Disaster detection task planning method, device and equipment based on unmanned aerial vehicle cluster and storage medium

Abstract

The invention discloses a task allocation and optimization method, system and device based on a hierarchical task network, and the method comprises the steps: generating an unmanned plane task set through expert knowledge, and initializing task allocation parameters; then task level decomposition is carried out, a task allocation initial solution is established, and a to-be-executed sub-task set is formulated for each unmanned aerial vehicle; establishing a task constraint and an optimization target, and solving a task planning problem by adopting an improved genetic algorithm; and finally, a conflict-free task sequence is obtained through a conflict resolution and replanning algorithm. The system comprises an unmanned aerial vehicle cluster system, a ground station communication module, a central processing unit and a task planning solving module. An expert knowledge base for realizing task execution of the unmanned aerial vehicle cluster and a pre-execution instruction are stored in the device. By using the invention, the unmanned aerial vehicle cluster can efficiently execute the specified task set as required. The method can be widely applied to disaster rescue fields such as fields and urban areas.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and unmanned aerial vehicle mission planning, and particularly relates to a method, device, equipment and storage medium for disaster detection mission planning of an unmanned aerial vehicle cluster. Background Art

[0002] With the maturity of the industrial-grade unmanned aerial vehicle application and advanced sensor industry chain, unmanned aerial vehicles have been rapidly popularized in scenarios such as disaster rescue and fire emergency response. Under the current background of lightweight upgrading of flight control hardware and specialization of navigation and control algorithms, unmanned aerial vehicles flying in harsh scenarios can also maintain good mission performance. According to the change of the payload carried by the unmanned aerial vehicle, it can complete various different disaster situation tasks. At present, the applications of unmanned aerial vehicles in disaster scenarios mainly include disaster monitoring, disaster area inspection, post-disaster search and rescue, disaster situation assessment, communication maintenance, etc. Conducting research on regional detection and mission planning of multi-unmanned aerial vehicle systems for emergency rescue in complex disaster environments, including problems such as static mission planning, dynamic mission planning and trajectory planning, is of great significance for improving the functionality and robustness of multi-unmanned aerial vehicle systems in complex unknown environments.

[0003] For the multi-UAV detection mission carried out in a disaster scenario, task planning for it can reduce the possibility of task conflicts between UAVs and improve the detection efficiency of the overall UAV cluster system. In the existing published literature, for example, in the existing invention patent application document "Task Dynamic Allocation System and Method Based on Artificial Intelligence" with the publication number CN117973820A, the existing method includes: providing a task dynamic allocation system and method based on artificial intelligence, related to the field of dynamic task allocation. The method includes: obtaining various task-related information and UAV state information in the task sequence to be executed; extracting task features such as the priority, time limit, and location information of each task in the task sequence to be executed to obtain an execution task feature vector, and extracting state features such as the location, speed, power, and available load capacity of each UAV in the UAV formation to obtain a UAV state feature vector; after fusing the execution task feature vector and the UAV state feature vector, performing relevant interference compensation based on the category probability value to obtain a compensated task allocation feature vector; generating a task classification strategy based on the task allocator. It can effectively solve the problems in UAV formation task allocation, improve the task completion degree and execution efficiency, and adapt to the demand changes in complex task environments. And in the existing invention patent application document "Firefighting UAV Cluster System and Firefighting Method" with the publication number CN108241349A, the existing method includes: disclosing a firefighting UAV cluster system and a firefighting method. The task ground station is respectively connected to the fire monitoring / locating UAV subgroup, the fire bomb dropping UAV subgroup, and the special rescue UAV subgroup through an air self-organizing network communication link, and the UAV ground station is respectively connected to the fire monitoring / locating UAV subgroup, the fire bomb dropping UAV subgroup, and the special rescue UAV subgroup through a ground-air data transmission communication link. The fire monitoring / locating UAV subgroup is respectively connected to the fire bomb dropping UAV subgroup and the special rescue UAV subgroup through an air self-organizing network communication link. Through the air self-organizing network communication link and the ground-air data transmission communication link, the present invention realizes the transmission and sharing of control instructions and fire monitoring information with each UAV subgroup. The whole system has a high degree of intelligence and high working efficiency, and can complete firefighting tasks quickly and efficiently.

