Energy supply and task path optimization method for mobile platform cluster operation

Abstract

The present disclosure provides a kind of energy supply and task path optimization method of mobile platform cluster job, it is related to multi-agent task allocation and path planning technical field.The method encodes carrier layer, platform layer, task layer information into individual, wherein carrier layer information is the collection of transport vehicle stopping place, including path starting point set, and path termination point set;Platform layer information is the matching relationship of unmanned platform and path starting point and path termination point;Task layer information includes the matching relationship of unmanned platform and various tasks including battery swap task, and task execution sequence;Initialize individual in the order of carrier layer, platform layer, task layer, generate initial population;Population optimization is carried out using cultural gene algorithm, and the multi-agent cluster job energy supply and task path collaborative planning scheme is output.The present application can simultaneously plan unmanned platform energy supply and task execution path.

Description

Technical Field

[0001] This invention relates to the field of multi-agent task allocation and path planning technology, specifically to a method for energy replenishment and task path optimization in mobile platform cluster operations. Background Technology

[0002] With the rapid development of intelligent technologies, multi-unmanned platforms have been widely used in numerous fields due to their many advantages, such as environmental and ecological monitoring, commodity logistics and transportation, power facility inspection, smart city security, and disaster area emergency rescue. In multi-unmanned systems, the efficiency of collaborative operations is effectively improved through multi-agent capability collaboration, multi-party resource integration, and multi-source information sharing. However, the diversity of task types, the differences in platform capabilities, and the challenges of synergy between heterogeneous platforms lead to more complex task allocation strategies and increased coupling of multi-dimensional system parameters in multi-unmanned systems, posing a severe challenge to improving the collaborative effectiveness of these systems.

[0003] This invention relates to a multi-type mission requirement scenario: multiple transport vehicles carrying unmanned platforms. Each transport vehicle can carry one or more unmanned platforms, with the number depending on the vehicle's loading capacity. The transport vehicles transport the unmanned platforms to a candidate location. The unmanned platforms then leave the transport vehicles to perform their missions. During mission execution, the unmanned platforms can refuel at the candidate locations. After completing the mission, they return to the candidate locations, board the transport vehicles, and are then taken away. The unmanned platforms can be heterogeneous, and there is no tight coupling between the heterogeneous unmanned platforms and the transport vehicles. The unmanned platforms can be carried back by any transport vehicle and can also refuel at any transport vehicle.

[0004] In the aforementioned task requirements scenario, the task allocation and path planning problem for energy-constrained multi-agent swarm operations falls under the domain of combinatorial optimization problems. The key challenge is to address how, within a given task environment, to allocate and coordinate multiple unmanned platforms with different functions by combining task requirements and agent performance. This is achieved through designing energy replenishment and return-to-base strategies, as well as unmanned platform-task adaptation criteria, and rationally planning task execution paths. The goal is to maximize swarm operation efficiency while meeting capability constraints, task requirements, and other relevant constraints. This includes, but is not limited to, minimizing task completion time, reducing overall operational costs or losses, and improving task completion quality and effectiveness. Ultimately, the aim is to integrate dispersed individual behaviors into a highly coordinated intelligent collaborative whole through systematic planning, resulting in intelligence emerging at the system level.

[0005] Given that unmanned platforms are electrically powered, their range cannot compare to that of fuel-powered vehicles. Therefore, to effectively address range limitations and improve operational efficiency, in the scenario studied, mobile transport vehicles can be used for the deployment and energy replenishment of unmanned platforms. Thus, before mission execution, the location of transport vehicles needs to be optimized to deploy the unmanned platforms, ensuring they can commence operations in the shortest possible time. During mission execution, transport vehicles provide energy replenishment to the unmanned platforms to ensure system sustainability. After mission completion, transport vehicles carry the unmanned platforms back to base or to subsequent mission areas. This operational process ensures the continuity and efficiency of mission execution, maximizes the operational coverage of unmanned platforms with limited range, and enhances the efficiency of group operations.

[0006] The aforementioned problem scenario involves multiple collaborative relationships between transport vehicles, between transport vehicles and unmanned platforms, and between heterogeneous unmanned platforms. It exhibits multi-level coupling between tasks, platforms, and vehicles, requiring collaborative decisions regarding transport vehicle parking locations, platform-task adaptation relationships, and unmanned platform task execution paths. Therefore, the multi-agent cluster operation optimization problem in this scenario is highly similar to the location-routing problem (LRP). The location-routing problem integrates three aspects of decision-making:

[0007] Determine a subset of candidate locations (facility site selection problem), which corresponds to the set of locations for unmanned platforms to be loaded (unloaded) by transport vehicles in this invention. Optimize the parking positions of transport vehicles to achieve energy replenishment and location deployment for multiple unmanned platforms.

[0008] The task-platform correspondence (task allocation problem) corresponds to the task-to-platform assignment in this invention. By optimizing the platform-task adaptation relationship, the capabilities and advantages of the unmanned platform can be fully utilized.

[0009] Task execution path optimization (vehicle routing problem) corresponds to the task execution path optimization of each unmanned platform in this invention, in order to minimize task completion time and improve task execution efficiency.

[0010] Currently, the optimization objectives of LRP (Location-Based Ratio) are typically to minimize transportation costs, maximize system revenue, and maximize customer satisfaction. Algorithms for solving the location-path problem are mainly divided into exact methods and intelligent optimization algorithms. The former mainly includes column generation, branch pricing, and dynamic programming. Since exact algorithms can only solve small-scale problems within an acceptable timeframe, they struggle to obtain feasible solutions for medium- to large-scale LRPs within a limited time. Therefore, many scholars have designed various high-performance intelligent optimization algorithms based on the characteristics of the LRP problem. The algorithmic frameworks involved mainly include: genetic algorithms, simulated annealing algorithms, tabu search algorithms, ant colony algorithms, and neighborhood search algorithms. By introducing heuristic rules, parallel computing, and hybrid strategies, intelligent optimization algorithms demonstrate strong global search capabilities and computational efficiency when solving large-scale LRPs. However, research reveals that current research on the location-path problem has several limitations when applied to task allocation and path planning problems in energy-constrained multi-mobile platform cluster operations with diverse task requirements:

[0011] (1) There is little introduction of mobile carriers and vehicle-to-station operations.

[0012] Current research on the location-path planning problem primarily focuses on logistics and transportation or home healthcare, where vehicles typically originate and terminate at warehouses or hospitals, lacking mobility and requiring vehicles to return to their starting points. Therefore, it lacks consideration for inter-station vehicle operations and collaborative planning of mobile carriers. In practical applications, such as disaster relief or urban security, allowing unmanned platforms to terminate their journeys at locations different from their starting points effectively reduces energy consumption during travel. Collaboration among multiple transport vehicles and unmanned platforms can improve task execution efficiency under limited resource conditions, making this problem of significant research value and practical importance. However, the introduction of inter-station unmanned platform operations and mobile carrier collaboration mechanisms leads to complex coupling between the task sets corresponding to each mobile carrier, making it difficult to solve effectively using task grouping and decomposition methods. Therefore, specifically designed intelligent optimization algorithms are needed for efficient solution.

