Forest fire air-ground resource scheduling method based on differential evolution and adaptive auction

Abstract

The application discloses a kind of forest fire air-ground resource scheduling method based on differential evolution and adaptive auction, based on attribute graph and influence net and introduce dynamic fire spread mechanism, construct the multi-resource collaborative task graph model of task urgency and target value that can be dynamically updated, to effectively reflect the spatio-temporal coupling relationship between tasks, provide accurate semantic basis for joint optimization;Design a hybrid double-layer structure coding method suitable for multi-objective differential evolution and auction mechanism joint optimization, for simultaneously expressing task selection information and resource allocation indication in scheduling solution;Construct the joint optimization algorithm of fusion multi-strategy differential evolution and adaptive auction, generate the scheduling scheme with fire situation suppression ability and actual scheduling implementability.The application can provide efficient and stable collaborative scheduling scheme for air and ground heterogeneous resources within a limited time, realize the strategy support for emergency decision in dynamic forest fire rescue scene, improve resource utilization and shorten response delay.

Description

Technical Field

[0001] This invention belongs to the field of aviation emergency rescue, specifically involving a method for scheduling air and ground resources for forest fires based on differential evolution and adaptive auction. Background Technology

[0002] With global warming, forest fires are becoming increasingly frequent and severe, further impacting the global climate and posing immeasurable dangers. When forest fires occur, ground rescue personnel often cannot accurately assess the fire's spread in real time, necessitating aerial firefighting capabilities such as drones and helicopters for reconnaissance and monitoring. Therefore, there is an urgent need to study the coordinated air-ground coordination of forest fire response, providing a theoretical basis for timely fire control and scientifically sound firefighting efforts, and offering scientific advice and precise decision-making for sudden forest fires.

[0003] Most current scheduling methods employ a step-by-step solution-building approach during execution, which can easily get trapped in local optima and neglect the potential for collaborative scheduling among global resources. Furthermore, forest fires are highly dynamic and uncertain, with sudden changes in fire spread paths and wind direction, resource depletion, and communication interruptions causing scheduling plans to quickly become invalid. However, traditional methods often provide one-time static solutions, lacking rolling update mechanisms and making it difficult to adjust existing scheduling results in real time to adapt to environmental changes. While some research has alleviated the complexity of forest fire resource scheduling to some extent, significant shortcomings remain in task-resource joint optimization, efficient matching of heterogeneous resources, and real-time scheduling. For example, existing algorithms often separate task selection from resource allocation, lacking global collaborative optimization strategies or dynamic feedback mechanisms during resource game processes, making it difficult to adapt to the rapid evolution of fire situations and the practical needs of multi-source resource coordination. Therefore, there is an urgent need to introduce a scheduling optimization method with global optimization capabilities, dynamic response mechanisms, air-ground collaborative modeling capabilities, and scalability to large-scale task environments. Summary of the Invention

[0004] Purpose of the invention: This invention provides a forest fire air and land resource scheduling method based on differential evolution and adaptive auction, which significantly improves the efficiency of rescue mission completion and resource utilization, enables rapid response and dynamic optimization scheduling in large-scale forest fire scenarios, thereby enhancing overall fire fighting efficiency and providing intelligent and reliable technical support for forest fire emergency command.

[0005] Technical solution: The present invention provides a forest fire air and land resource scheduling method based on differential evolution and adaptive auction, comprising the following steps:

[0006] Based on attribute graphs and influence networks, and incorporating a dynamic fire spread mechanism, a multi-resource collaborative task graph model is constructed that can dynamically update task urgency and target value to reflect the spatiotemporal coupling relationship between tasks and provide a precise semantic basis for joint optimization. The multi-resource collaborative task graph modeling method involves: combining attribute graphs and influence networks to form an operational task graph model that supports "probabilistic logic" and "resource-task mapping"; introducing a dynamic fire spread mechanism based on a forest fire spread model to achieve real-time mapping of the dynamic evolution of the fire situation; and performing resource constraint judgments and designing objective functions to achieve joint optimization of task selection and resource allocation.

[0007] Design a hybrid two-layer structure encoding method suitable for joint optimization of multi-objective differential evolution and auction mechanism, which can simultaneously express task selection information and resource allocation indication in scheduling solution;

[0008] Based on a multi-strategy adaptive differential evolution algorithm, a global optimal search is achieved for the task set, and the subset of tasks with the highest fire-fighting efficiency at the current moment is output, thereby maximizing fire control. By using an improved auction algorithm based on a centralized architecture, the optimal allocation scheme of tasks and resources is found, thereby maximizing resource matching.

[0009] Furthermore, the task graph model includes the construction of the task graph and model variables and constraints;

[0010] The task graph, which is a set of tasks and their relationships depicted in the form of an attribute graph, is represented as follows:

[0011]

[0012] in, For the set of task nodes, For a set of logical effect nodes, Let be a set of directed edges. This is the set of currently available resources;

[0013] Job task nodes Describe using the following tuples:

[0014]

[0015] in, For node name, This specifies the resource type for performing the task. For the target list, List of required resources for the task. The operational resources that may be consumed during task execution and their quantity. Task priorities are updated in real time, taking into account urgency and risk-reward ratio. The basic probability of execution for this node;

[0016] Logical effect node Describe using the following tuples:

[0017]

[0018] in, The name of the logical effect node. Used to describe the type of logical effect node The basic probability of a node achieving this effect, List of required resources for the task. This represents the urgency of the task; the higher the value, the higher the priority for triggering it. The expected benefits of fire suppression or resource protection upon completion of the mission;

[0019] Directed edge Describe using the following tuples:

[0020]

[0021] in, Let be the name of the directed edge. As the starting node, For the termination node, This represents the degree to which a child node promotes its parent node's logical effect when the child node is executed. , This represents the degree to which a child node promotes the execution of its parent node's logic when the child node does not execute. ;

[0022] resource Describe using the following tuples:

[0023]

[0024] in, For resource types, , For resource availability, These are the resource's speed, range, and available time.