[0004] In the process of multi-UAV mission planning, it is necessary to ensure that each specified detection mission is not repeatedly assigned, that is, each mission forms a one-to-one relationship with the UAV. From the specific implementation content of the aforementioned prior art, it can be seen that the current research represented by the aforementioned prior solutions mainly has the following deficiencies: some literatures have too simple modeling assumptions for disaster areas and are not applicable to actual scenarios; the research and modeling of environmental information and UAV sensor models are not sufficient, and the actual complex and unknown disaster situations cannot be effectively characterized; for actual disaster rescue scenarios, the dynamic evolution characteristics of the disaster situation and the full utilization of detection information are not fully considered; the expert experience of disaster rescue is not used in the actual UAV system and multi-UAV mission planning problem, and the actual detection constraints are not further considered to achieve an efficient UAV swarm mission planning solution and apply it to complex dynamic scenarios. Summary of the Invention

[0005] To solve the above technical problems, the purpose of the present invention is to provide a disaster detection mission planning method, device, equipment and storage medium based on UAV swarms, which can inject the expert experience of disaster rescue into the UAV swarm system in the form of prior knowledge, and plan the task execution sequence based on the dynamic hierarchical task network for the UAV swarm to achieve efficient detection of disaster areas.

[0006] The technical problems to be solved by the present invention are: how to solve the technical problems in the prior art of insufficient and inaccurate modeling of disaster areas, difficult guiding application of expert experience in disaster rescue in the actual UAV swarm system, and lack of reasonable task allocation for the planning requirements of UAV swarms.

[0007] The present invention solves the above technical problems by adopting the following technical solutions: A disaster detection mission planning method, device, equipment and storage medium based on UAV swarms includes: The present invention focuses on two core contents of the multi-UAV detection mission planning problem in the disaster area: the decision-making scheme generation module and the task planning module; The decision-making scheme generation module extracts the actual characteristics of the disaster area by analyzing the characteristics of common disaster situation environment map modeling methods, clarifies the modeling methods and constraints applicable to the planning problems in the disaster scenario, and lays a foundation for the UAV swarm disaster detection mission planning method. The present invention designs a conversion of the decision-making understanding of expert knowledge dedicated to the emergency field into a task scheme that can be autonomously executed by the UAV, and proposes a task planning algorithm for multi-UAV disaster detection. Aiming at the problem that it is difficult for the machine system to structurally express the expert rescue decision in the emergency rescue task, a conversion scheme from expert knowledge in the disaster rescue field to detection tasks is designed, and a hierarchical task network is used for subsequent task decomposition and planning, thereby optimizing the detection task planning scheme; The drone cluster task planning module contains two mechanisms, one is the task decision decomposition and coordination conflict elimination mechanism within the multi-drone system, and the other is the task allocation mechanism between multiple drones. The purpose of both is to maximize the role of the drone cluster system. Secondly, the drone cluster system task planning module involves coupled optimization problems, which can be divided into sub-problems such as scheduling planning, trajectory constraints, and task allocation. In addition, scheduling planning involves many situations such as aircraft dynamics constraints, terrain constraints, task priority constraints, and task time window constraints. Among them, the commonly used method of task decision decomposition is to gradually decompose the composite tasks in the task set into atomic tasks that the drone cluster system can understand, and the atomic tasks cannot be decomposed further. A composite task is often composed of multiple atomic tasks, and there are related constraints between the atomic tasks; To achieve the above technical objectives, according to a first aspect of the present invention, the present invention provides a method for generating a decision solution based on expert knowledge, comprising: S1. The decision-making scheme generation method is based on the existence of a task set in the disaster scenario. Generate, specifically, the tasks in the task set include atomic tasks and composite tasks. In the detection problem, atomic tasks include detection detection tasks and continuous exploration mission Among them, the detection task requires the drone to reach a certain range above the detected area and perform imaging detection of the area according to a given viewing angle without staying for a long time; for the continuous detection task, the drone is required to reach a certain range above the detected area and remain in a hovering state for a certain period of time. During this period, the drone's viewing angle needs to be aligned with the detected area. The area alignment can be achieved by adjusting the angle of the pod carried by the drone. The supplementary detection task generated in a dynamic scene is regarded as a derivative of the detection task. The detection task is expressed as:

[0008] In the formula, Indicates the location of the drone. is the position range required for the detection task; similarly, The pitch and yaw angles of the pod carried by the drone relative to the detection mission center, which meet the relative constraints required by the mission ; The UAV swarm system usually needs to meet the corresponding trajectory constraints, dynamic constraints, endurance constraints, task sequence coupling constraints and task execution time window constraints when planning tasks in disaster areas. The specific constraint description and analysis include: Specifically, the disaster environment constraints and the drone trajectory constraints are: The constraints of the disaster situation environment mainly refer to the limitations of the flyable areas of drones in the disaster situation environment, including the geographical information constraints of the disaster area, as well as the uncertain limitations such as no-fly zones, areas with unstable communication, and dynamic disaster areas. The constraints of the disaster situation environment will affect the specific flight path after the mission planning is completed. In addition, when conducting mission planning, the path cost and time cost are usually used to evaluate the quality of the planning results, and the path cost and time cost also need to be calculated according to the flight path constraints of the drone; Specifically, the dynamic constraints of the drone are as follows: When conducting the detection mission planning, the dynamic constraints of the drone need to be considered. It mainly includes the maximum flight speed of the drone, the detection imaging constraints of the drone, etc. Specifically, during the actual mission planning process, the flight speed of the drone should be within the allowable flight speed threshold. Secondly, when the drone is conducting detection, too fast a flight speed will make it impossible to meet the requirements of the detection imaging model; Specifically, the constraints of the drone's endurance are as follows: For the actual drone system, the longest endurance distance constraint of the drone also needs to be considered. The limited endurance time and distance limit the allowable detection time and flight distance of the drone. During the mission planning process, the path cost calculated according to the task sequence assigned to the drone individual shall not exceed the maximum flight endurance distance of the drone, and the drone needs to return to the starting point or the designated landing location before the end of the endurance time; Specifically, the coupling constraints of the task sequence are as follows: The coupling constraints between detection tasks include the sequence relationship and the cooperation relationship. For two tasks with a sequence relationship and ( ), the task with a later sequence coupling needs to be carried out after the task with a previous sequence is completed. Tasks with a cooperation relationship need to be completed jointly by multiple drones, and the execution times of each drone for the tasks with a cooperation relationship can be staggered; Specifically, the task execution time window constraint is as follows: The time window constraint of the mission planning means that the task needs to be completed within the specified time window. Taking the detection mission as an example, according to the priority of the detection mission or the coupling relationship of the task sequence, the detection mission with a time window constraint needs to be completed as soon as possible; S2. The above flight path planning problem is proposed in the context of disaster area detection and combined with the actual disaster detection requirements, and the established optimization problem is described as follows: For the drone detection imaging flight path planning problem of the disaster map, its mathematical model is established as follows:

[0009] In the formula, is the path sequence length of the UAV , is the maximum endurance time of the UAV is the maximum allowable distance when the UAV executes a single-step plan.

[0010] S3. The task decision understanding and planning constraint method proposed in S2 is described in a unified framework. In the disaster scenario, the multi-UAV system adopts a centralized communication framework, that is, each UAV can receive the route and task information assigned by the ground station terminal, and transmit back the sensor information collected by the on-board load, such as RGB images, etc., and the environmental information maps of all UAVs are shared.

[0011] Considering the above constraints, the following objective function is designed for multi-UAV mission planning:

[0012] In the formula, is the overall cost function, to are the path information benefit efficiency function, the path cost efficiency function, the task execution cost efficiency function, and the task balance cost efficiency function respectively, to are the weight coefficients.

[0013] In the formula, the efficiency function defines the total path information benefit of the multi-UAV system, where is the grid decision variable, representing the UAV at the map grid point and time 's probability information gain.

[0014] In the formula, the efficiency function represents the total path cost of the multi-UAV system, represents the path point sequence of the UAV .

[0015] In the formula, the efficiency function represents the total task cost of the multi-UAV system, is the task decision variable, representing the UAV to complete the task 's cost value, including the time cost and the endurance cost of the UAV.

[0016] In the formula, the efficiency function represents the task balance cost, and respectively represent the UAV system At the task balancing cost and path balancing cost at the moment, where is the weight coefficient.