[0013] (2) Insufficient ability to explore high-dimensional complex solution spaces

[0014] Existing algorithms face limitations when exploring complex solution spaces, including the curse of dimensionality, susceptibility to local optima, and the trade-off between computational efficiency and real-time performance. They struggle to efficiently solve high-dimensional, tightly coupled complex problems, especially in multi-mobile platform cluster operations. While the introduction of transport vehicle collaboration mechanisms improves task execution efficiency, it also exacerbates the multi-level nesting of constraints in task-resource-location relationships, further increasing the difficulty of problem-solving and reducing the practical effectiveness of existing algorithms in multi-agent cluster operations. Therefore, it is necessary to design targeted algorithms based on problem characteristics to effectively address the energy replenishment and task path collaborative optimization problems in multi-agent cluster operations with diverse task requirements, improve the ability to explore complex solution spaces, and significantly enhance the effectiveness and intelligence of multi-agent cluster operations. Summary of the Invention

[0015] In view of this, the present invention provides an energy replenishment and task path optimization method for mobile platform cluster operations. The task planning problem of energy-constrained multi-agent cluster operations in various task demand scenarios is modeled as a location-path planning problem with mobile carriers and energy replenishment. The method combines task requirements and unmanned platform performance to reasonably select the parking location of transport vehicles and plan the energy replenishment and task execution path of unmanned platforms. While meeting capacity constraints, task requirements and other relevant constraints, the method minimizes the task completion time as much as possible.

[0016] To solve the above-mentioned technical problems, the present invention is implemented as follows.

[0017] A method for energy replenishment and task path optimization in mobile platform cluster operations includes:

[0018] Step 1: Determine the candidate location, transport vehicle, unmanned platform, and mission information;

[0019] Step 2: Model the task planning problem. The constraints of the planning problem include: (1) The starting points of the transportation vehicles form the set of the path start points. The set of route termination points consists of the docking locations of the transport vehicles that return after the unmanned platform completes various tasks. , The elements in the data also serve as optional locations for unmanned platforms to perform battery swapping tasks; the set of path start and end points. and the set of path termination points Elements in the list are selected from candidate locations; the set of path start locations. and the set of path termination points (1) The elements in the content are the same or different; (2) The unmanned platform shall not consume all energy while driving, and the unmanned platform shall be fully charged after each energy replenishment;

[0020] Step 3: Encode the carrier layer, platform layer, and task layer information into individuals; wherein the carrier layer information is the set of transport vehicle stopping locations, including the set of path start and end points. and the set of path termination points Platform-level information includes the matching relationship between the unmanned platform and the starting and ending points of the path; task-level information includes the matching relationship between the unmanned platform and various tasks, including battery swapping, as well as the task execution sequence.

[0021] Step 4: Initialize individuals in the order of carrier layer, platform layer, and task layer to generate the initial population;

[0022] Step 5: Utilize the cultural gene algorithm to optimize the population and output a collaborative planning scheme for energy replenishment and task paths for multi-agent cluster operations.

[0023] Preferably, in step one, the information for determining the candidate location, transport vehicle, unmanned platform, and mission is as follows:

[0024] Determine the location and number of candidate locations, the number and loading capacity of transport vehicles, the number of unmanned platforms, the mission execution capability matrix and endurance time vector, and the location and requirements of the mission.

[0025] Elements in the task execution capability matrix of unmanned platforms for Type of unmanned platform performs tasks per unit time The amount completed.

[0026] Preferably, in step two, the overall optimization objective of the task planning problem is to minimize the cluster job completion time.

[0027] Preferably, in step two, the constraints of the planning problem further include:

[0028] (3) Loading capacity constraints of transport vehicles: A transport vehicle can carry a maximum of An unmanned platform;

[0029] (4) Unmanned platform Arrive at the mission Time spent The following conditions must be met:

[0030]

[0031] In the formula, For type unmanned platform The speed of travel; For unmanned platforms In the region and The driving distance between;M The total number of tasks. L The total number of candidate locations. This represents the total number of unmanned platforms.

[0032] (5) The task requirements cannot be broken down, and each unmanned platform cannot travel back and forth to the task location; they must leave after the task is completed.

[0033] (6) Each transport vehicle can park independently, meaning that at most one transport vehicle can park at each candidate location.

[0034] Preferably, step four, which initializes individuals in the order of carrier layer, platform layer, and task layer to generate the initial population, specifically includes:

[0035] Step 41: Initialize the carrier layer and determine the starting path set for the unmanned platform. and path termination set For each task, determine the nearest candidate location, and for each candidate location, count the number of times that candidate location is the nearest candidate point for the task. The number of times Reflects the density of task distribution around the candidate location; based on the number of times Determine the global selection probability of candidate locations ,frequency The higher the probability, the greater the global selection probability. The larger the value; based on the global selection probability. The starting set of paths is determined by using a round-robin method. and path termination set Elements in;

[0036] Step 42: Perform platform-level initialization and decide on the allocation of candidate locations for unmanned platforms, including the allocation of path start and end locations for unmanned platforms.

[0037] Based on the set of starting and ending points of the path determined in task and step 41. The distance between each element is used to group all tasks, and each task group corresponds to a path start point. The total time required for each unmanned platform to be assigned to each task group to execute the corresponding task is calculated. The unmanned platform-task group matching is performed according to the task assignment scheme with the minimum total time required. The path start point corresponding to the task group is used as the path start point for matching the unmanned platform.

[0038] Based on the mission execution status of each unmanned platform, the endurance constraints of the unmanned platform, and the loading capacity constraints of the transport vehicle, the distance greedy principle is used to determine the path termination location matched by the unmanned platform.

[0039] Step 43: Perform task layer initialization, and determine the matching tasks and task execution paths for the unmanned platform:

[0040] A weighted ranking value for tasks is constructed using the transfer cost of unmanned platforms and the task execution time. When matching tasks, random numbers are generated. If random number Less than the set first threshold If a candidate task is randomly selected from the task group as a pending follow-up task, otherwise the task with the smallest weighted ranking value is selected as the pending follow-up task. Calculate whether the remaining energy after completing the pending follow-up task can reach the nearest transport vehicle. If it can, the pending follow-up task is confirmed as a follow-up task. If not, determine to add a battery swapping task and return to the nearest transport vehicle to perform the battery swapping operation. For each specified follow-up task, update the weighted ranking value of the remaining tasks and then perform follow-up task allocation until all tasks are allocated and the task execution path is obtained.

[0041] Preferably, in step 43, the weighted ranking value for constructing the task using the unmanned platform transfer cost and task execution time is:

[0042] Let the population size be To set intervals generate Weight coefficients For the first When initializing an individual, select the first... Weight coefficients As a weighting coefficient for the cost of relocating unmanned platforms, As a weighting coefficient for task execution time; then the first task in the task group x Candidate tasks Weighted sort value for: ,in For candidate tasks The cost of transfer, For candidate tasks Task execution time.

[0043] Preferably, in step five, when performing population optimization, the individual crossover includes crossover selection at three levels: carrier layer, platform layer, and task layer.

[0044] Carrier layer crossover selection: for parent individuals and The set of starting and ending points of the offspring individuals is determined through cross-selection. and the set of path termination points The method is: to determine the or Recorded as ;Sure At that time, the parental individual The set of starting and ending points of the paths of the various unmanned platforms involved is denoted as . Parental individuals The set of starting and ending points of the paths of the various unmanned platforms involved is denoted as . ;Sure At that time, the parental individual The set of path termination points for all the unmanned platforms involved is denoted as Parental individuals The set of path termination points for all the unmanned platforms involved is denoted as ; and The set of common elements in is , and The union of the unique elements of the two is ;if The number of elements is the total number of transport vehicles. P Then the set ;like The element in is less than P Then the set , where set exist The selection method used in the middle is the round betting method. This represents taking the union of sets;

[0045] Platform-level cross-selection: Combining the allocation of unmanned platforms and candidate locations in the parent generation, the set of path start locations for offspring individuals is derived. Each candidate location element in the middle corresponds to the unmanned platform number in the parent generation, and forms a set. ;in, Candidate locations p The corresponding set of unmanned platform IDs;

[0046] set up for The number of unmanned platforms in China The maximum load capacity of the transport vehicle on the unmanned platform; according to , , Candidate locations are processed in order of priority.

