convergence efficiency and engineering feasibility of the algorithm.

[0217] Step 4-2, Construction of the multi-objective optimization function;

[0218] To address the collaborative transportation needs in scenarios requiring continuous emergency supplies support, the following multi-objective optimization function set is constructed:

[0219] Energy consumption objective function; used to minimize the overall flight energy consumption of the UAV in the current scheduling cycle, its value is determined by the flight distance and transport flow on each transport edge;

[0220] The response time objective function is used to minimize the maximum time for key demand nodes to obtain the first batch of supplies or reach the guarantee threshold, reflecting the efficiency of emergency response.

[0221] Fairness objective function; used to measure the balance of material support among different demand nodes, reflecting the system's ability to coordinate support for multiple nodes under continuous supply conditions;

[0222] Risk objective function; used to minimize the cumulative exposure of drones in high-risk areas or along high-risk routes, reflecting transportation safety.

[0223] The aforementioned objectives conflict with each other, and are addressed in a unified manner through multi-objective optimization.

[0224] Step 4-3, Non-dominated sorting and crowding calculation;

[0225] The scheduling system uses the NSGA-II algorithm to perform fast non-dominated sorting of individuals in the current population, dividing the population into several non-dominated layers.

[0226] For individuals within the same non-dominated layer, their crowding distance is further calculated to characterize the sparsity of their distribution in the target space. Individuals with higher crowding have higher priority in maintaining solution set diversity.

[0227] Step 4-4, genetic operator operations and offspring generation;

[0228] After completing the non-dominated sorting and crowding calculation, the scheduling system sequentially performs genetic operations to generate the offspring population, including:

[0229] Cross-operation: Cross-recombining the coding segments involving material allocation and path selection in the candidate schemes to generate new collaborative schemes while ensuring that flow conservation and capacity constraints are not violated;

[0230] Mutation operation: Make slight perturbations to the material interception ratio, transfer flow allocation, or drone task allocation order of some nodes to enhance the search capability of the algorithm.

[0231] All the above genetic operations are based on the network constraints constructed in step (3) to ensure that the generated offspring individuals always meet the physical and scheduling feasibility requirements.

[0232] Steps 4-5: Elite retention and population renewal;

[0233] After merging the parent and offspring populations, the scheduling system performs non-dominated sorting again and selects a preset number of individuals to form a new generation population, based on the principle of prioritizing non-dominated hierarchy and then crowding distance.

[0234] The elite retention mechanism ensures that high-quality non-dominated solutions already obtained are not lost during the iteration process.

[0235] Steps 4-6: Iteration termination condition and optimal strategy output;

[0236] The NSGA-II algorithm iteration terminates when any of the following conditions are met:

[0237] Reach the preset maximum number of iterations;

[0238] The algorithm converges when the variation of the non-dominated solution set is less than the threshold for several consecutive generations.

[0239] After termination, the final non-dominated solution set is output as the multi-objective optimization cooperative transportation strategy cluster for the current scheduling cycle. The scheduling system can select one or more schemes from this strategy cluster for execution based on actual emergency management preferences, or use them as the initial input for rolling optimization in the next scheduling cycle. The resulting three-dimensional Pareto front diagram of emergency material scheduling is shown below. Figure 4 As shown in the figure. Compared with the traditional NSGA-II algorithm, the MCMF initial solution accelerates convergence by 53 generations. This innovative method is suitable for rapid response in emergency dispatching. The performance comparison graph of MCMF initial solution vs. randomly initialized NSGA-II is shown in the figure. Figure 5 , Figure 6 , Figure 7 and Figure 8 As shown.

[0240] This invention provides a dynamic decision-making method and system for multi-UAV collaborative transportation in emergency environments. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A dynamic decision-making method for multi-UAV collaborative transportation in emergency environments, characterized in that, Includes the following steps: Step 1: After an emergency occurs, obtain the material status information reported by the nodes in the disaster area. The material status information includes the node's material inventory, demand scale, urgency level, and risk level. Step 2: Based on the material status information, construct a local decision-making model of hunters and prey with disaster-stricken nodes as the main body. Each node with material demand is a hunter node, and the suitability of potential material supply nodes as prey nodes is independently evaluated, forming a local priority and competition relationship between nodes. Step 3: Based on the local decision-making results of the hunter and prey, establish a multi-objective minimum cost maximum flow (MCMF) model, dynamically adjust the cost parameters of the corresponding nodes and edges in the MCMF model, construct a transportation network that satisfies the capacity constraint and flow conservation constraint, and solve for candidate transportation schemes. Step 4: Using candidate transportation schemes as initial population input, the improved non-dominated sorting genetic algorithm NSGA-II is used to perform non-dominated optimization on energy consumption, time, fairness and risk, generating an optimal cooperative transportation strategy cluster that approximates the Pareto front.