[0017] To achieve the above technical objectives, according to the second aspect of the present invention, the present invention provides a task planning method for an unmanned aerial vehicle (UAV) cluster, including: S4. For the task planning problem in UAV cluster disaster detection, design a fast UAV cluster task planning algorithm based on the hierarchical task network algorithm and the improved genetic algorithm, analyze the feasibility and applicability of the algorithm under different disaster scenarios, and evaluate the planning results using a variety of evaluation indicators; The hierarchical task network algorithm has powerful advantages in task decomposition and task constraint processing such as endurance constraint, time window constraint, and precedence constraint. It processes the number of tasks through a method set, operators, etc. to obtain a feasible solution. The hierarchical task network decomposes and solves the constraints of the tasks in the task set, and can obtain a feasible solution for the task planning of a coarse-grained multi-UAV system. On this basis, it is also necessary to optimize the task sequence, and balance the path cost and time cost of the allocation result through the total cost function and the effectiveness function. Since the disaster points in the tile map are discretely distributed, the ground station uses interactive commands as the input for the UAV cluster waypoint planning in the map. The method based on gradient descent can quickly find the direction of the spread of the disaster, thereby finding the source point of the disaster, saving the endurance time of the UAV, and quickly obtaining the disaster information in the area.

[0018] Preferably, the genetic algorithm adopted by the present invention is a bionic algorithm based on the genetic process in the biological world. Through heuristic inheritance, crossover, and mutation operations, it inherits the optimal solution generated in each iteration process to the initial solution of the next iteration process. It has high efficiency in the solution process and can quickly obtain an optimized solution for task allocation optimization. Although the genetic algorithm cannot ensure that the optimal solution is obtained in the final iteration, that is, the iteration only converges to a local optimum, it can still obtain a relatively good solution that can generally be accepted by the multi-UAV system. This is a balanced choice considering the balance between the time cost and the quality of the solution set when solving the task problem in a complex disaster scenario. Therefore, considering from the aspects of the quality of the solution and the iteration convergence speed, the genetic algorithm is preferably selected as the method for task allocation optimization; S5. Further, the initial solution pair generated by the hierarchical task network algorithm is used to generate the encoding of the next-generation population until a new population is formed. Then, iterative operations of the genetic algorithm are performed, including selection, mutation, and crossover, etc., and the optimal solution in each generation is selected and passed on to the next generation until the maximum iteration period is reached. For the convenience of calculation, the initial solution generated by the hierarchical task network algorithm needs to be encoded and assigned to the UAV sequence. The operation functions required include an initialization function, a selection function, a crossover function, and a mutation function. Among them, the initialization function mainly encodes the initial solution generated by the hierarchical task network algorithm and initializes it according to the set number of genes; the selection function needs to preferentially extract excellent gene individuals in the current population as the initial individuals of the next generation according to the preset selection probability; the crossover and mutation operators are used to generate the next-generation individuals. Combining the foregoing task planning model, a fitness function for allocation optimization is set, so as to provide a basis for the selection function; Preferably, the present invention uses a selection method that combines roulette wheel random selection and tournament selection. In this method, individuals to be competed are obtained through roulette wheel selection in the selection stage, and then better individuals are obtained through tournament selection, so that the result of roulette wheel random selection is better. This selection method not only ensures the relative balance of computational complexity but also improves the diversity of the population through roulette wheel random selection. During the tournament selection process, individuals generated through roulette wheel selection are obtained. When the number of individuals is greater than 2, the competition process starts, and then the best one is selected to enter the offspring population. Repeat this operation until the scale of the new population reaches the requirement; The specific steps of S5 include: S51: Select an individual from the offspring population ; S52: Select a certain number of rows in the encoding matrix according to the mutation probability, and perform transposition mutation and random mutation operations on them. The remaining genes that have not undergone mutation operations are retained in order to generate new individuals ; S53: Merge the population and the population obtained after the mutation operation , and calculate the fitness of each encoding matrix. After arranging them in ascending order, retain the first encoding matrices.

[0019] Preferably, considering that the rate of change of the fitness convergence speed in the population with iteration generally varies, an adaptive method for adjusting the mutation probability is set:

[0020] In the formula, is the mutation probability, and are the minimum fitness and the maximum fitness in the population respectively, is the fitness of the individual to be mutated, and are the current iteration cycle and the maximum iteration cycle respectively; is the sigmoid function, .