[0047] for Candidate locations, offspring from candidate locations The unmanned platform that departed had the same serial number as its parent.

[0048] for Candidate locations, based on mission execution time The unmanned platforms were sorted and selected from the top ones. The unmanned platform with the shortest execution time and no conflicts between multiple candidate locations is assigned to the current candidate location of the offspring.

[0049] If only one candidate location exists, it meets the requirements. First, assign unmanned platform elements that do not conflict with multiple candidate locations to the current candidate location. If there are still unmanned platforms remaining after assignment, assign all remaining unmanned platforms to the current candidate location of the offspring.

[0050] If multiple locations exist, all of which meet the criteria First, allocate unmanned platform elements that do not conflict with multiple candidate locations; if there are still unmanned platforms remaining after allocation, assign the remaining unmanned platforms based on the task execution time and the loading capacity constraints of the transport vehicles.

[0051] In the platform-level cross-selection process, the set is updated after each candidate location is assigned. PL Elements in;

[0052] Task-level cross-selection: Based on the location assigned to the unmanned platform after platform-level cross-selection, determine the set of tasks in the parent generation that use that location as the starting or ending point of the path. The union of the two task sets is taken as the dominant gene set of the current unmanned platform. The dominant gene set is used to assign task execution sequences to the unmanned platform.

[0053] Preferably, in the task-layer cross-selection, allocating task execution sequences to the unmanned platform using the dominant gene set includes two stages:

[0054] In the first phase, the first task to be performed by each unmanned platform is to generate random numbers. If the random number Less than the set second threshold Then, unmanned platforms are randomly selected sequentially, and a task is chosen from the set of advantageous genes corresponding to the unmanned platform and added to the execution path of that unmanned platform as the first task; if the random number Not less than the second threshold Then, calculate the execution cost of the optional tasks in the set of advantageous genes corresponding to each unmanned platform, and add the task with the minimum execution cost to the task execution path of the corresponding unmanned platform as the first task;

[0055] In the second phase, the subsequent tasks for each unmanned platform are selected: a "first-come, first-served" strategy is adopted, allowing the unmanned platform that first completes its current task and enters an active state to select a subsequent task from its corresponding set of advantageous genes and add it to the task execution path; if the remaining energy of an unmanned platform is insufficient to execute the tasks in its set of advantageous genes, it continues to determine whether other unexecuted tasks can be executed. If so, the executable tasks are added to the task execution path; otherwise, the subsequent tasks are selected based on the set of offspring path termination locations. The candidate locations are selected and the battery swapping task locations are added to the task execution path, and the task selection process of the first and second stages is repeated. After each task selection, the elements in the set of advantages of each unmanned platform and the elements in the set of unexecuted tasks are updated.

[0056] Preferably, when performing population optimization in step five, a multi-granularity neighborhood search is used to perform local search of the solution space in the local search, including path adjustment based on task point exchange, path adjustment based on candidate location exchange, path adjustment based on task fragment exchange, and path adjustment based on task reallocation mechanism between paths.

[0057] Preferably, the path adjustment based on the inter-path task reallocation mechanism includes:

[0058] Calculate the distance between paths corresponding to each task execution sequence, and select the two paths with the shortest distance. and ;

[0059] In two paths and Randomly delete a set percentage of tasks;

[0060] Calculate the task execution time increment for each deleted task in the deleted task set when inserted into different positions in all current paths; select the deleted task with the smallest time increment and insert it into the corresponding position.

[0061] Update the insertion position, the remaining tasks in the deleted task set, and the corresponding task execution time increments, and continue to select the next deleted task with the smallest task execution time increment to insert into the corresponding position;

[0062] After completing the insertion operation of all deleted tasks, adjust or add energy replenishment locations for unmanned platforms that do not meet the endurance constraints.

[0063] Beneficial effects:

[0064] This invention addresses the task planning problem of energy-constrained multi-agent swarm operations in various task scenarios. It models the problem as a location-path planning problem with mobile carriers and energy replenishment. By combining task requirements and unmanned platform performance, it rationally selects the parking locations of transport vehicles and plans the energy replenishment and task execution paths for the unmanned platforms. While satisfying capability constraints, task requirements, and other relevant constraints, it minimizes the task completion time, thus disclosing a method for energy replenishment and task path optimization in mobile platform swarm operations. The main contributions of this invention are reflected in the following aspects:

[0065] (1) This invention analyzes and characterizes a variety of complex coupling constraints, fully considers the nonlinear and strong coupling characteristics in practical applications, and constructs a location-path planning problem model with a mobile carrier and energy supply.

[0066] (2) In a preferred scheme, in order to effectively balance the quality and diversity of the initial population, a hierarchical heuristic-assisted population initialization rule was designed, involving collaborative decision-making at the carrier layer, platform layer and task layer, and comprehensively considering task distribution, platform execution capability and workload balance; in order to effectively improve the quality of the planning scheme, the transfer cost of unmanned platform and task matching degree were introduced, and the rapid generation of high-quality initial population was realized.

[0067] (3) In an optimal scheme, in order to improve the quality of the population and the convergence speed, a crossover operator with vehicle location exploration and dominant gene rearrangement is proposed. The offspring individuals select and rearrange the superior genes of the parents and explore new task sequences and location schemes, which significantly improves the search efficiency of the algorithm in the complex solution space, accelerates the convergence speed of the algorithm, and enhances the population's ability to explore potential high-quality solutions.

[0068] (4) In a preferred solution, a multi-granularity neighborhood search structure was designed based on the characteristics of the problem, including path adjustment based on task point exchange, path adjustment based on candidate location exchange, path adjustment based on task fragment exchange, and path adjustment based on task redistribution mechanism between paths, thus realizing efficient collaboration between global and local search. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of the CLRP-MCER problem planning scheme of the present invention.

[0070] Figure 2 Flowchart of the enhanced Memetic algorithm for hierarchical initialization and dominant gene rearrangement.

[0071] Figure 3 This is a schematic diagram illustrating the individual coding of vehicles transporting CLRP-MCER problems under the condition that all vehicles are fully loaded.

[0072] Figure 4 This is a schematic diagram illustrating the individual coding of a CLRP-MCER problem transport vehicle when it is not fully loaded.

[0073] Figure 5 This is a schematic diagram of the individual decoding method in the CLRP-MCER problem.

[0074] Figure 6 A schematic diagram of population initialization rules for hierarchical heuristic assistance.

[0075] Figure 7 A schematic diagram of a crossover operator with vehicle location exploration and dominant gene rearrangement.

[0076] Figure 8 Diagram illustrating the assignment of candidate locations to unmanned platforms.

[0077] Figure 9 This is a schematic diagram of the task redistribution mechanism between paths. Detailed Implementation

[0078] This invention models the task allocation and path planning problem in energy-constrained multi-agent cluster operations under various task requirements as a variant of the location-path planning problem, namely, the location-path planning problem with mobile carriers and energy replenishment, which is called CLRP-MCER (Complete location-routing problem with mobile carriers and energy replenishment). Based on the Memetic algorithm, this invention introduces domain knowledge and proposes a method and apparatus for the collaborative optimization of energy replenishment and task path in heterogeneous mobile platform cluster operations.

[0079] For an abstract description of the above problems, see [link to relevant documentation]. Figure 1 The unmanned platform is mounted on a transport vehicle and transported to the mission area. The transport vehicle unloads the unmanned platform, which then completes its task set according to a planned path. During mission execution, if the unmanned platform's energy is low, it moves to the transport vehicle's docking point to perform a battery swap. Once all tasks are completed, the unmanned platform returns to the transport vehicle. Given a set of candidate locations, the transport vehicle's starting and stopping points are selected from these candidate locations.

