Heterogeneous unmanned ship task dynamic optimization method based on intelligent distributed CBBA

Abstract

The invention relates to a heterogeneous unmanned ship task dynamic optimization method based on intelligent distributed CBBA, and relates to the technical field of unmanned autonomous cooperative control, the method comprises the following steps: firstly, using an intelligent distributed CBBA algorithm to realize task initial allocation, and constructing a cost function integrating distance cost, time cost and time window default penalty; and a task sequence is adjusted through a greedy strategy so as to meet strict time window constraints. And secondly, in the task execution process, a real-time monitoring and dynamic re-planning mechanism is established, and emergencies can be rapidly dealt with. A redistribution mechanism is introduced, the weight of a cost function is dynamically adjusted through a genetic algorithm, and loss and scheme optimization benefits caused by task execution subject change are balanced. Compared with the prior art, the task execution efficiency, reliability and adaptability of the heterogeneous unmanned ship cluster in a complex dynamic environment are effectively improved, and the method is suitable for complex application scenes such as maritime search and rescue, material delivery, patrol monitoring and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned autonomous cooperative control, and particularly relates to a heterogeneous unmanned ship task dynamic optimization method based on intelligent distributed CBBA. BACKGROUND

[0002] With the development and application of unmanned ship cluster technology, multi-unmanned ship cooperative execution of complex tasks is increasingly popular in military and civilian fields. For example, in the scenarios of maritime patrol, emergency rescue, and material supply, the cooperative operation of unmanned ships with different functions can greatly improve the efficiency and reliability of the task. However, the existing task allocation and scheduling methods still face many challenges in the dynamic task optimization of heterogeneous unmanned ship clusters. First, traditional methods often use centralized or static planning, which does not fully consider the performance differences (such as speed, load, and power) of heterogeneous unmanned ships and the time requirements of tasks, making it difficult to adapt to changes in the task environment in a timely manner. Second, the multi-task allocation problem is usually an NP-hard problem, and it is very difficult to solve the optimal solution in real time when the number of tasks and the size of unmanned ships increase, so more efficient algorithms are needed. Third, in the actual execution process, unmanned ships may encounter sudden situations such as failure, addition, or task changes, and the existing solutions lack a mechanism for dynamic re-planning during task execution, resulting in task failure or reduced efficiency.

[0003] There are already some research results on unmanned ship / unmanned aerial vehicle task allocation. For example, the consensus bundle algorithm (CBBA) as a distributed task allocation algorithm is widely used in task scheduling of multi-agent systems. However, the basic CBBA algorithm usually assumes that the number of tasks and participants is fixed, and does not fully consider the time window constraints of tasks and dynamic events in the execution process, and its adaptability to heterogeneous unmanned platforms is also limited. To solve the problem of unmanned aerial vehicle task package construction conflict, Chinese patent CN116149370A attempts to improve the CBBA algorithm through clustering, allowing unmanned aerial vehicles to preferentially select tasks that are closer to them to construct task packages and improve efficiency, but the clustering results are limited to distance. Chinese patent CN116993073A introduces time window constraints into CBBA to improve the algorithm's ability to handle task timing requirements, but it still does not solve the problem of task re-allocation in dynamic environments. Chinese patent CN116911535A introduces a distance reward and penalty factor and a time window constraint indicator variable to construct a cost function to allocate task packages, which improves the adaptability to time window constraint scenarios, but the task re-allocation mechanism in dynamic environments is not perfect, and the ability to respond to sudden situations such as unmanned ship failure and new tasks is insufficient.

[0004] In addition, there are documents that apply machine learning or meta-algorithms to unmanned vehicle task optimization. A related patent discloses a water surface unmanned vehicle cluster patrol task allocation method, which solves the patrol path by improving the genetic algorithm and adopts a voting mechanism to select the optimal scheme, and then re-plans the remaining task route according to the surrounding unmanned vehicle navigation cost. However, this method is mainly aimed at a single patrol task scenario, lacks adaptation to complex scenarios such as multiple types of tasks and unmanned vehicle failures, and is difficult to meet the collaborative task requirements of heterogeneous unmanned ship clusters in dynamic and variable environments. Chinese patent CN119831218A uses a nearest neighbor clustering algorithm based on a greedy strategy to cluster and group and construct a task cluster package, combines a clustering consensus consistency package algorithm, considers the kinematics of unmanned vehicles and task coupling constraints, but has problems such as insufficient task re-allocation efficiency and flexibility in dynamic task environments, imperfect unmanned vehicle failure handling mechanisms, and the like, which affect task continuity and reliability.