[0025] The specific model variables and constraints are as follows:

[0026] Model variables refer to whether the child nodes of each selection logic effect node are executed. Let there be a total of Each selection logic effect node ,node have If there are 1 child node, then the solution variables and constraints are expressed as follows:

[0027]

[0028] in, ,like Then the first The first selection logic effect node Each child node executes. Then it will not be executed;

[0029] Representing resources Should the task be executed? ,when A time indicates execution, otherwise no execution. If the total number of resources is... The total number of tasks is The task allocation constraints are then expressed as follows:

[0030] .

[0031] Furthermore, the forest fire spread model approximates the burning area as a region with a radius of [missing information]. A circle, and in a unit of time Inner radius increased The formula for the spread of the fire point area is expressed as:

[0032]

[0033] in, unit of time The area affected by the spread of the internal fire. Perimeter;

[0034] Firefighting resources in a unit of time The area of ​​the region where the fire was extinguished is expressed as: The calculation formula is as follows:

[0035]

[0036] in, The fire-fighting capacity of resources, that is, the fire-fighting capacity per unit time. The area of ​​the region where internal energy extinguishes the fire;

[0037] The change in the area of ​​the combustion zone Represented as:

[0038]

[0039] Therefore, the rate of change of the combustion zone area can be expressed as:

[0040]

[0041] in, The speed of fire spread is expressed as:

[0042]

[0043] in, The basic propagation speed is determined by the type of fuel. This is an environmental regulation function, defined as follows:

[0044]

[0045] in, For wind speed, This is the wind speed adjustment coefficient. For slope, This is the slope adjustment coefficient. Humidity ratio Humidity suppression coefficient;

[0046] make If is a constant, then we get:

[0047]

[0048] Assume there is a total One fire point and Each fire extinguishing resource is allocated to the fire point. The resource set is Then its total fire extinguishing capacity is expressed as:

[0049]

[0050] in, Representing resources Firefighting capabilities;

[0051] In summary, the wildfire spread model can be represented as follows:

[0052] .

[0053] Furthermore, the specific process for determining the constraints of the aforementioned resources is as follows:

[0054] S1. Input the task graph obtained from the attribute graph modeling. Available resources include air and ground firefighting and rescue equipment;

[0055] S2. Starting from the root node of the task graph, use a depth-first traversal method to visit each logical effect node and task node layer by layer.

[0056] S3. If it is a sequential logic effect node, check the resource requirements of its child node tasks one by one. If they are less than or equal to the available resources, execute the task and release the remaining resources; otherwise, determine that the resources are insufficient and terminate the judgment process, and return "False".

[0057] S4. If it is a parallel logic effect node, summarize the total demand of each child node. If it is less than or equal to the available resources, execute each subtask in parallel and release the remaining resources; otherwise, return "False".

[0058] S5. If the selected logical effect node is selected, first determine whether to execute it. If the resource requirements meet the constraints, execute and release the resources; otherwise, return "False".

[0059] S6. If it is a job task node, if the demand does not exceed the available resources, execute and release the resources; otherwise, return "False".

[0060] S7. If all nodes satisfy the constraints, output the judgment result "True"; if any node has insufficient resources, output "False".

[0061] Furthermore, the objective function is designed as follows:

[0062] The quality of a forest fire response mission is measured by its ability to suppress fire spread, urgency of response, and spatial accessibility. The objective function for maximizing fire source suppression benefits is expressed as:

[0063]

[0064] in, Choose a vector for the task, where 1 represents the task. Selected, 0 indicates not selected. For the task The benefit function is expressed as:

[0065]

[0066] in, For the task The area of ​​the fire source in the fire zone represents the potential controllable fire area. For current resources to tasks Shortest distance, Based on the urgency of the task, These are normalized weight coefficients, and satisfy... ;

[0067] The optimization objective is to maximize the total fire source suppression benefit of the task set, but since differential evolution aims to minimize it, a negative sign is used to transform it into: Even if the task selection is reasonable, scheduling will fail if resources are allocated improperly. It is necessary to optimize resource suitability for tasks, execution costs, and allocation conflicts. The objective function for minimizing resource matching costs is expressed as:

[0068]

[0069] in, Assign a resource vector to represent a task. The assigned resource number, For resources Execute the task The allocation cost is expressed as:

[0070]

[0071] in, For resources To the mission distance, For the task With resources Adaptability difference, For the same time resources The penalty value for being assigned to multiple tasks. Assign a cost weighting factor;

[0072] The optimization objective is to minimize the total execution cost of the task-resource allocation combination, expressed as: .

[0073] Furthermore, the hybrid two-layer structure encoding method specifically includes:

[0074] The basic search unit in each policy-adaptive differential evolution algorithm consists of two parts: the task selection part controls which tasks will enter the current scheduling round, encoded as a sequence of length [length missing]. The bit string is suitable for binary differential evolution operations; the initial encoding of the resource indication part is generated through a proximity-first, random allocation rule, and this part is used as the initial candidate resource pool during the resource bidding stage to participate in task bidding and assignment; specifically:

[0075]

[0076] in, The number of candidate tasks. To select variables, representing the first... Whether a task is selected or not determines the subset of tasks that will participate in subsequent auctions and resource allocation. , The resource allocation hint variable is represented as a task. Preset bid resource number;

[0077] This code uses a binary task selection structure, supports standard binary operations in differential evolution, and uses resource indicator variables to assist the auction, providing the initial bidding range for the auction module and enhancing the rationality of resource allocation. In addition, this code supports incomplete task allocation, and tasks that are not selected can be skipped naturally, adapting to dynamically changing task numbers and forest fire environments.