[0080] See Figure 1The dashed lines in the figure represent the paths of the unmanned platform before and after energy replenishment. Due to the introduction of the inter-station operation required for battery swapping—the battery swapping task—coupling relationships arise between the task sets corresponding to each candidate location. The mobility of the transport vehicles means that the elements in the candidate location sets, which serve as the starting and ending points of the path, may differ. Furthermore, as the number of transport vehicles and candidate locations increases, the solution space of the CLRP-MCER problem in this invention rapidly expands. Simultaneously, as the capacity of the transport vehicles increases, more complex coupling relationships emerge between the task sets corresponding to each candidate location. These factors make effectively solving the CLRP-MCER problem quite difficult, while the proposed algorithm can effectively address these issues, exhibiting a significant advantage in task completion time compared to existing technologies.

[0081] The present invention will now be described in detail with reference to the accompanying drawings.

[0082] Step 1: Determine the candidate location, transport vehicle, unmanned platform, and mission information.

[0083] This step includes determining the location and number of candidate sites, the number and loading capacity of transport vehicles, the number of unmanned platforms, the mission execution capability matrix and endurance time vector, and the location and requirements of the mission. Specifically, the elements in the mission execution capability matrix of the unmanned platforms... for Type of unmanned platform performs tasks per unit time The amount completed.

[0084] This step also determines algorithm parameters, including population size and the number of times indicators are evaluated.

[0085] Step 2: Construct a task planning problem model.

[0086] A known number of heterogeneous unmanned platforms are deployed to designated locations by transport vehicles to perform scattered tasks with varying requirements. Within their endurance time, they return to the transport vehicles for energy replenishment or are transported to subsequent task areas after completing their tasks. Considering the different capabilities of various types of unmanned platforms in performing different types of tasks, it is necessary to optimize the task execution paths of each platform, including the start and end points of the paths and energy replenishment locations, to minimize the completion time of all tasks. The constraints and assumptions of the planning problem are as follows:

[0087] (1) Planning problem constraints

[0088] Each unmanned platform has a limited endurance and needs to return to the transport vehicle's docking point to replenish energy within the allotted time. Assume the unmanned platform... When fully charged, the battery capacity is ;

[0089] Each transport vehicle has a maximum loading capacity. ;

[0090] The task requirements cannot be broken down, and each unmanned platform cannot travel back and forth to the task location; they must leave after the task is completed.

[0091] Each candidate location may accommodate a maximum of one transport vehicle.

[0092] (2) Assumptions of the planning problem

[0093] Assuming that all unmanned platforms have the same volume, the number of platforms that each transport vehicle can carry is not affected by the platform type;

[0094] It is assumed that the endurance of each unmanned platform is fixed and depends only on the platform type, and is not affected by the task being performed.

[0095] More than The energy replenishment time (battery replacement) of multiple unmanned platforms at the same transport vehicle is negligible because the energy replenishment time is much shorter than the inter-regional travel time and mission execution time.

[0096] Assume that the transport vehicle will proceed to the next location after completing the unloading of all unmanned platforms (if the transport vehicle needs to change its parking location), and can arrive at the designated location before the unmanned platforms arrive.

[0097] (3) Description and modeling of unmanned platform capabilities

[0098] ① Unmanned platform task execution capability

[0099] Due to the influence of task type and scale, the time required for the same unmanned platform to execute different tasks varies; due to the influence of the unmanned platform's execution capability, the time required for the same task to be executed by different unmanned platforms also varies, as shown in formula (1.1), where, This is a matrix representing the amount of tasks completed by multiple unmanned platforms per unit time. for Type of unmanned platform performs tasks per unit time The amount, ,common Types of unmanned platforms M Type of task:

[0100] (1.1)

[0101] ②Endurance of unmanned platforms

[0102] During cluster operations, the time consumption of unmanned platforms mainly comes from two aspects: travel between different areas and task execution. Here, "area" is a general term for tasks and candidate locations (parking locations of transport vehicles). Due to the heterogeneity of unmanned platforms, the selection of transport vehicle locations, matching of transport vehicles and unmanned platforms, selection of unmanned platforms and task execution paths are mutually influential and coupled, and all four need to be jointly optimized.

[0103] (4) Task requirement modeling

[0104] Assuming an unmanned platform Arrive at the mission The time is ,Task The total demand is Then the following relationship holds:

[0105] (1.2)

[0106] in, For the number of tasks, The number of candidate locations. For the number of unmanned platforms, As a decision variable, if the platform is unmanned By region ( or Move to area ( or ),but ,on the contrary, . For the task The completion time. and Indicate candidate locations and .

[0107] (5) Modeling the mission planning problem of heterogeneous unmanned platforms

[0108] The CLRP-MCER problem can be defined on undirected graphs. Above, among which, For point sets, including task sets and candidate location set The latter can serve as both a location for unloading (loading) unmanned platforms from transport vehicles and a location for replenishing energy for each unmanned platform. Let be the edge set, where each element represents the distance traveled by the unmanned platform between different regions (including the mission area and candidate locations). The transport vehicle set is... The quantity is and Unmanned platforms are integrated into .

[0109] The mathematical model for the CLRP-MCER problem is as follows:

[0110] (1.3)

[0111] (1.4)

[0112] (1.5)

[0113] (1.6)

[0114] (1.7)

[0115] (1.8)

[0116] (1.9)

[0117] (1.10)

[0118] (1.11)

[0119] (1.12)

[0120] (1.13)

[0121] (1.14)

[0122] (1.15)

[0123] Formula (1.3) is the optimization objective of the problem. f This indicates minimizing the cluster job completion time. For the task The completion time.

[0124] Formulas (1.4)-(1.5) indicate that each transport vehicle stops independently, meaning that at most one transport vehicle can stop at each candidate location. Here, the set of transport vehicle stopping locations at the beginning of the operation is called the path starting location set. The set of locations where transport vehicles return after completing various tasks is called the path termination point set. , The elements in the list also serve as optional locations for the unmanned platform to perform battery swapping tasks. (Set of path start and end points) Set of path termination points The elements in the array may differ, but the quantity is always the same. This indicates the mobility characteristic of the transport vehicle. ( ) is the decision variable. If the transport vehicles are at the beginning (end) of the cluster operation... Located at candidate site ,but ( ),on the contrary, ( ); This represents the number of candidate locations.

[0125] Formula (1.6) indicates that there is a loading capacity constraint on transport vehicles; a transport vehicle can carry a maximum of [number] loads. One unmanned platform, among which ( ) is the decision variable. If the unmanned platform is at the beginning (end) of the cluster operation... By transport vehicle Loading, then ( ),on the contrary, ( ).

[0126] Formula (1.7) represents the candidate locations as the start and end points of the mission execution paths for each unmanned platform. and This is the set of candidate locations corresponding to two site selections for multiple transport vehicles. and These represent unmanned platforms. From candidate locations Departure and move to mission The incident, unmanned platform Completed the mission Return to candidate location If the above events occur, then the corresponding... =1 ( =1). The second formula in formula (1.7) indicates that the path termination point may be visited multiple times because there will be a battery swapping task, so the unmanned platform may return to the transport vehicle multiple times.

[0127] Formulas (1.8)-(1.10) stipulate that the unmanned platform must not consume all its energy during the journey, and that the unmanned platform must be fully charged after each energy replenishment. , and Unmanned platforms Battery consumption rate, maximum battery capacity, and reach area The remaining battery power at that time For unmanned platforms in the region ( or )and ( or The driving distance between ( ). Indicates unmanned platform Arrive at the mission Time; As a decision variable, if the platform is unmanned By region ( or Move to area ( or ),but ;on the contrary, .

[0128] Formula (1.11) demonstrates the unmanned platform By region Move to area In this situation, unmanned platforms Arrive at the mission Time The calculation method. For type unmanned platform Movement speed.