[0005] In summary, the existing technology still has deficiencies in the intelligentization and dynamic optimization of heterogeneous unmanned ship cluster task allocation. For example, there are problems such as lack of strict constraints on task time limits and execution order, lack of efficient re-allocation strategies in emergency situations, and insufficient use of the scalability and fault tolerance of distributed algorithms, which need to be solved. Therefore, it is necessary to provide a new technical solution to overcome the above deficiencies and realize intelligent distributed dynamic optimization of heterogeneous unmanned ship tasks. SUMMARY

[0006] The purpose of the present application is to provide a heterogeneous unmanned ship (Unmanned Surface Vehicle, USV) task dynamic optimization method based on intelligent distributed CBBA (Consensus-Based Bundle Algorithm), which uses an improved consensus bundle algorithm (CBBA) and intelligent optimization strategies to achieve efficient allocation and dynamic optimization of tasks for a heterogeneous unmanned ship cluster with multiple functions such as reconnaissance ships, supply ships, evacuation ships, and general-purpose ships.

[0007] The purpose of the present application can be achieved by the following technical solutions:

[0008] A heterogeneous unmanned ship task dynamic optimization method based on intelligent distributed CBBA, comprising the following steps:

[0009] S1, initialize data and parameters: obtain the state data and task environment data of each unmanned ship in the heterogeneous unmanned ship cluster, preprocess the data, and distribute the processed task parameters to each unmanned ship through encrypted communication;

[0010] S2, task pre-allocation: define a cost function, based on the intelligent distributed CBBA algorithm, cluster all tasks to be executed according to the number of unmanned ships, and the corresponding unmanned ship preferentially bids for the tasks in each subgroup; based on the cost function, the cost of each unmanned ship executing the candidate task is calculated, and the task package is constructed by initially selecting the task combination with the lowest cost; the tasks are bid by communication between each unmanned ship, and the iteration is performed until the initial task allocation scheme without conflict is obtained;

[0011] S3, time window adjustment: check the time window constraints of each task in the initial task allocation scheme, and re-allocate the tasks that exceed the time window requirements using the greedy strategy to meet the time window constraints of all tasks;

[0012] S4, conflict resolution: when multiple unmanned ships bid for the same task in the initial task allocation scheme, the conflict resolution rule is used to resolve the conflict; repeat the bidding and conflict resolution until all tasks are uniquely assigned, and obtain a task allocation scheme that meets the constraint conditions, and calculate the global objective function corresponding to the scheme;

[0013] S5, scheme optimization iteration: the weight coefficients of the cost function are evolved and adjusted using the genetic algorithm, and the task allocation scheme is iteratively optimized by repeating steps S2-S4 until the preset iteration number is reached or the global objective function converges, and the optimal task allocation scheme after optimization is output;

[0014] S6, task execution monitoring: control each unmanned ship to execute the task according to the optimal task allocation scheme instruction, and monitor the task execution state in real time, detect the emergency event, and execute step S7 when any emergency event is detected;

[0015] S7, state update and task pool reconstruction: update the available list of unmanned ships and the task pool according to the type of the emergency event;

[0016] S8, task re-allocation: based on the updated task pool and the state of the unmanned ship, the intelligent distributed CBBA algorithm is re-executed to allocate all uncompleted tasks, and a new task allocation scheme is obtained, wherein an additional re-allocation cost penalty is given to the task whose execution unmanned ship changes to reduce the loss caused by replacing the execution subject; the new task allocation scheme is issued to the corresponding unmanned ship to continue executing the remaining tasks;

[0017] S9, cycle until completion: real-time monitor the task execution state, repeat the steps of emergency event detection, state update and task pool reconstruction, and task re-allocation, process multiple emergency events occurring in time sequence in turn, until all tasks are completed.

[0018] The unmanned ship state data includes the current position, navigation speed, remaining power and completed task quantity of the unmanned ship.

[0019] The environmental data includes temperature, humidity, wind speed and wind direction of the working sea area.

[0020] The preprocessing includes filling of missing data, removal of abnormal data and cleaning of repeated data, and standardization or normalization conversion of the unmanned ship state data and environmental data.

[0021] The cost function includes distance cost, time cost and time window penalty, the distance cost is determined based on the time / energy cost corresponding to the sailing distance of the unmanned ship from the current location to the execution task, the time cost is determined based on the execution time and waiting time consumed by the task itself, and the time window penalty is defined as: if the required completion time of the task exceeds the specified time window, the penalty value is obtained by multiplying the exceeding time by a large penalty coefficient; the distance cost, time cost and time window penalty are combined by weighted linear combination to form the total cost function.