[0078] The scheduling area is divided into multiple task grids, each grid representing a candidate firefighting task point. The resource number in the selected task grid indicates the scheduling resource recommended for that task.

[0079] Furthermore, the multi-strategy adaptive differential evolution algorithm is as follows:

[0080] The above model variables and constraints are transformed as follows:

[0081]

[0082] in, That is, the first Each logical effect node should execute the number of its child nodes.

[0083] Furthermore, the specific solution process of the multi-strategy adaptive differential evolution algorithm is as follows:

[0084] S1. Population Initialization: Considering scheduling feasibility, initialize one elite population and three evolutionary populations. Each individual corresponds to a candidate air-ground cooperative scheduling scheme, and its gene assignment is as follows:

[0085]

[0086] in, For the initial population, the first The first individual The value of the dimensional gene, , , To select the number of logical effect nodes, and It is the first The minimum and maximum values ​​that the dimensional gene can take, where , ;

[0087] S2. Parameter Adaptive Selection: By learning from historical high-quality fire dispatch schemes, the mutation and crossover parameters are adaptively adjusted to enhance the algorithm's global search capability during the rapid fire spread phase and its local fine-tuning capability during the fire control phase. This is expressed as:

[0088]

[0089] in, and The first The mutation scaling factor and crossover probability of each generation. Indicates a normal distribution. This indicates the calculation of the average value. , The mean of a normal distribution is . , Let these represent the mutation scaling factor and the set of crossover probabilities that can generate excellent mutants, respectively. This reflects the role of historical information in updating control parameters;

[0090] The parameters are adjusted using the Sigmoid function. Perform nonlinear dynamic adjustment:

[0091]

[0092] in, For parameters The maximum value, For the maximum number of generations, The formula for calculating the Sigmoid function, representing the degree of evolution, is as follows: ;

[0093] S3. Population Mutation and Detection: Four mutation strategies are adopted, and the "roulette wheel" method is used to prioritize the scheduling scheme with higher fire source suppression efficiency and lower resource matching cost, as shown below:

[0094]

[0095] in, For the first The first generation of the population A variant. , , , and The first The first generation of the population , , , and Individual, The top in the current population For a random outstanding individual, the function sequence Constructing the Fibonacci sequence ;

[0096] Boundary detection of genes in variants is performed using a "boundary absorption" strategy:

[0097] ;

[0098] S4. Population Crossover and Selection: After mixing the genes of mutants and individuals from the original population, experimental subjects are obtained, generating new candidate schemes that fuse different scheduling schemes. The crossover calculation formula is:

[0099]

[0100] By employing a "greedy strategy," selecting superior individuals from the parent population and the experimental subject set to obtain the next generation population, we have:

[0101]

[0102] in, For the first The first generation of the population One test subject, For the first The corresponding first generation in the population Individual, , The fitness solution function is based on the fire source suppression benefits and resource matching costs.

[0103] S5. Population Information Sharing: After each iteration, the top 30% of individuals in fitness are added to the candidate elite set, and the poorly performing individuals in the elite population are replaced. After the iteration ends, the best individual in the elite population is the final air-ground cooperative scheduling solution.

[0104] Furthermore, the improved auction algorithm based on a centralized architecture is specifically as follows:

[0105] Pricing strategies utilizing distance information: when resources For the task When submitting a quote, the lower limit of the price increase. The calculation formula is:

[0106]

[0107] in, For the custom price increase lower limit distance weighting coefficient, For the quantity of resources, For resources With the task The distance between them The range of values ​​is Control parameters;

[0108] An adaptive operator for the number of quotes: using an adaptive operator. To control The value of , The calculation formula is as follows:

[0109]

[0110]

[0111] in, It is a positive constant. Representing resources For the task The number of times a quote is made. and Indicates the maximum and second-largest benefits that can be obtained by performing the task;

[0112] Multiple auctions with dynamically updated task value: The value update formula is expressed as:

[0113]

[0114] in, These are customizable weighting coefficients used to balance historical value with the current environment. For resources Execute the task The success rate For customizable about The weighting coefficients, , For the task The comprehensive objective value is calculated using the following formula:

[0115]

[0116] in, The fire spread area within the mission area. For wind speed, For terrain slope, For humidity, With each indicator weighted and the sum of the weights equal to 1, this update mechanism can dynamically enhance the mission value of areas with severe fires, harsh terrain, high wind speeds, and low humidity, and accurately guide resource responses to high-risk areas.

[0117] Furthermore, the solution process for the improved auction algorithm based on a centralized architecture is as follows:

[0118] Treating resources as bidders and tasks as bidding objects, the average movement distance of each resource is the objective function. The total value of completing the firefighting mission is the objective function. Let the maximum number of resources that can be allocated to each task be . The specific process is as follows:

[0119] S1. Initialize the location of resources and target tasks, available resource set, resource list, resource value, resource weight, and maximum resource capacity;

[0120] S2. Based on the input parameters, generate a random resource bidding sequence as the initial bidding order;

[0121] S3. For each unassigned target task, perform the following operations: When there are unassigned resources, select the target task; if the number of resources allocated to the task has reached a certain threshold... Then choose from the remaining tasks;

[0122] S4. Calculate the price of each resource, and the task will be assigned to the resource with the highest price.

[0123] S5. After each round of bidding, save the current resource allocation plan and update it. , ;

[0124] S6. Perform a non-dominated sort on all the solutions to obtain the optimal Pareto solution set.