[0129] Constraint (1.12) indicates that the task must be fully executed and the task requirements cannot be divided, meaning that each unmanned platform cannot travel back and forth to the task location and must leave after the task is completed.

[0130] Equation (1.13) represents the flow balance constraint. and These represent unmanned platforms. From the region Move to task And in completing the task Move to the area after ;

[0131] Equation (1.14) is for eliminating constraints in sub-loops. and For the task and Index in the planned path.

[0132] Formula (1.15) represents the decision variables. It is a Boolean variable.

[0133] Table 1 describes the decision variables in the CLRP-MCER problem.

[0134] Table 1. Symbols and descriptions of decision variables

[0135]

[0136] Steps three through five below describe the enhanced Memetic algorithm for hierarchical initialization and dominant gene rearrangement designed in this invention.

[0137] The Memetic algorithm (MA) combines the powerful global search capability of genetic algorithms with the efficient local optimization capability of local search, making it a high-performance algorithmic framework that has been successfully applied to various complex optimization problems with good results. Unlike genetic algorithms, the mutation operation in the MA algorithm is guided by problem knowledge, which can effectively improve the algorithm's ability to explore complex solution spaces and its search efficiency. Therefore, based on the characteristics of the CLRP-MCER problem and the Memetic algorithm framework, this invention proposes an enhanced Memetic algorithm with hierarchical initialization and dominant gene rearrangement to efficiently solve the CLRP-MCER problem.

[0138] Step 3: Design the CLRP-MCER problem codec.

[0139] ① Encoding method:

[0140] The CLRP-MCER problem encoding includes carrier layer, platform layer, and task layer information, which encodes the individuals in the subsequent step population. The carrier layer information is the set of transport vehicle stopping locations, including the set of path start and end points. and the set of path termination points The platform layer information includes the matching relationship between the unmanned platform and candidate locations (path start and path end points); the task layer information includes the matching relationship between the unmanned platform and the task, as well as the task execution sequence. The tasks here include various tasks such as battery swapping.

[0141] In this embodiment, the following is adopted: Figure 3 and Figure 4 The coding method shown represents individuals. In the example, uppercase letters represent candidate location numbers, Roman numerals represent unmanned platform numbers, and Arabic numerals represent mission numbers. Figures 3-4 The introduction is as follows:

[0142] exist Figure 3 In this scenario, there are 3 transport vehicles and 6 unmanned platforms. The loading capacity of each transport vehicle is 2, therefore, all remote transport platforms are fully loaded. Candidate locations A / B / C constitute the initial set. Candidate locations B / C / E constitute the termination set. Unmanned platforms II and IV depart from candidate location A, sequentially execute tasks 1→13→15→14 and 5→6→7 respectively, and then return to candidate locations C and B. Element 0 represents energy replenishment. In this example, unmanned platform II replenishes energy at candidate location C before executing task 15 due to endurance limitations. Unmanned platforms I and VI depart from candidate location B, sequentially execute tasks 9→4→8 and 11→10→20→19 respectively, and then return to candidate locations B and E. Similarly, unmanned platform VI replenishes energy at candidate location B before executing task 20. Similarly, unmanned platforms III and V depart from candidate location C, sequentially execute tasks 12→3→2 and 16→17→18 respectively, and then return to candidate locations C and E.

[0143] exist Figure 4 In this scenario, there are 3 transport vehicles and 5 unmanned platforms. The loading capacity of each transport vehicle is 2. Except for the remote transport platform located at candidate location B, all other transport vehicles are fully loaded at the start of the mission. Candidate locations A, B, and C constitute the initial set. Candidate locations B / C / E constitute the termination set. The mission execution and energy replenishment status of the unmanned platform and Figure 3 Similarly, this will not be repeated here; since the transport vehicle located at candidate location B is not fully loaded, the empty position is filled by element NA.

[0144] ②Decoding method:

[0145] An event-triggered mechanism is used for individual decoding. Whenever an unmanned platform completes its current task, it becomes active and selects a subsequent task to execute based on the encoding scheme. When the encoding scheme in the task layer is "0", the unmanned platform moves to the corresponding candidate location to replenish its energy. Figure 5 The middle section shows Figures 3-4 The diagram shows the decoding of each encoding scheme. Each unmanned platform executes tasks sequentially according to the given encoding order. Task 14 and Task 4 are the last tasks completed in two different scenarios. The dashed lines represent the round-trip paths of the unmanned platform before and after energy replenishment, and the solid lines represent the paths of the unmanned platform leaving / returning to the candidate location and executing the task.

[0146] Step 4: Generate a high-quality initial population using hierarchical heuristic-assisted initialization rules: Initialize individuals in the order of carrier layer, platform layer, and task layer to generate the initial population.

[0147] See Figure 2The second step involves considering various factors, including candidate location selection, location-platform matching, platform-task allocation, and execution path planning. Therefore, heuristic methods designed considering only a single factor cannot generate high-quality execution plans for computational examples with different characteristics. Based on the above analysis, a layered heuristic-assisted initialization rule (LHa-IR) was designed, comprehensively considering task distribution, platform execution capabilities, and workload balancing. It mainly includes three decision-making aspects: the carrier layer, the platform layer, and the task layer. Figure 6 As shown, the details are as follows:

[0148] Step 41: Carrier Layer Initialization: Decision-making on the set of parking locations for transport vehicles, i.e., the starting path set for the unmanned platform. and path termination set .

[0149] To minimize energy consumption during unmanned platform movement between areas and shorten the overall operation time of the cluster, candidate locations with a relatively dense distribution of surrounding tasks should be prioritized to form the initial path set. and path termination set Therefore, for each task, the nearest candidate location is determined, and for each candidate location, the number of times the candidate location is the nearest candidate point for the task is counted. The number of times Reflects the density of task distribution around the candidate location; based on the number of times Determine the global selection probability of candidate locations ,frequency The higher the probability, the greater the global selection probability. The larger the probability, the better; then, based on the global selection probability of each candidate location. The method of round betting is used to determine the results. and The elements in the design aim to maximize the quality of the initial individuals while satisfying the need for diversity.

[0150] Global selection probability of each candidate location Calculated according to formula (4.1), where, Indicates distance from candidate location The number of recent tasks.

[0151] (4.1)

[0152] Step 42: Platform layer initialization: Matching unmanned platforms with candidate locations, including matching unmanned platforms with path start locations and unmanned platforms with path end locations.

[0153] Selection of the starting point of the unmanned platform path: based on the set determined in the task and step 41. The distances between elements in the algorithm cluster the tasks, with each task group corresponding to a path start point. The total time required for each unmanned platform to execute its corresponding task in each task group is calculated (ignoring endurance constraints and inter-task distances). Assume each task group is numbered as follows: ,but For unmanned platforms Task grouping The total time spent on each task As a decision variable, if the platform is unmanned Task grouping Each task in the middle, then ,on the contrary, The number of unmanned platforms loaded on each transport vehicle must not exceed the upper limit, and each unmanned platform can only be assigned to one task zone.

[0154] According to formula (4.2), the assignment of multiple unmanned platforms should follow the principle of minimizing the total time required. That is, the unmanned platform-task group matching should be carried out according to the task assignment scheme with the minimum total time required, and the starting point of the path corresponding to each task group should be used as the starting point of the path matching for that unmanned platform.

[0155] (4.2)

[0156] (4.3)

[0157] (4.4)

[0158] The selection of the unmanned platform's path termination location: Based on the mission execution status, combined with the unmanned platform's endurance constraints and the transport vehicle's loading capacity constraints, a distance-greedy principle is used to determine the termination location. One specific operation is as follows: When deciding on the path termination location, the unmanned platform adopts the "first-come, first-served" principle, meaning the unmanned platform that completes the last task on the path first has priority in selecting the path termination location. Considering the transport vehicle's loading capacity constraints, each time the unmanned platform selects a path termination location, within the transport vehicle's capacity limits, the distance-greedy principle is used to match the available candidate locations.