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: 1) The present invention combines expert experience knowledge and task planning algorithms to establish a task planning model for UAV swarms, and applies an improved genetic algorithm to solve the task allocation problem, which can meet specific usage requirements under actual large-scale verification conditions.

[0022] 2) Compared with the original genetic algorithm in terms of the task planning results of the method of the present invention, the improved genetic algorithm can effectively reduce the fitness value of the cost function and improve the usage efficiency of UAV swarms, and has advantages in both instantaneity and effectiveness in disaster detection tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Schematic diagram of UAV swarm task allocation

[0024] Figure 2 Flow chart of a disaster detection task planning method based on UAV swarms DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Based on the embodiments in this application, other similar embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application. In addition, the directional terms mentioned in the following embodiments, such as "upper", "lower", "left", "right", etc., are only with reference to the directions of the drawings. Therefore, the directional terms used are for illustration rather than to limit the present invention.

[0026] The present invention will be further described below in conjunction with the drawings and preferred embodiments.

[0027] Embodiment 1: S1. Referring to Table 1, the decision-making scheme generation method provided by the present invention is generated based on the task set in the disaster scenario. Let be the set of UAVs executing tasks in the current scenario, be the number of UAVs. For each UAV , the set of tasks to be executed is set as , and the set of tasks it has completed is The method includes steps S101 - S102; S101, Establishment of domain knowledge The relationship between the decision-making schemes involved in emergency management and disaster rescue processes and the mission constraints and execution priorities of the UAV system is described in Table 1, covering the expert decision-making schemes in most actual disaster scenarios. Among them, the priority is a relative value compared to other tasks, and it does not indicate the importance of the task in the actual scenario. In addition, for the decision-making schemes not fully listed, corresponding scheme tables can also be compiled according to the actual situation; Table 1 Emergency rescue domain knowledge for multi-UAV system migration

[0028] S101. Task planning continuously decomposes the composite tasks in the task set through predefined domain knowledge until only atomic tasks are used for execution. Before solving the hierarchical task network, it is also necessary to establish domain knowledge and define the planning domain, operators, and method set; It can be understood that as the number of individuals in the multi-UAV system increases, a single UAV is more likely to encounter emergencies such as airframe failures, environmental interference, and mission conflicts. Blindly increasing the number of UAVs may not only result in low mission efficiency but also lead to mission failures. On the contrary, when the number of UAVs is less than a certain threshold, it cannot guarantee the efficient completion of tasks either; S102, Definition of the planning domain, operators, and method set Set the performance parameters of the multi-UAV system , and respectively represent the maximum endurance time and maximum speed of the UAV . In the mission planning stage, it includes a static task set given at the initial moment and dynamic tasks that occur during the mission execution. After completing the definition of domain knowledge, it is also necessary to formulate the operators and method set of the mission, and these two together constitute the planning domain of the HTN method . The operator illustrates the constraint conditions required for the execution of atomic tasks and the target results after the completion of atomic tasks. The method set provides the method for decomposing the mission into atomic tasks; According to the operator definition in the PDDL language, the operators of the detection mission and the time window mission are specifically defined as follows: Preferably, define the UAV cluster detection mission: where the operator of the detection mission is a nine-tuple:

[0029] Among them, is the head of the operator, is a first-order expression, representing the corresponding constraint conditions before, during, and after the execution of the task, represents the parts that need to be deleted from the overall task result before and after the execution of the task, represents the parts that need to be added to the overall task result before and after the execution of the task. Specifically, for the detection task, it needs to meet the detection imaging constraint during the task execution, and in the case of long-range flight, it is required to meet the UAV endurance constraint. After the task execution, the detected area is newly added to the overall task, and the completed part in the task sequence of the undetected area can be deleted.

[0030] Preferably, define the time window task: the operator of the detection task is a ten-tuple:

[0031] wherein, is the time window constraint. For the time window task, it is required to retrieve the time window information and constraint conditions before the task execution , and during the continuous execution of the task, it is judged whether it is within the allowable time window at . The definitions of the remaining parts are similar to those of the detection task.