2. The method according to claim 1, characterized in that, Step 1 includes the following steps: Step 1-1: Emergency Triggering and Dispatch System Initialization; Following a public emergency, the dispatch system is automatically triggered, entering the emergency dispatch initialization state, determining the currently affected area, and generating a set of affected nodes. , where each node Each disaster-stricken area, temporary resettlement site, or emergency material support node serves as the basic decision-making unit for subsequent coordinated dispatch. Step 1-2: Multi-node state awareness and asynchronous information reporting: Each node At any moment Report the actual material status information vector to the scheduling system : ， in, Represents a node At any moment The total amount of emergency supplies currently available; Represents a node The size of the population currently served or covered; Represents a node The indicators of material supply recovery capacity or external supply accessibility; Represents a node The credibility or reliability weight of the information reported; Steps 1-3: Information credibility weighting and anomaly suppression processing; After receiving the state information vectors from each node, the scheduling system performs weighted processing on the data reported by the nodes to form a reliable state estimate. ， ； in, Represents a node At any moment Weighted credible inventory of materials; Represents a node At any moment The weighted trusted service population size; Steps 1-4: Calculate the supply and demand level indicators of materials at each node; Based on the weighted state information, the scheduling system calculates the time of each node. per capita material quantity index : ； Steps 1-5: Global statistics update and dynamic threshold preparation; The scheduling system calculates statistical reference values ​​at the system level based on the average material quantity per person across all nodes. ， ， ， in This represents the average amount of supplies per person across all nodes. Indicates at time The maximum value of the per capita material quantity index for all disaster-affected nodes; Indicates at time The minimum value of per capita material quantity for all affected nodes.

3. The method according to claim 2, characterized in that, Step 2 includes: Step 2-1: The system at time... Calculate the global dynamic threshold : ， in It is the standard deviation of per capita material quantity; These are adjustable adaptive weight parameters; Step 2-2: Establish prey node determination rules; If node satisfy: ， Then the node Nodes identified as prey are denoted as follows: ;in Indicates time The set of prey nodes; For any node where there is a shortage of supplies When satisfied At that time, node Considered a hunter node, its goal is to acquire prey nodes. The assessment of potential support recipients is ongoing. Steps 2-3: Modeling the competition constraints among hunters; Introduce normalized weights: ， in Indicates at time Hunter node For prey nodes Normalized hunting preference weights; Introducing a contention load function for prey nodes : ， in Represents the set of all hunter nodes; Hunter Node For all prey nodes Calculate the local hunting utility value independently: ， in, The function representing the resource redundancy of a prey node; Represents the hunter node With prey node The flight distance or time cost between; This indicates the risk inherent in the prey node j itself; These are the weighting coefficients; when When the load exceeds the prey node's capacity, the corresponding utility value will be dynamically penalized: ， in Indicates at time Hunter node From the prey node The initial local revenue value of acquiring resources; Indicates at time Hunter node From the prey node The actual profit value of acquiring resources, adjusted for competition penalties; It is a coefficient greater than 0; Simultaneously, a competition penalty item is introduced. : ， in Indicates at time The set of all nodes identified as hunters; Let be the decision variable, representing the hunter. Should you choose your prey? ,if A value of 1 indicates selection, otherwise it indicates no selection; For indicator functions; Steps 2-4: Establish a local decision-making model for the hunter node and construct the reward functions for the hunter and prey; Hunter Node Targeting the prey node Constructing a local reward function : ， in It is a node With prey node Spatial distance or flight time between them; It refers to the energy consumption or risk cost per unit of material transportation; These are weighting coefficients; Each hunter node Independently solve for local optimal prey: ； in Indicates at time Hunter node The optimal prey node number is selected after independent evaluation based on the local reward function; Employing an active prey transport mechanism: The hunter node only expresses needs and evaluates preferences; Actual transportation is determined by the prey node. Initiate a proactive attack on the hunter node.

4. The method according to claim 3, characterized in that, In step 3, when constructing the minimum cost maximum flow (MCMF) model: Prey node With the Hunter Node The potential transport relationships between them are mapped as directed edges. Unit cost of the side Defined as: ， in Basic transportation costs; This is the distance cost weighting coefficient; The weighting factor is used to calculate the partial returns.