[0159] Step 43: Task layer initialization: Decision-making unmanned platform matching tasks and task execution paths.

[0160] To effectively improve the quality of the planning scheme, the transfer cost (TC) and capability matching degree (AM) of unmanned platforms are introduced. The transfer cost (TC) refers to the time required for the unmanned platform to reach the next task. In order to shorten the task completion time as much as possible, the task execution path of each unmanned platform needs to be optimized. Therefore, the movement distance between the unmanned platform and the task can be used as a reference factor when selecting tasks. The capability matching degree (AM) mainly includes the task execution time, which refers to the time required for the unmanned platform to perform the task.

[0161] During the decision-making process, the active unmanned platform selects the subsequent task to be executed or the candidate return location. The priority of the candidate tasks is determined by the transfer cost (TC) and capability matching degree (AM) of the unmanned platform. of The weighted and sorted approach is used to better balance the transfer costs and requirements of different tasks.

[0162] Specifically, during the task-level initialization phase, the unmanned platform transfer cost (TC) and the capability matching degree (AM) expressed as task execution time are sorted in ascending order of weighted value. When matching tasks, random numbers are generated. If random number Less than the set first threshold If a candidate task is randomly selected from the task group as a pending follow-up task, then the task with the smallest weighted ranking value is selected as the pending follow-up task. When the unmanned platform selects each follow-up task, it calculates whether the remaining energy after completing the pending follow-up task can reach the nearest transport vehicle. If it can, the pending follow-up task is confirmed as a follow-up task; otherwise, a battery swapping task is added, and the platform first returns to the nearest transport vehicle for battery swapping. For each designated follow-up task, the weighted ranking value of the remaining tasks is updated, and the execution location of the follow-up task or battery swapping task is assigned again until all tasks are assigned and the task execution path is obtained.

[0163] In a preferred embodiment, the weighted sorting value is set as follows:

[0164] Assuming the population size is To set intervals generate Weight coefficients ,interval The size is When generating corresponding individuals, appropriate weight coefficients are used respectively, such as when generating the first individual. When there are individual entities, the weighting coefficient is Then the weighting factor for the transfer cost TC is The weighting coefficient for capability matching degree AM is: That is, in generating individuals At that time, candidate tasks weighted value The calculation method is as follows: Then, according to Arrange the tasks in ascending order to determine the order of each task. When matching tasks, if the random number does not meet the conditions, then select... The smallest task is assigned as the next task.

[0165] Step 5: Utilize the cultural gene algorithm to optimize the population and output a collaborative planning scheme for energy replenishment and task paths for multi-agent cluster operations.

[0166] The inventive point in this step lies in the design of a crossover operator with vehicle location exploration and dominant gene rearrangement, as well as the use of multi-granularity neighborhood search in local search.

[0167] See Figure 2 This step specifically includes the following steps:

[0168] Step 51: Use a crossover operator with vehicle location exploration and dominant gene rearrangement to generate offspring individuals and explore the complex solution space.

[0169] Offspring generation is a crucial step in evolutionary computation. Effective gene inheritance and spatial exploration can enhance population diversity while ensuring individual quality. Therefore, it is necessary to design efficient offspring generation rules based on problem characteristics to obtain high-quality offspring individuals.

[0170] Based on the characteristics of the CLRP-MCER problem, this invention proposes a crossover operator with vehicle location exploration and superior gene rearrangement (VleSGr-C). It should be noted that VleSGr-C does not directly inherit gene sequences from the parent generation, but rather selects and rearranges superior genes from the parent generation, and explores new task sequences and location schemes.

[0171] VleSGr-C is mainly discussed from three aspects: carrier layer, platform layer, and task layer. Now, let's take... Figure 7 For example, the following is an introduction, in which and Each represents a parent individual:

[0172] Carrier layer cross selection: Taking the cross selection of the path start location set as an example, based on whether the path start positions are the same, the parent generation is selected. The set of starting and ending points of the path and father generation The set of starting and ending points of the path The elements in the database are divided into two types: and , respectively represent and The set of common elements in the two sets and the union of their unique elements. If , The initial set of offspring individuals is represented by the number of elements in the set. ;like ,but ,in In the set The selection process uses a round-robin betting method, and the set... The probability of choosing each element and Proportional and satisfying . The method for determining each element is similar to the process described above.

[0173] like Figure 7 As shown, ,and Therefore, it is necessary to The middle selects two elements through a round betting method, and... Together constitute Assuming the selected element is ,but Similarly, we can conclude that... .

[0174] Platform-level cross-selection: Combining the allocation of unmanned platforms and candidate locations in the parent generation, a set is derived. Each candidate location element in the middle corresponds to the unmanned platform number in the parent generation, and forms a set. ,in, Candidate locations p The corresponding set of unmanned platform IDs.

[0175] set up for The number of unmanned platforms in China This represents the maximum load capacity of the transport vehicle on the unmanned platform. To inherit the best traits from previous generations, the processing priority for matching candidate locations with unmanned platforms is as follows: [Prioritizing locations with higher priority]. ,Again ,at last Priority order.

[0176] for If a candidate location is selected, then the offspring will be selected from the candidate locations. The unmanned platform that departed had the same serial number as its parent.

[0177] for The candidate locations are then selected based on the mission execution time. Sorting the unmanned platforms by number, select the top one. The unmanned platform with the shortest total execution time and no conflicts between multiple candidate locations is assigned to the current candidate location of its offspring. No conflicts between multiple candidate locations means that an unmanned platform that has already been assigned to a candidate location cannot be assigned to other unmanned platforms.

[0178] For example in Figure 7 In, set The set of unmanned platform IDs corresponding to each element in the parent generation is: For candidate locations For example, the optional platform number is It was finally determined to be .

[0179] If only one candidate location exists, it meets the requirements. First, assign unmanned platform elements that do not conflict with multiple candidate locations to the current candidate location; if there are still unmanned platforms remaining after assignment, assign all remaining unmanned platforms to the current candidate location of the offspring.

[0180] If multiple locations exist, all of which meet the criteria First, allocate unmanned platform elements that do not conflict with multiple candidate locations; if unmanned platforms remain after allocation, then assign the remaining unmanned platforms based on the task execution time and transport vehicle loading capacity constraints, such as... Figure 8 As shown.

[0181] It should be noted that after each candidate location is assigned, it must be done according to the following... Figure 8 Update the collection as shown The elements in.

[0182] Task-level cross-selection: This method combines the dominant genes identified in the parent generation with the rearrangement of these gene sequences to improve the quality of offspring while effectively inheriting the superior genes from the parents. The dominant genes of each unmanned platform are the union of the task numbers corresponding to the candidate locations matched with that platform in the parent generation. For example, for an unmanned platform... In offspring From candidate locations Departure, and the father generation and Candidate Locations The sets of tasks corresponding to the starting or ending points of the path are respectively and Then its dominant gene set is the union of the two, that is... .

[0183] The system utilizes a set of dominant genes to assign task execution sequences to unmanned platforms. In a preferred embodiment, the allocation of offspring individual paths, i.e., task execution sequences, is divided into two stages:

[0184] In the first phase, the first task to be executed by each unmanned platform is selected: at the beginning of the path construction for this phase, random numbers are generated. ,like If so, then randomly select unmanned platforms in sequence, and select a task from the set of advantageous genes corresponding to the unmanned platform to add to the execution path of the unmanned platform as the first task; otherwise, calculate the execution cost corresponding to the selectable task in the set of advantageous genes corresponding to each unmanned platform. The execution cost can be calculated by the sum of the movement time of the task and the execution time of the task; and add the task with the minimum execution cost to the execution path of the corresponding unmanned platform as the first task.