[0125] Beneficial Effects: Compared with existing technologies, the beneficial effects of this invention are as follows: This invention can solve the problem of air-ground coordinated resource scheduling in forest fires under dynamic constraints, providing an efficient and stable coordinated scheduling scheme for heterogeneous air and ground resources within a limited time, improving resource utilization and reducing response delay; This invention can effectively improve the intelligence and practicality of scheduling strategies, providing a scalable and robust solution for air-ground coordinated intelligent scheduling in dynamic forest fire rescue scenarios, and providing intelligent support for emergency decision-making by relevant departments. Attached Figure Description

[0126] Figure 1 This is a flowchart of the present invention;

[0127] Figure 2 This is a schematic diagram of a simple task graph model provided by the present invention;

[0128] Figure 3 This is a schematic diagram showing the distribution of aviation, ground fire stations, and mission points in the verification scenario provided by the present invention;

[0129] Figure 4 This is a schematic diagram of simulation results comparing the fitness of the present invention with the number of iterations with other methods;

[0130] Figure 5 This is a schematic diagram of simulation results comparing different resource quantities provided by the present invention with other methods. Detailed Implementation

[0131] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings to further illustrate the technical solutions of the present invention. However, the present invention is not limited to these embodiments.

[0132] This invention provides a forest fire air and ground resource scheduling method based on differential evolution and adaptive auction. The problem to be solved is the coordinated scheduling of air and ground resources for forest fires: heterogeneous air and ground resource clusters depart from aviation and ground fire stations respectively to carry out fire fighting and rescue missions at target fire points, thereby achieving coordinated scheduling of air and ground resources and efficient rescue.

[0133] Define a two-dimensional plane The area is a forest fire zone. Environmental factors considered within the zone include wind speed, terrain slope, relative humidity, and vegetation type. It is assumed that the number of fire points and their specific locations involved in the rescue are known. Each task is associated with a specific fire point area and is assigned corresponding environmental attributes. The initial radius of each fire point is randomly generated in the range of [200, 600] m, and the fire source area expands dynamically in a circular pattern.

[0134] The number of open land resources deployed is Air resources include light, medium, and heavy helicopters and fixed-wing aircraft, while ground resources include tracked fire trucks, water tanker fire trucks, foam fire trucks, and high-pressure spray trucks. The attributes of each air and ground resource are known (including maximum range, fire extinguishing capacity, speed, and maximum endurance).

[0135] like Figure 1 As shown, the specific implementation process is as follows:

[0136] Step 1: Based on attribute graphs and influence networks, and introducing a dynamic fire spread mechanism, construct a multi-resource collaborative task graph model that can dynamically update task urgency and target value to reflect the spatiotemporal coupling relationship between tasks and provide a precise semantic basis for joint optimization. The attribute graph and influence network are combined to form an operational task graph model that supports "probabilistic logic" and "resource-task mapping"; a dynamic fire spread mechanism is introduced based on the forest fire spread model to realize real-time mapping of the dynamic evolution of the fire situation; resource constraint judgments and objective functions are designed separately to achieve joint optimization of task selection and resource allocation.

[0137] Figure 2 A simple task graph is provided, where the model variables are whether the child nodes of each selection logic effect node are executed. Let there be a total of... Each selection logic effect node ,node have If there are 1 child node, then the solution variables and constraints are expressed as follows:

[0138]

[0139] in, ,like Then the first The first selection logic effect node Each child node executes. Then it will not be executed.

[0140] Representing resources Should the task be executed? ,when A time indicates execution, otherwise no execution. If the total number of resources is... The total number of tasks is The task allocation constraints are then expressed as follows:

[0141] .

[0142] The combustion zone is approximated as a region with a radius of... A circle, and in a unit of time Inner radius increased The formula for the spread of the fire point area is expressed as:

[0143]

[0144] in, unit of time The area affected by the spread of the internal fire. Perimeter.

[0145] Firefighting resources in a unit of time The area of ​​the region where the fire was extinguished is expressed as: The calculation formula is as follows:

[0146]

[0147] in, The fire-fighting capacity of resources, that is, the fire-fighting capacity per unit time. The area within which the fire can be extinguished.

[0148] The change in the area of ​​the combustion zone Represented as:

[0149]

[0150] Therefore, the rate of change of the combustion zone area can be expressed as:

[0151]

[0152] in, The speed of fire spread is expressed as:

[0153]

[0154] in, The basic propagation speed is determined by the type of fuel. This is an environmental regulation function, influenced by factors such as wind speed, slope, and humidity, and is defined as follows:

[0155]

[0156] in, For wind speed, This is the wind speed adjustment coefficient. For slope, This is the slope adjustment coefficient. Humidity ratio This is the humidity suppression coefficient.

[0157] make If is a constant, then we can obtain:

[0158] .

[0159] Assume there is a total One fire point and Each fire extinguishing resource is allocated to the fire point. The resource set is Then its total fire extinguishing capacity is expressed as:

[0160]

[0161] in, Representing resources Its fire-fighting capabilities.

[0162] In summary, the wildfire spread model can be represented as:

[0163] .

[0164] The specific process for determining resource constraints is as follows:

[0165] S1. Input the task graph obtained from the attribute graph modeling. Available resources include air and ground firefighting and rescue equipment.

[0166] S2. Starting from the root node of the task graph, use a depth-first traversal to visit each logical effect node and task node layer by layer.

[0167] S3. If it is a sequential logic effect node, check the resource requirements of its child node tasks one by one. If they are less than or equal to the available resources, execute the task and release the remaining resources; otherwise, determine that the resources are insufficient and terminate the judgment process, returning "False".

[0168] S4. If it is a parallel logic effect node, summarize the total demand of each child node. If it is less than or equal to the available resources, execute each subtask in parallel and release the remaining resources; otherwise, return "False".

[0169] S5. If the selected logical effect node is selected, first determine whether to execute it. If the resource requirements meet the constraints, execute and release the resources; otherwise, return "False".

[0170] S6. If it is a job task node, if the demand does not exceed the available resources, execute and release the resources; otherwise, return "False".

[0171] S7. If all nodes satisfy the constraints, output the judgment result "True"; if any node has insufficient resources, output "False".