5. The method according to claim 4, characterized in that, In step 3, a multi-objective minimum cost maximum flow network is constructed. It includes the following structures: Node set Including super source nodes Prey node set Hunter node set and Super Hub Node ; in For at any time The first in the prey node set The last node, i.e., the last prey node; For at any time The first in the prey node set The last node, i.e., the last hunter node; Directed edge set ,include: Source to the edge of the prey ,capacity ,cost ;in Represents the prey node At any moment The amount of surplus materials that can be used for external allocation; The prey approached the hunter. ,capacity ,cost Weighted sum of multiple factors: ， in, For preference weights and competition intensity The dynamic adjustment terms constituted; ; It is the cost of transportation time; It is a path risk factor; This is the distance cost weighting coefficient; This is the time cost weighting coefficient; This refers to the risk cost weighting coefficient. These are the weighting coefficients for the local decision adjustment term; In the hunter local model, each hunter node Form a non-binary preference vector for all prey nodes. : ， in For at any time Hunter node For the prey node set Prey node Normalized preference weights; , Represents the prey node For the hunter node The degree of relative compatibility; Hunters to the edge of the river ,capacity ,cost ; The multi-objective minimum cost maximum flow (MCMF) model satisfies the following flow conservation constraints: Source node constraints: ， That is, from the super source node The total amount of all outflowing resources cannot exceed the sum of the total amount of resources currently available for allocation across all hunting nodes; Prey node constraints: ， That is, from any prey node The total amount of materials transported out cannot exceed the current available resources of the node itself; Hunter node constraints: ， That is, any hunter node The total amount of all materials received is equal to the amount that will eventually be consumed or retained, and the total amount cannot exceed the demand gap of the node. Based on this, solve the following minimum-cost maximum flow problem to obtain candidate transportation solutions: , ， in, This represents the maximum emergency response capacity achievable at the current moment.

6. The method according to claim 5, characterized in that, In step 4, the improved non-dominated sorting genetic algorithm NSGA-II includes the following steps: Step 4-1: Encode the candidate transportation schemes into population individuals. Each individual represents a complete UAV cooperative transportation decision scheme, including the following set of decision variables: Actual resource flow from prey node to hunter node ; Corresponding transportation route selection variables ; Service priority or scheduling timing parameters of hunter nodes ; Adjusting weights based on path risk or congestion ; Step 4-2: To address the multi-dimensional decision-making requirements of the collaborative transportation of emergency supplies, the following multi-objective optimization function set is constructed: ， Among them, the energy consumption objective function is... for: ， This indicates the overall energy consumption cost of the drone during the transportation mission; Time objective function for: This is used to minimize the arrival time of materials at the node with the highest demand, reflecting the efficiency of emergency response; Fairness objective function for: ， in To determine the actual amount of supplies obtained, For demand; Risk objective function for: ， in The path risk coefficient reflects the uncertainty of the flight environment and safety risks. Step 4-3: Use the fast non-dominated sorting method to divide the population into multiple non-dominated layers: ； Any two individuals and If the following conditions are met: ， Then it is called Dominate ; Prioritize retaining low-level non-dominated solutions to ensure that the solution set approximates the Pareto front; Step 4-4: Execute genetic operators to generate offspring population: Adaptive crossover operation: The crossover probability is dynamically adjusted with the number of iterations. ， in This represents the current iteration number. This represents the maximum number of iterations. Indicates the first In the next iteration, the algorithm uses an adaptive crossover probability; This is the preset maximum crossover probability; This is the preset minimum crossover probability; Risk-guided mutation manipulation: For high-risk individuals, enhance mutation probability. ， in This represents the current average risk value of the population. This is the adjustment coefficient; For individuals Risk-adjusted mutation probability; The baseline probability of variation; Represents an individual The risk objective function value; Steps 4-5: Merge parent and offspring populations using an elite retention strategy, and select individuals to form a new generation population based on non-dominated hierarchy and crowding distance; use rapid non-dominated sorting to divide the population into two or more non-dominated hierarchies, and calculate the improved crowding distance for individuals within the same non-dominated hierarchy. ; Steps 4-6: Terminate the algorithm when the maximum number of iterations is reached or the non-dominated front converges for multiple consecutive generations, and output the final non-dominated solution set as the optimal cooperative transportation strategy cluster.

7. The method according to claim 6, characterized in that, In steps 4-5, the improved congestion distance is calculated using the following formula. : ， in is the weighting coefficient; M is the total number of optimization objectives; This indicates that in the current non-dominated layer, the first... The maximum value of each objective function; This indicates that in the current non-dominated layer, the first... The minimum value of an objective function; Indicates that in the order of the first After sorting the objective function values ​​in ascending order, the individual The next adjacent individual's first One objective function value; Indicates that in the order of the first After sorting the objective function values ​​in ascending order, the individual The next adjacent individual's first The objective function value.

8. A multi-UAV cooperative transportation dynamic decision-making system for implementing the method as described in any one of claims 1 to 7 in an emergency environment, characterized in that, include: The information sensing module is used to acquire and process the material status information reported by each node; The local decision-making module executes the hunter-prey local decision-making model and outputs node preferences and competition relationships; the global optimization module constructs and solves the multi-objective minimum cost maximum flow model and generates candidate transportation schemes; the multi-objective evolution module executes the improved non-dominated sorting genetic algorithm to optimize the candidate schemes in multiple objectives and outputs a Pareto optimal policy cluster.

9. An electronic device comprising a processor and a memory, the memory storing program code, characterized in that, When the program code is executed by the processor, the method as described in any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that, It stores a computer program or instructions that, when run on a computer, perform the steps of the method as described in any one of claims 1 to 7.

Images