[0185] The second phase involves selecting subsequent tasks for each unmanned platform. In this phase, each platform has tasks to execute, and a "first-come, first-served" strategy is adopted. The platform that first completes its current task and enters an active state selects a subsequent task from its corresponding set of advantageous traits and adds it to its task execution path. If the platform's remaining energy is insufficient to execute the tasks in its set of advantageous traits, it continues to determine if other unexecuted tasks can be executed. If so, the executable task is added to the task execution path; otherwise, the remaining energy is used to determine the next task based on the set of advantageous traits. The candidate locations for the battery swapping task are selected and added to the task execution path, and the task selection process in the above two stages is repeated. After each task selection, the elements in the set of advantageous genes of each unmanned platform and the elements in the set of unexecuted tasks need to be updated.

[0186] Once the specified number of offspring are obtained, proceed to step 52.

[0187] Step 52: Use multi-granularity neighborhood search to perform local search of the solution space.

[0188] A single neighborhood structure can easily cause the algorithm to get stuck in a local optimum, while multiple neighborhood structures working together can effectively enhance the algorithm's exploration ability. Therefore, in the algorithm proposed in this invention, a multi-granularity neighborhood search method is designed from different perspectives, including path adjustment based on task point exchange, path adjustment based on candidate location exchange, path adjustment based on task fragment exchange, and path adjustment based on task redistribution mechanism between paths.

[0189] ① Path adjustment based on task point exchange: given a solution Choose any two tasks from a randomly selected task execution path. and Exchange positions and adjust or add energy replenishment locations for unmanned platforms that do not meet endurance constraints.

[0190] ② Path adjustment based on candidate location exchange: A roulette wheel selection method is used, with each candidate location selected from the set... and Add a candidate location. Due to limitations on the number of transport vehicles, the collection... and Each candidate location must be removed again using a roulette wheel selection method. The paths to the removed candidate locations (starting point, ending point, or refueling location) are then adjusted, and refueling locations are adjusted or added for unmanned platforms that do not meet the endurance constraints.

[0191] ③ Path adjustment based on task fragment swapping: given a solution Randomly select two task execution paths, and choose any two tasks. and The gene fragments were exchanged, and energy replenishment locations were adjusted or added for unmanned platforms that did not meet the endurance constraints.

[0192] ④ Path adjustment based on inter-path task reallocation mechanism: Considering that two paths with smaller spatial distances are more likely to undergo effective task adjustment, calculate the distance between each path and select the two closest paths. and ; on two paths and A certain percentage of tasks are randomly deleted; then a task redistribution mechanism between paths is used to repair each path.

[0193] The distance between each path can be calculated using the following formula (5.1). Where, and Representing paths and The number of tasks.

[0194] (5.1)

[0195] like Figure 9 As shown, the path adjustment method based on the inter-path task reallocation mechanism is as follows:

[0196] First, calculate the task execution time increment for each task in the deleted task set inserted into different positions along all paths, i.e., the difference between the distance moved before and after task insertion and the task execution time. Then, select the task with the lowest cost to insert into the corresponding position. Finally, update the insertable positions, the remaining tasks in the deleted task set, and their corresponding task execution time increments, and continue selecting the next deleted task with the lowest task execution time increment to insert into the corresponding position. For example... Figure 9 As shown, inserting task number 4 into position 3 of the path of unmanned platform II yields the minimum time increment. After completing the insertion of all deleted tasks, adjust or add energy replenishment positions for unmanned platforms that do not meet the endurance time constraint.

[0197] Step 53: Update the population and the best individuals using a biased population evaluation mechanism while maintaining population diversity.

[0198] Step 54: Determine whether the algorithm termination condition is met. If not, return to step 51. If yes, output the multi-agent cluster operation energy replenishment and task path planning scheme, i.e., the location path planning scheme of transport vehicle-unmanned platform.

[0199] This concludes the process.

[0200] Based on differences in overall planning scenarios and coupling relationships, the present invention makes specific improvements to the energy replenishment and task path optimization method for mobile platform cluster operations in the following aspects:

[0201] A hierarchical heuristic-assisted population initialization rule was designed, taking into account task distribution, unmanned platform capabilities and endurance constraints. By introducing transfer costs and task matching degree, a high-quality initial population was generated rapidly.

[0202] A crossover operator with vehicle location exploration and dominant gene rearrangement is proposed. By selecting and rearranging superior parent genes and exploring new gene sequences and location schemes, the algorithm's search efficiency in complex solution spaces is improved.

[0203] By adopting a multi-granularity neighborhood search structure, the algorithm achieves synergy between global and local searches, effectively improving its ability to explore the solution space.

[0204] The specific embodiments described above only illustrate the design principles of the present invention. The shapes and names of the components in this description may differ and are not limited. Therefore, those skilled in the art can modify or make equivalent substitutions to the technical solutions described in the foregoing embodiments; and these modifications and substitutions do not depart from the inventive spirit and technical solutions of the present invention, and should all fall within the protection scope of the present invention.

Claims

1. A method for energy replenishment and mission path optimization of a mobile platform cluster operation, for a multi-type task demand scenario: multiple transportation vehicles carrying unmanned platforms, one transportation vehicle can carry one or more unmanned platforms; the transportation vehicle transports the unmanned platform to a candidate location, the unmanned platform leaves the transportation vehicle to perform a task, and in the process of task execution, the unmanned platform can be replenished with energy at the candidate location transportation vehicle, and after completing the task, it arrives at the candidate location to get on the vehicle and is taken away by the transportation vehicle; heterogeneous unmanned platforms and transportation vehicles are not tightly coupled; characterized in that, The method comprises: Step 1: determining information of candidate sites, transport vehicles, unmanned platforms and tasks; Step 2: modeling a task planning problem, planning constraints comprising: The starting points of the transport vehicles constitute the set of route start points. The locations where transport vehicles return after completing various tasks form the path termination points. This set of elements also serves as the optional locations for the unmanned platform to perform battery swapping tasks; the set of path start and end points. and the set of path termination points Elements in the list are selected from candidate locations; the set of path start locations. and the set of path termination points The elements in the text are the same or different; unmanned platforms cannot consume all energy while driving, and are full of electricity after each energy supply; Step three: encode the carrier layer, platform layer and task layer information into individuals; the carrier layer information is a set of transportation vehicle stopping places, including a set of path starting places and a set of path ending places ; the platform layer information is the matching relationship between the unmanned platform and the path starting place and the path ending place; the task layer information includes the matching relationship between the unmanned platform and various tasks including the battery replacement task and the task execution sequence; ​ Step 4: initializing individuals in order of carrier layer, platform layer and task layer to generate an initial population; specifically comprising: Step 41: Perform carrier layer initialization, determine the path starting set and the path ending set of the unmanned platform Step 42: For each task, determine the nearest candidate location, and for each candidate location, count the number of times the candidate location is the nearest candidate location for the task Step 43: The number of times reflects the task distribution density around the candidate location; the higher the number of times , the greater the global selection probability of the candidate location Step 44: According to the global selection probability, determine the elements in the path starting set and the path ending set respectively by using the roulette method Step 45: According to the global selection probability, determine the elements in the path starting set and the path ending set respectively by using the roulette method ​ Step 42: performing platform layer initialization to decide assignment of unmanned platforms to candidate sites, including assignment of unmanned platform-path starting sites and unmanned platform-path ending sites: a set of path starting locations determined according to the tasks and the steps 41 The distance between each element groups all tasks, and each task group corresponds to a path starting location; the total time required for each unmanned platform to be assigned to each task group to perform the corresponding task is calculated respectively, and the unmanned platform-task group matching is performed according to the task assignment scheme with the minimum total time required, and the path starting location corresponding to the task group is taken as the path starting location matched by the unmanned platform; combining task execution conditions of each unmanned platform, endurance constraints of the unmanned platform and loading capacity constraints of the transport vehicle, and using a distance greedy principle to determine path ending sites matched with the unmanned platform; Step 43: performing task layer initialization to decide matched tasks and task execution paths of the unmanned platform: A weighted ranking value of a task is constructed by using a transfer cost and a task execution time of an unmanned platform, a random number is generated when tasks are matched , if the random number is less than a set first threshold value , a candidate task is randomly selected in a task group as a pending subsequent task, otherwise, a task with the smallest weighted ranking value is selected as the pending subsequent task; whether the remaining energy after executing the pending subsequent task can reach the nearest transport vehicle is calculated, if yes, the pending subsequent task is confirmed as a subsequent task, if not, it is determined to add a battery replacement task, and the nearest transport vehicle is returned to perform a battery replacement operation; the weighted ranking value of a remaining task is updated every time a subsequent task is specified, and subsequent task allocation is performed again until all task allocation is completed, and a task execution path is obtained; Step 5: performing population optimization to output a cluster operation energy supply and task path collaborative planning scheme.