[0172] The objective function for the multi-resource collaborative task graph model is defined as follows:

[0173] The objective function for maximizing fire source suppression benefits is expressed as:

[0174]

[0175] in, Choose a vector for the task, where 1 represents the task. Selected, 0 indicates not selected. For the task The benefit function is expressed as:

[0176]

[0177] in, For the task The area of ​​the fire source in the fire zone represents the potential controllable fire area. For current resources to tasks Shortest distance, The urgency of the mission is determined by factors such as the rate of fire spread, wind direction, and slope. These are normalized weight coefficients, and satisfy... .

[0178] The optimization objective is to maximize the total fire source suppression benefit of the task set, but since differential evolution aims to minimize it, a negative sign is used to transform it into:

[0179]

[0180] The objective function for minimizing resource matching costs is expressed as:

[0181]

[0182] in, Assign a resource vector to represent a task. The assigned resource number, For resources Execute the task The allocation cost is expressed as:

[0183]

[0184] in, For resources To the mission distance, For the task With resources Poor adaptability, such as mismatch between fire extinguishing rating and load capacity, etc. For the same time resources The penalty value for being assigned to multiple tasks. The weighting coefficient for the allocation cost.

[0185] The optimization objective is to minimize the total execution cost of the task-resource allocation combination, expressed as: .

[0186] Step 2: Design a hybrid two-layer structure encoding method suitable for joint optimization of multi-objective differential evolution and auction mechanism, which can simultaneously express task selection information and resource allocation indication in scheduling solution.

[0187] To design a hybrid two-layer encoding method suitable for joint optimization of multi-objective differential evolution and auction mechanisms, each individual consists of two parts. The task selection part controls which tasks will enter the current scheduling round, and is encoded as a length of... The bit string is suitable for binary differential evolution operations, while the initial encoding of the resource indicator part can be generated through rules such as proximity priority and random allocation. This part is used as the initial candidate resource pool during the resource bidding stage to participate in task bidding and assignment; specifically:

[0188]

[0189] in, The number of candidate tasks. To select variables, representing the first... Whether a task is selected or not determines the subset of tasks that will participate in subsequent auctions and resource allocation. , The resource allocation hint variable is represented as a task. Preset bid resource number;

[0190] The scheduling area is divided into multiple task grids, each grid representing a candidate firefighting task point. The resource number in the selected task grid indicates the scheduling resource recommended for that task.

[0191] Step 3: Based on the aforementioned multi-resource collaborative task graph model and hybrid two-layer structure encoding, a joint optimization algorithm integrating multi-strategy differential evolution and adaptive auction is constructed: First, based on the multi-strategy adaptive differential evolution algorithm, a global optimal search is performed on the task set, outputting the subset of tasks with the highest firefighting efficiency at the current moment, thereby maximizing fire control; then, through an improved auction algorithm based on a centralized architecture, the optimal allocation scheme of tasks and resources is found, thereby maximizing resource matching. The specific process is as follows:

[0192] The above model variables and constraints are transformed as follows:

[0193]

[0194] in, That is, the first Each logical effect node should execute the number of its child nodes.

[0195] The specific solution process of the multi-strategy adaptive differential evolution algorithm is as follows:

[0196] S1. Population Initialization: Considering scheduling feasibility, initialize one elite population and three evolutionary populations. Each individual corresponds to a candidate air-ground cooperative scheduling scheme, and its gene assignment is as follows:

[0197]

[0198] in, For the initial population, the first The first individual The value of the dimensional gene, , , To select the number of logical effect nodes, and It is the first The minimum and maximum values ​​that the dimensional gene can take, among which , .

[0199] S2. Parameter Adaptive Selection: By learning from historical high-quality fire dispatch schemes, the mutation and crossover parameters are adaptively adjusted to enhance the algorithm's global search capability during the rapid fire spread phase and its local fine-tuning capability during the fire control phase. This is expressed as:

[0200]

[0201] in, and The first The mutation scaling factor and crossover probability of each generation. Indicates a normal distribution. This indicates the calculation of the average value. , The mean of a normal distribution is . , Let these represent the mutation scaling factor and the set of crossover probabilities that can generate excellent mutants, respectively. This reflects the role of historical information in updating control parameters.

[0202] The parameters are adjusted using the Sigmoid function. Perform nonlinear dynamic adjustment:

[0203]

[0204] in, For parameters The maximum value, For the maximum number of generations, The formula for calculating the Sigmoid function, representing the degree of evolution, is as follows: .

[0205] S3. Population Mutation and Detection: Four mutation strategies are adopted, and the "roulette wheel" method is used to prioritize the scheduling scheme with higher fire source suppression efficiency and lower resource matching cost, as shown below:

[0206]

[0207] in, For the first The first generation of the population A variant. , , , and The first The first generation of the population , , , and Individual, The top in the current population For a random outstanding individual, the function sequence Constructing the Fibonacci sequence ;

[0208] Boundary detection of genes in variants is performed using a "boundary absorption" strategy:

[0209] .

[0210] S4. Population Crossover and Selection: After mixing the genes of mutants and individuals from the original population, experimental subjects are obtained, generating new candidate schemes that fuse different scheduling schemes. The crossover calculation formula is:

[0211] .

[0212] By employing a "greedy strategy," selecting superior individuals from the parent population and the experimental subject set to obtain the next generation population, we have:

[0213]

[0214] in, For the first The first generation of the population One test subject, For the first The corresponding first generation in the population Individual, , This is a fitness solution function based on fire source suppression benefits and resource matching costs.

[0215] S5. Population Information Sharing: After each iteration, the top 30% of individuals in fitness are added to the candidate elite set, and the poorly performing individuals in the elite population are replaced. After the iteration ends, the best individual in the elite population is the final air-ground cooperative scheduling solution.