[0032] S2. Specifically, the detection task for the disaster area is defined in two ways, including: the detection location is circled by emergency management personnel, and the other is to autonomously generate the area detection task; To clarify the input of the task planning, the task types to be involved are defined here, including: 1) Definition of simple detection task, including: The UAV only needs to reach above the area to be detected for image acquisition. The time spent in this part of the task planning is relatively short; 2) Definition of continuous detection task, including: The UAV needs to stay above the area to be detected for a specified time, and continuously transmit the information of this area during this period. The time spent in this part of the task planning is relatively long; 3) Definition of supplementary detection task, including: After completing the detection task of a certain area, the omitted areas caused by poor flight path planning or detection effect need to be detected again, and the situation where the disaster point has moved and needs to be detected again; 4) Definition of complex detection task, including: Multiple UAVs are required to detect a certain location, and the access order of different UAVs to this location is not additionally specified in the present invention; 5) Definition of time window task, including: The UAV needs to complete the specified detection task within the specified time window. If it fails to complete the task within the time window or the task duration exceeds the time window, the task is regarded as failed. 6) Composite task definition, including: It consists of multiple of the above basic task forms and there is a certain task sequence. For example, the supplementary detection task usually occurs after the detection task is completed. S3. When performing task allocation, the following conditions also need to be followed: S301. The UAV performs detection and shooting tasks over the disaster area and needs to avoid some inaccessible areas such as smoke areas, no-fly zones, and areas with harsh conditions. S302. The boundary range of the disaster area is known, but the actual situation and disaster distribution within the area can only be determined when arriving during the flight process. S303. Within the disaster area, the disaster points are time-varying, which maps from the dynamic evolution process of real-world disasters such as forest fires, floods, debris flows and other scenarios. S304. The UAV needs to complete two detection tasks: one is to perform as much coverage detection of the disaster area as possible, and the second is to detect and shoot the spreading direction of the disaster and the disaster points. S305. When performing the shooting task, the images of the disaster area should be kept as complete as possible, and there should be a certain overlap degree between adjacent captured images, and the presentation of the disaster area and the disaster trajectory should be achieved. S306. The endurance time of the UAV is limited. Its endurance ability includes the flight path length and the hovering time, and is predefined at the beginning of the simulation.

[0033] When performing task allocation for the multi-UAV system, the following mechanism is satisfied: 1) The multi-UAVs should generate as few conflicts as possible in a single task plan. 2) The information gain of the multi-UAVs in the disaster area should be as large as possible. 3) The path cost of the multi-UAVs in the disaster area should be as small as possible; the task plan of the multi-UAVs should be balanced. 4) Specifically, it is reflected in the balance of the path length and the balance of the task execution time. 5) Reasonably plan the number of UAVs dispatched.

[0034] S4. The specific operation steps are as follows: S41. Fitness function The fitness function involved in this embodiment includes three terms: the maximum flight distance of a single drone in the drone system, the average flight distance during the task execution process, and the time-flight distance balance cost within the drone system. The flight distance is calculated based on the task assignment result, the kinematic model of the drone, and the Dubins path. The maximum flight distance is the longest flight distance generated in the drone system during the task execution process. The average flight distance is the average flight distance of the drone system during the task execution process. The time-flight distance balance cost represents the effective utilization degree of the task assignment result for the drone system. The total fitness is the weighted sum of the three sub-fitness values. The smaller the fitness, the more reasonable the rescue task assignment result is.

[0035] S42. Initialization function and encoding / decoding

[0036] The initialization function is generated using the initial solution generated by HTN and the encoding rules. In this embodiment, a two-dimensional matrix with the number of rows and columns being the number of drones and the number of tasks respectively is used to represent an encoding. As Figure 1 shown, where the rows represent the drone numbers and the columns represent the task sequences. An element in the two-dimensional matrix represents that a corresponding task is undertaken by a certain drone. The task assignment scheme represented by the encoding matrix saves the task sequences that each drone flight needs to execute, including the task numbers and orders to be executed, and the order of specific tasks needs to be maintained without being disrupted during subsequent genetic operations.