2. The method of claim 1, wherein, In the step 1, the information of the candidate sites, the transport vehicles, the unmanned platforms and the tasks is: determining positions and quantities of the candidate sites, quantities and loading capacities of the transport vehicles, quantities of the unmanned platforms, a task execution capability matrix and an endurance time vector of the unmanned platforms, positions and demands of the tasks; Elements in a mission execution capability matrix of an unmanned platform To Type of unmanned platform Amount of mission completed per unit time.

3. The method of claim 1, wherein, In the step 2, a total optimization target of the task planning problem is to minimize an end time of the cluster operation.

4. The method of claim 1, wherein, In the step 2, the planning constraints further comprise: Transport vehicle loading capacity constraint: a transport vehicle can load at most one unmanned platform; Unmanned platform Arrival task Time at Satisfies the following: In the formula, For type unmanned platform The speed of travel; For unmanned platforms In the region and The distance traveled between locations; M represents the total number of tasks, and L represents the total number of candidate locations. This represents the total number of unmanned platforms. task demands cannot be split, and each unmanned platform cannot go back and forth at a task and needs to leave after completing the task; each transport vehicle is independently parked, that is, at most one transport vehicle can be parked at each candidate site.

5. The method of claim 1, wherein, In the step 43, the weighted ordering value of the task is constructed using a transfer cost of the unmanned platform and a task execution time, and the weighted ordering value of the task is: Let the population size be At a set interval Generate weighting coefficients For the initialization of the first individual, select the first weighting coefficient as the weighting coefficient of the transfer cost of the unmanned platform, and as the weighting coefficient of the task execution time; then the weighted ranking value of the xth candidate task in the task group is , wherein is the transfer cost of the candidate task , and is the task execution time of the candidate task .​ 6. The method of claim 1, wherein, The step 5 uses a cultural gene algorithm to perform population optimization, and when performing population optimization operations, individual crossover comprises cross selection of three levels of the carrier layer, the platform layer and the task layer; Carrier layer cross selection: for parent individuals and , the path starting point set and the path ending point set of the offspring individuals are determined by cross selection in the following way: the to-be-determined or is recorded as ; when is determined, the path starting point set of each unmanned platform involved in the parent individual is recorded as , and the path starting point set of each unmanned platform involved in the parent individual is recorded as ; when is determined, the path ending point set of each unmanned platform involved in the parent individual is recorded as , and the path ending point set of each unmanned platform involved in the parent individual is recorded as ; The common element set in and is , The union of the unique elements of and is ; if the number of elements in is the total number of transport vehicles P, then the set ; if the number of elements in is less than P, then the set , where the set is selected by round gambling in , represents taking the union set; Platform layer cross selection: combined with the allocation of unmanned platforms-candidate sites in the parent, the path starting site set of the child individual is obtained Each candidate site element in the parent corresponds to the unmanned platform number, and constitutes a set ; wherein, is the unmanned platform number set corresponding to the candidate site p; Set For The number of unmanned platforms in The maximum loading capacity of the transport vehicle for the unmanned platform; according to , , The priority order of each candidate site is processed; For a candidate location of the child, the number of unmanned platforms starting from the candidate location of the child is the same as that of the parent. for Candidate locations, based on mission execution time The unmanned platforms were sorted and selected from the top ones. The unmanned platform with the shortest execution time and no conflicts between multiple candidate locations is assigned to the current candidate location of the offspring. If there is only one candidate location, the following conditions are met If there is only one candidate location, the following conditions are met If there is only one candidate location, the following conditions are met If there are multiple locations, all of which are consistent then first assign the UAV elements for which there are no multiple candidate locations; if there are still UAVs remaining after assignment, then assign the remaining UAVs in combination with the task execution time and transport vehicle load capacity constraints; In the platform layer cross selection, elements in a set PL are updated after completing assignment of each candidate site; Task layer cross selection: according to sites assigned to the unmanned platform after the platform layer cross selection, a task set in which each site is used as a path starting site or a path ending site is determined in the parent generation, and a union set of the two task sets is used as an advantage gene set of the unmanned platform; the advantage gene set is used to assign a task execution sequence to the unmanned platform.

7. The method of claim 6, wherein, In the task layer cross selection, assigning the task execution sequence to the unmanned platform using the advantage gene set comprises two stages: In the first stage, a first task to be executed by each unmanned platform is selected: generating a random number If the random number is less than a second threshold value , the unmanned platforms are randomly selected in turn, and a task is selected from the corresponding set of dominant genes of the unmanned platform and added to the execution path of the unmanned platform as the first task. If the random number is not less than the second threshold value , the execution cost of the tasks in the corresponding set of dominant genes of each unmanned platform is calculated, and the task with the minimum execution cost is added to the execution path of the corresponding unmanned platform as the first task. In the second stage, a subsequent task to be executed by each unmanned platform is selected: a "first completion first selection" strategy is adopted, and the unmanned platform that first completes the current task enters an active state, selects a subsequent task from the corresponding advantage gene set, and adds the subsequent task to the task execution path; if the remaining energy of the unmanned platform is insufficient to execute the tasks in the advantage gene set, it is determined whether other unexecuted tasks can be executed, if yes, the executable tasks are added to the task execution path, otherwise, a candidate location in the offspring path termination location set is selected to add a battery replacement task location to the task execution path, and the task selection process of the first stage and the second stage is repeatedly continued; after each task selection, the elements in the advantage gene set of each unmanned platform and the elements in the unexecuted task set are updated.

8. The method of claim 1, wherein, The step 5 uses the cultural gene algorithm to perform population optimization, and in local search, a multi-granularity neighborhood search is used to search locally in a solution space, including path adjustment based on task point exchange, path adjustment based on candidate site exchange, path adjustment based on task segment exchange and path adjustment based on a task reallocation mechanism between paths.

9. The method of claim 8, wherein, The path adjustment based on the task reallocation mechanism between paths comprises: Calculate the path distance corresponding to each task execution sequence, select the two paths with the shortest distance and ; in two paths and randomly deleting a set proportion of tasks; calculating task execution time increments corresponding to insertion of each deleted task in the deleted task set into different positions of all current paths respectively; and selecting a position corresponding to insertion of a deleted task with the smallest time increment. The remaining tasks in the deleted task set and the corresponding task execution time increment are updated, and the next task with the minimum task execution time increment is selected for insertion into the corresponding position; After completing the insertion operation of all deleted tasks, the energy supplement position is adjusted or added for the unmanned platform that does not meet the endurance time constraint.

Images