[0216] Then, the improved auction algorithm based on the centralized architecture is as follows:

[0217] (1) Pricing strategy using distance information: when resources For the task When submitting a quote, the lower limit of the price increase. The calculation formula is:

[0218]

[0219] in, For the custom price increase lower limit distance weighting coefficient, For the quantity of resources, For resources With the task The distance between them The range of values ​​is The control parameters.

[0220] (2) Adaptive operator for the number of bids: Use an adaptive operator To control The value of , The calculation formula is as follows:

[0221]

[0222]

[0223] in, It is a positive constant. Representing resources For the task The number of times a quote is made. and These represent the maximum and second-highest potential gains from performing the task.

[0224] (3) Multiple auctions with dynamic updates to task value: The value update formula is expressed as:

[0225]

[0226] in, These are customizable weighting coefficients used to balance historical value with the current environment. For resources Execute the task The success rate For customizable about The weighting coefficients, , For the task The comprehensive objective value is determined by factors such as the fire situation, wind speed, slope, and humidity in the area, and its calculation formula is as follows:

[0227]

[0228] in, The fire spread area within the mission area. For wind speed, For terrain slope, For humidity, With each indicator weighted and the sum of the weights equal to 1, this update mechanism can dynamically enhance the mission value of areas with severe fires, harsh terrain, high wind speeds, and low humidity, and accurately guide resource responses to high-risk areas.

[0229] Treating resources as bidders and tasks as bidding objects, the average movement distance of each resource is the objective function. The total value of completing the firefighting mission is the objective function. Let the maximum number of resources that can be allocated to each task be . The specific process of the improved auction algorithm based on a centralized architecture is as follows:

[0230] S1. Initialize the location of resources and target tasks, available resource set, resource list, resource value, resource weight, and maximum resource capacity.

[0231] S2. Based on the input parameters, generate a random resource bidding sequence as the initial bidding order;

[0232] S3. For each unassigned target task, perform the following operations: When there are unassigned resources, select the target task; if the number of resources allocated to the task has reached a certain threshold... Then, choose from the remaining tasks.

[0233] S4. Calculate the price of each resource, and the task will be assigned to the resource with the highest price.

[0234] S5. After each round of bidding, save the current resource allocation plan and update it. , .

[0235] S6. Perform a non-dominated sort on all the solutions to obtain the optimal Pareto solution set.

[0236] Figure 3 This invention provides a specific scenario setting. The region is a representative forest fire-prone area with complex terrain and diverse vegetation types, which can well represent the uncertainties and challenges of real fire rescue. Air and ground fire stations are considered as point 0 and point-1, respectively. Assuming the distribution of each point is known, the distances between points and the environmental conditions within each area are also known. Air resources are set to include 1 light helicopter, 2 medium helicopters, 2 heavy helicopters, and 3 fixed-wing aircraft. Ground resources include 1 tracked fire truck, 1 water tanker fire truck, 1 foam fire truck, and 1 high-pressure spray truck. Their specific performance parameters are shown in Table 1. The initial positions of the air and ground resource clusters are located at point 0 and point-1, respectively.

[0237] Table 1. Specific Performance Parameters of Open Space Resources

[0238]

[0239] Figure 4The simulation results provided by this invention are as follows: With 10 task points and 12 available land resources, the DE algorithm, MHDP algorithm, AA algorithm, TSGAA algorithm, and the method proposed in this invention were independently run 30 times under the same parameter settings and initial environment. The average fitness of each generation was used to plot the convergence curve of fitness as a function of iteration number. The average fitness values ​​of the different algorithms were 0.802 (DE), 0.800 (MHDP), 0.826 (AA), 0.826 (TSGAA), and 0.926 (the method proposed in this invention), respectively. The optimal fitness values ​​were 0.861 (DE), 0.900 (MHDP), 0.850 (AA), 0.878 (TSGAA), and 0.950 (the method proposed in this invention), respectively. The method proposed in this invention showed a faster convergence speed in the initial stage and tended to stabilize after about 60 generations, indicating that the proposed strategy achieved good synergy between global search and local resource matching. This invention demonstrates its strong optimization capabilities and scheduling adaptability, enabling air and ground resources to respond quickly and coordinate efficiently to complete forest fire rescue missions.

[0240] Figure 5 This is a schematic diagram illustrating the simulation results compared with other methods under different resource quantities provided by this invention: When the available resource quantity is set to 12, 24, and 36, the fitness distribution characteristics obtained by using the DE algorithm, MHDP algorithm, AA algorithm, TSGAA algorithm, and the method proposed in this invention are compared. The method proposed in this invention exhibits extremely high consistency in multiple runs across all resource scales: its fitness standard deviation is the smallest, and there are no obvious abnormal fluctuations, indicating that the method proposed in this invention can stably converge to a high-quality solution under different initial population or perturbation conditions. This verifies the synergistic advantages of the population exploration ability of the differential evolution operator and the rapid assignment ability of the auction mechanism in terms of robustness and convergence stability.

[0241] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A forest fire open space resource scheduling method based on differential evolution and adaptive auction, characterized in that, The implementation process is as follows: Based on attribute graphs and influence networks, and incorporating a dynamic fire spread mechanism, a multi-resource collaborative task graph model is constructed that can dynamically update task urgency and target value to reflect the spatiotemporal coupling relationship between tasks, providing a precise semantic basis for joint optimization. The multi-resource collaborative task graph modeling method involves: combining attribute graphs and influence networks to form an operational task graph model that supports "probabilistic logic" and "resource-task mapping"; introducing a dynamic fire spread mechanism based on a forest fire spread model to achieve real-time mapping of the dynamic evolution of the fire situation; and performing resource constraint judgments and designing objective functions to achieve joint optimization of task selection and resource allocation. Design a hybrid two-layer structure encoding method suitable for joint optimization of multi-objective differential evolution and auction mechanism, which can simultaneously express task selection information and resource allocation indication in scheduling solution; Based on a multi-strategy adaptive differential evolution algorithm, a global optimal search is achieved for the task set, and the subset of tasks with the highest fire-fighting efficiency at the current moment is output, thereby maximizing fire control. By using an improved auction algorithm based on a centralized architecture, the optimal allocation scheme of tasks and resources is found, thereby maximizing resource matching.

2. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction as described in claim 1, characterized in that, The task graph model includes the construction of the task graph and model variables and constraints; The task graph, which is a set of tasks and their relationships depicted in the form of an attribute graph, is represented as follows: in, For the set of task nodes, For a set of logical effect nodes, Let be a set of directed edges. This is the set of currently available resources; Job task nodes Describe using the following tuples: in, For node name, This specifies the resource type for performing the task. For the target list, List of required resources for the task. The operational resources that may be consumed during task execution and their quantity. Task priorities are updated in real time, taking into account urgency and risk-reward ratio. The basic probability of execution for this node; Logic effect node Describe using the following tuples: in, The name of the logical effect node. Used to describe the type of logical effect node The basic probability of a node achieving this effect, List of required resources for the task. This represents the urgency of the task; the higher the value, the higher the priority for triggering it. The expected benefits of fire suppression or resource protection upon completion of the mission; Directed edge Describe using the following tuples: in, Let be the name of the directed edge. As the starting node, For the termination node, This represents the degree to which a child node promotes its parent node's logical effect when the child node is executed. , This represents the degree to which a child node promotes the execution of its parent node's logic when the child node does not execute. ; resource Describe using the following tuples: in, For resource types, , For resource availability, These are the resource's speed, range, and available time. The specific model variables and constraints are as follows: Model variables refer to whether the child nodes of each selection logic effect node are executed. Let there be a total of Each selection logic effect node ,node have If there are 1 child node, then the solution variables and constraints are expressed as follows: in, ,like Then the first The first selection logic effect node Each child node executes. Then it will not be executed; Representing resources Should the task be executed? ,when A time indicates execution, otherwise no execution. If the total number of resources is... The total number of tasks is The task allocation constraints are then expressed as follows: 。 3. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction as described in claim 1, characterized in that, The forest fire spread model approximates the burning area as a region with a radius of... A circle, and in a unit of time Inner radius increased The formula for the spread of the fire point area is expressed as: in, unit of time The area affected by the spread of the internal fire. Perimeter; Firefighting resources in a unit of time The area of ​​the region where the fire was extinguished is expressed as: The calculation formula is as follows: in, The fire-fighting capacity of resources, that is, the fire-fighting capacity per unit time. The area of ​​the region where internal energy extinguishes the fire; The change in the area of ​​the combustion zone Represented as: Therefore, the rate of change of the combustion zone area can be expressed as: in, The speed of fire spread is expressed as: in, The basic propagation speed is determined by the type of fuel. This is an environmental regulation function, defined as follows: in, For wind speed, This is the wind speed adjustment coefficient. For slope, This is the slope adjustment coefficient. For humidity ratio, Humidity suppression coefficient; make If is a constant, then we get: Assume there is a total One fire point and Each fire extinguishing resource is allocated to the fire point. The resource set is Then its total fire extinguishing capacity is expressed as: in, Representing resources Firefighting capabilities; In summary, the wildfire spread model can be represented as follows: 。 4. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction as described in claim 1, characterized in that, The specific process for determining the constraints of the resources is as follows: S1. Input the task graph obtained from the attribute graph modeling. Available resources include air and ground firefighting and rescue equipment; S2. Starting from the root node of the task graph, use a depth-first traversal method to visit each logical effect node and task node layer by layer. S3. If it is a sequential logic effect node, check the resource requirements of its child node tasks one by one. If they are less than or equal to the available resources, execute the task and release the remaining resources; otherwise, determine that the resources are insufficient and terminate the judgment process, and return "False". S4. If it is a parallel logic effect node, summarize the total demand of each child node. If it is less than or equal to the available resources, execute each subtask in parallel and release the remaining resources; otherwise, return "False". S5. If the selected logical effect node is selected, first determine whether to execute it. If the resource requirements meet the constraints, execute and release the resources; otherwise, return "False". S6. If it is a job task node, if the demand does not exceed the available resources, execute and release the resources; otherwise, return "False". S7. If all nodes satisfy the constraints, output the judgment result "True"; if any node has insufficient resources, output "False".

5. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction as described in claim 1, characterized in that, The objective function is designed as follows: The quality of a forest fire response mission is measured by its ability to suppress fire spread, urgency of response, and spatial accessibility. The objective function for maximizing fire source suppression benefits is expressed as: in, Choose a vector for the task, where 1 represents the task. Selected, 0 indicates not selected. For the task The benefit function is expressed as: in, For the task The area of ​​the fire source in the fire zone represents the potential controllable fire area. For current resources to tasks Shortest distance, Based on the urgency of the task, These are normalized weight coefficients, and satisfy... ; The optimization objective is to maximize the total fire source suppression benefit of the task set, but since differential evolution aims to minimize it, a negative sign is used to transform it into: Even if the task selection is reasonable, scheduling will fail if resources are allocated improperly. It is necessary to optimize resource suitability for tasks, execution costs, and allocation conflicts. The objective function for minimizing resource matching costs is expressed as: in, Assign a resource vector to represent a task. The assigned resource number, For resources Execute the task The allocation cost is expressed as: in, For resources To the mission distance, For the task With resources Adaptability difference, For the same time resources The penalty value for being assigned to multiple tasks. Assign a cost weighting factor; The optimization objective is to minimize the total execution cost of the task-resource allocation combination, expressed as: .

6. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction as described in claim 1, characterized in that, The hybrid two-layer structure encoding method is specifically as follows: The basic search unit in each policy-adaptive differential evolution algorithm consists of two parts: the task selection part controls which tasks will enter the current scheduling round, encoded as a string of length [length missing]. The bit string is suitable for binary differential evolution operations; the initial encoding of the resource indication part is generated through a proximity-first, random allocation rule, and this part is used as the initial candidate resource pool during the resource bidding stage to participate in task bidding and assignment; specifically: in, The number of candidate tasks. To select variables, representing the first... Whether a task is selected or not determines the subset of tasks that will participate in subsequent auctions and resource allocation. , The resource allocation hint variable is represented as a task. Preset bid resource number; This code uses a binary task selection structure, supports standard binary operations in differential evolution, and uses resource indicator variables to assist the auction, providing the initial bidding range for the auction module and enhancing the rationality of resource allocation. In addition, this code supports incomplete task allocation, and tasks that are not selected can be skipped naturally, adapting to dynamically changing task numbers and forest fire environments. The scheduling area is divided into multiple task grids, each grid representing a candidate firefighting task point. The resource number in the selected task grid indicates the scheduling resource recommended for that task.

7. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction as described in claim 1, characterized in that, The multi-strategy adaptive differential evolution algorithm is as follows: The above model variables and constraints are transformed as follows: in, That is, the first Each logical effect node should execute the number of its child nodes.

8. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction according to claim 1, characterized in that, The specific solution process of the multi-strategy adaptive differential evolution algorithm is as follows: S1. Population Initialization: Considering scheduling feasibility, initialize one elite population and three evolutionary populations. Each individual corresponds to a candidate air-ground cooperative scheduling scheme, and its gene assignment is as follows: in, For the initial population, the first The first individual The value of the dimensional gene, , , To select the number of logical effect nodes, and It is the first The minimum and maximum values ​​that the dimensional gene can take, where , ; S2. Parameter Adaptive Selection: By learning from historical high-quality fire dispatch schemes, the mutation and crossover parameters are adaptively adjusted to enhance the algorithm's global search capability during the rapid fire spread phase and its local fine-tuning capability during the fire control phase. This is expressed as: in, and The first The mutation scaling factor and crossover probability of each generation. Indicates a normal distribution. This indicates the calculation of the average value. , The mean of a normal distribution is . , Let these represent the mutation scaling factor and the set of crossover probabilities that can generate excellent mutants, respectively. This reflects the role of historical information in updating control parameters; The parameters are adjusted using the Sigmoid function. Perform nonlinear dynamic adjustment: in, For parameters The maximum value, For the maximum number of generations, The formula for calculating the Sigmoid function, representing the degree of evolution, is as follows: ; S3. Population Mutation and Detection: Four mutation strategies are adopted, and the "roulette wheel" method is used to prioritize the scheduling scheme with higher fire source suppression efficiency and lower resource matching cost, as shown below: in, For the first The first generation of the population A variant. , , , and The first The first generation of the population , , , and Individual, The top in the current population For a random outstanding individual, the function sequence Constructing the Fibonacci sequence ; Boundary detection of genes in variants is performed using a "boundary absorption" strategy: ； S4. Population Crossover and Selection: After mixing the genes of mutants and individuals from the original population, experimental subjects are obtained, generating new candidate schemes that fuse different scheduling schemes. The crossover calculation formula is: By employing a "greedy strategy," selecting superior individuals from the parent population and the experimental subject set to obtain the next generation, we have: in, For the first The first generation of the population One test subject, For the first The corresponding first generation in the population Individual, , The fitness solution function is based on the fire source suppression benefits and resource matching costs. S5. Population Information Sharing: After each iteration, the top 30% of individuals in fitness are added to the candidate elite set, and the poorly performing individuals in the elite population are replaced. After the iteration ends, the best individual in the elite population is the final air-ground cooperative scheduling solution.

9. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction according to claim 1, characterized in that, The improved auction algorithm based on a centralized architecture is as follows: Pricing strategies utilizing distance information: when resources For the task When submitting a quote, the lower limit of the price increase. The calculation formula is: in, For the custom price increase lower limit distance weighting coefficient, For the quantity of resources, For resources With the task The distance between them The range of values ​​is Control parameters; An adaptive operator for the number of quotes: using an adaptive operator. To control The value of , The calculation formula is as follows: in, It is a positive constant. Representing resources For the task The number of times a quote is made. and Indicates the maximum and second-largest benefits that can be obtained by performing the task; Multiple auctions with dynamically updated task value: The value update formula is expressed as: in, These are customizable weighting coefficients used to balance historical value with the current environment. For resources Execute the task The success rate For customizable about The weighting coefficients, , For the task The comprehensive objective value is calculated using the following formula: in, The fire spread area within the mission area. For wind speed, For terrain slope, For humidity, With each indicator weighted and the sum of the weights equal to 1, this update mechanism can dynamically enhance the mission value of areas with severe fires, harsh terrain, high wind speeds, and low humidity, and accurately guide resource responses to high-risk areas.

10. The forest fire air and land resource scheduling method based on differential evolution and adaptive auction according to claim 1, characterized in that, The solution process for the improved auction algorithm based on a centralized architecture is as follows: Treating resources as bidders and tasks as bidding objects, the average movement distance of each resource is the objective function. The total value of completing the firefighting mission is the objective function. Let the maximum number of resources that can be allocated to each task be . The specific process is as follows: S1. Initialize the location of resources and target tasks, available resource set, resource list, resource value, resource weight, and maximum resource capacity; S2. Based on the input parameters, generate a random resource bidding sequence as the initial bidding order; S3. For each unassigned target task, perform the following operations: When there are unassigned resources, select the target task; if the number of resources allocated to the task has reached a certain threshold... Then choose from the remaining tasks; S4. Calculate the price of each resource, and the task will be assigned to the resource with the highest price. S5. After each round of bidding, save the current resource allocation plan and update it. , ; S6. Perform a non-dominated sort on all the solutions to obtain the optimal Pareto solution set.

Images