[0037] S43. By querying each element in the encoding matrix, a drone task sequence as Figure 1 shown can be constructed. The decoding program adds the tasks of the decision variable to the corresponding task lists of the corresponding drones, so as to obtain the task assignment result represented by the encoding matrix. Figure 2 represents that drone will execute task , drone will execute task , and so on. During the encoding process, since the order of task execution has been defined when generating the feasible solution, it is necessary to take into account the order of tasks in the time lines of the same and different drones, and ensure that the order before and after specific tasks conforms to the task logic. According to the task execution order of each drone in the encoding matrix, the time scale when the drone completes the task can be calculated. By recording the time scale when each task is completed, it can be judged whether the task sequence with task order constraints conforms to the corresponding task logic. When there is a conflict in the task sequence, the task execution order needs to be adjusted until the encoding matrix meets the specified task order constraints. By adopting the form of matrix gene encoding and the method of cooperating with the adjustment of the feasible solution order conflict, the complexity of encoding and decoding the task sequence can be effectively reduced, which is convenient for subsequent operations of genetic operators.

[0038] S44. The selection of the population will affect the performance of task planning and solution. The selection operator selects excellent individuals from the current population and uses these individuals as the parents of the next generation of individuals. Usually, the selection function is probability-based, and the probability of the selected individuals is associated with their fitness, so that individuals with higher fitness have higher advantages. The commonly used selection function is the roulette wheel selection method, in which the probability of selecting an individual is directly proportional to its fitness value. This is equivalent to using a roulette wheel and assigning fitness values to a part of each roulette wheel. Each time a random selection is made, the probability of the selected individual is proportional to the magnitude of the fitness value.

[0039] S441. Determine the number of individuals selected each time. Generally, 2 are selected.

[0040] S442. For the individuals generated by roulette wheel selection, according to the fitness value of each individual, select the individual with the optimal fitness value to enter the offspring population.

[0041] S443. Repeat step S442 until the number of individuals constituting the new generation population reaches the requirement.

[0042] S45. Perform the crossover operation. This step is the main operation in the improved genetic algorithm and largely determines the global search ability of the algorithm. In the genetic algorithm, the design of the crossover operator is closely related to the adopted coding method in addition to being related to specific optimization problems. For different coding methods, corresponding crossover operators should be adopted. Usually, the crossover probability is related to the individual fitness. For the stage where the fitness changes slowly, the crossover probability can be appropriately increased. Set the maximum number of evolutionary generations and the maximum continuous retention generations of the optimal individual in the population. When either condition is met, the optimization process terminates and the current optimal solution is output. The present invention performs the crossover operation on the coding matrix using the non-uniform crossover operator and k the point crossover operator. The coding matrix after coding is denoted as and the offspring coding matrix obtained after performing the crossover operation on it is denoted as ( is the population size). On the basis of sorting the population according to the fitness value, the optimal individual in the population is set as the main coding matrix for the next round of iteration. The specific crossover operation is as follows.

[0043] S451. Select the coding matrix as the parent.

[0044] S452. Sort the parent coding matrix according to the fitness value. First, use the non-uniform crossover operator to cross the odd-numbered coding matrices and the even-numbered coding matrices arranged in sequence pairwise, and keep the positions and orders of the task points with task order constraints during the crossover process.

[0045] S453: The offspring coding matrix obtained after the previous step of crossover operation is , and then the offspring individuals are sorted according to fitness , select the top 30% of the individuals to perform k-point crossover with the even terms of the remaining individuals. According to each crossover, two new individuals are generated, and then their fitness values are calculated. Then the remaining offspring populations are merged, and finally sorted according to the fitness values. The number of populations with the specified population quantity in the front is taken as the new population .

[0046] S46: Mutation operation. First, select a coding matrix, and then according to the mutation probability , randomly select a certain row in the coding matrix for mutation operation. It specifically represents the task sequence of the UAV. Since there are sequence constraints for the tasks in the task sequence, it is also necessary to select variable abnormal points according to the sequence constraints of the position where the task is located, and randomly select a position within the allowable range of the constraints to insert the specified task , so as to ensure that the generated task sequence does not break the previous constraint relationship. For the specified task with task sequence constraints , only the method of swap mutation is used to try in the feasible interval in turn, and the row that satisfies the first feasible solution is retained. For the task set without task sequence constraints , perform random mutation operation on it.

[0047] In summary, the disaster detection task planning method for the UAV cluster based on the embodiments of the present application is clarified. First, it obtains the actual characteristics of the disaster area and selects a modeling scheme applicable to the current disaster scenario to generate the constraint conditions of the UAV cluster. Secondly, the proposed task decision understanding and planning constraint method is incorporated into a unified framework for description. Finally, the task assignment is mathematically modeled and solved, and the task assignment of the UAV cluster is completed through an improved genetic algorithm. In this way, the problems in the task planning of the UAV cluster can be effectively solved, the efficiency and robustness in the execution process of the disaster detection task can be improved, and the UAV cluster can adapt to the task demand changes in the complex disaster environment.

[0048] All changes and modifications made without departing from the spirit and scope of the present invention, all equivalent technical solutions also belong to the scope of the present invention.

[0049] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same and similar parts among the embodiments can be referred to each other.

[0050] Those skilled in the art will understand that the embodiments of the present invention can be provided as a method, an apparatus, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD ROM, optical memory, etc.) containing computer-usable program code.

Claims

1. A disaster detection mission planning method based on drone swarm, characterized in that: Includes the following modules: Module 1: Decision-making plan generation module, which is used to convert disaster rescue expert knowledge into detection mission plans that can be executed by drones, including: Establish domain knowledge based on the task set in the disaster scenario (including atomic tasks and composite tasks), and define the planning domain, operators and method sets; Decompose the composite task into atomic tasks, including detection tasks, continuous detection tasks and supplementary detection tasks, and define the execution constraints of each task; Module 2: Task planning module, which is used to generate the initial task allocation plan through the hierarchical task network algorithm and optimize the task allocation in combination with the improved genetic algorithm. Specifically, it includes: The hierarchical task network algorithm is used to process the trajectory constraints, dynamic constraints, endurance constraints, task sequence coupling constraints and time window constraints between tasks to generate a feasible solution for coarse-grained task planning. An improved genetic algorithm is used to iteratively optimize the initial solution, including genetic operations based on adaptive mutation probability, to balance the path cost, task execution cost and task equilibrium cost.

2. The method according to claim 1, characterized in that: The optimization process of improving the genetic algorithm in the task planning module includes: A selection method based on a combination of roulette random selection and tournament selection is used to generate the offspring population; Through the mutation operation of adaptively adjusting the mutation probability, the mutation probability formula is: in, is the current individual fitness, is the maximum fitness of the population, is the current iteration number, is the maximum number of iterations.

3. The method according to claim 1, characterized in that The constraints of the atomic task include: Detection The detection mission requires the UAV to complete imaging detection in a specified area; Continuous detection missions require the drone to hover in a designated area and continuously collect data; The supplementary detection task is triggered by the dynamic evolution of the disaster situation and must meet the time window constraints and task sequence constraints.

4. The method according to claim 1, characterized in that The objective function of the task planning module is a weighted sum of multiple performance functions, including: Path information gain efficiency function, used to maximize the information gain of the disaster area; Path cost efficiency function, used to minimize the total flight path of the drone cluster; Task execution cost efficiency function, used to minimize the time and endurance cost of task execution; The task balancing cost-effectiveness function is used to balance the task allocation and path load among UAVs.

5. The method according to claim 1, characterized in that The mission planning module adopts a centralized communication framework. The ground station coordinates the drone cluster to perform detection tasks by sharing environmental maps and task allocation instructions, and receives sensor data sent back by the drones in real time.

6. A disaster detection mission planning device based on drone clusters, characterized in that: include: A decision solution generation unit, used to convert expert knowledge into structured task solutions; Task planning unit, used to generate optimized task allocation results through hierarchical task network algorithm and improved genetic algorithm; The communication control unit is used to realize command transmission and data synchronization between the drone cluster and the ground station.

7. The device according to claim 6, characterized in that The mission planning unit also includes: Constraint processing module, used to analyze trajectory constraints, task priority constraints and dynamic disaster evolution constraints; The encoding and decoding module is used to encode the task allocation plan into a two-dimensional matrix and generate the UAV's task sequence through decoding.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

10. The method according to claim 1, characterized in that The operator definitions in the hierarchical task network algorithm include: The detection task operator is a nine-tuple (task header, precondition, execution condition, postcondition, delete list, add list, endurance constraint, imaging constraint, time constraint). The time window task operator is a ten-tuple (task header, precondition, execution condition, postcondition, deletion list, addition list, endurance constraint, imaging constraint, time constraint, time window parameter).

Images