[0022] The clustering grouping of all to-be-executed tasks according to the number of unmanned ships specifically includes the following steps:

[0023] Multi-modal feature engineering and embedding are performed on all to-be-executed tasks and unmanned ships, and heterogeneous information is fused to generate task feature representation;

[0024] Based on the task feature representation, an HDBSCAN adaptive density clustering algorithm is used to cluster the tasks, wherein the HDBSCAN adaptive density clustering algorithm first constructs a minimum spanning tree of the data, then creates a series of interrelated cluster hierarchy structures by deleting edges with relatively higher weights in the minimum spanning tree, and adaptively identifies task clusters of different shapes and densities based on cluster stability analysis; after generating the task clusters, the task clusters are matched and optimized according to the task categories and the performance of the unmanned ships; the assignment relationship of the task clusters to the unmanned ships is optimized through multi-objective dynamic optimization, and one or more task clusters are determined as the priority task set of each unmanned ship until the assignment result is stable.

[0025] The intelligent distributed CBBA algorithm performs the following steps to realize initial task allocation: each unmanned ship selects a set of tasks with the lowest cost and meeting threshold conditions to join the task execution sequence of itself as an intelligent agent, forming a preliminary task package, in the process, the unmanned ship preferentially selects tasks in its priority task cluster to reduce potential execution cost and task conflict; then, each unmanned ship broadcasts the selected tasks and corresponding bid income to other unmanned ships in the cluster; after a round of bidding is completed, each unmanned ship updates the task selection set and enters the next round of bidding; the above steps are repeated iteratively until all tasks have a clear allocation and no conflict occurs, that is, all unmanned ships reach an agreement on the task allocation scheme, at this time, the task sequence set held by each unmanned ship constitutes a task allocation scheme.

[0026] The re-allocation using the greedy strategy specifically refers to removing the task that cannot be executed on time from the task sequence of the original execution unmanned ship, trying to insert it into the task sequence of other unmanned ships and selecting the insertion scheme with the minimum increase in total cost.

[0027] The conflict resolution rule specifically refers to preferentially assigning the task to the unmanned ship with fewer assigned tasks, and if the number is the same, assigning the task to the unmanned ship with lower execution cost.

[0028] The emergency type includes unmanned ship failure, addition of unmanned ship, task cancellation and task insertion.

[0029] Updating the available list of unmanned ships and the task pool according to the emergency type specifically refers to: if the unmanned ship fails, re-adding the unfinished task to the task pool and marking the unmanned ship as unavailable, if a new unmanned ship is added, adding the unmanned ship to the available resource list, if a task is cancelled, removing the task from the task pool, and if a new task is added, adding the new task to the task pool.

[0030] Compared with the prior art, the present application has the following beneficial effects:

[0031] (1) Strong intelligent dynamic adaptation capability: The present application considers both the task pre-allocation and execution process re-allocation stages, and through the event-driven mechanism, it can real-time perceive the unmanned ship failure, addition, task change and other conditions during task execution and timely start task re-allocation, ensuring that the cluster has dynamic adaptation capability to complex and variable task demands, greatly improving the reliability and robustness of task completion. Compared with the traditional method which can only be statically planned, the present application can quickly adjust the scheme after the occurrence of an emergency, avoiding task interruption or resource idling.

[0032] (2) Distributed coordination efficiency is high: the application adopts a distributed consensus bundling algorithm (CBBA) for task allocation, and each unmanned ship autonomously bids for tasks through communication iteration and reaches an agreement, without the need for a central controller to complete global coordination. This distributed architecture improves the scalability and fault tolerance of the system: when individual unmanned ships fail, the remaining unmanned ships can still coordinate to complete redistribution, without causing the entire system to fail. Compared with centralized algorithms, the application has higher communication efficiency and real-time performance in the multi-unmanned ship coordination scenario, and can meet the task allocation needs of large-scale unmanned ship clusters.

[0033] (3) Comprehensive optimization quality is good: the application integrates various optimization strategies into the framework of the basic CBBA algorithm, including a task cost threshold screening mechanism, a hard time window constraint processing, a task HDBSCAN clustering algorithm, a greedy insertion adjustment, and a genetic algorithm adaptive weight optimization. The cost threshold pre-screening reduces invalid calculations, and the vectorized batch calculation further speeds up the allocation process (the code execution speed is improved by about 10 times compared with conventional implementation). HDBSCAN adaptive density clustering enables the unmanned ship to preferentially allocate tasks to clusters that are closely related to its capabilities and adapt to multiple dimensions based on the composite characteristics of the tasks, significantly reducing the task search space and improving the intelligence of the allocation. The greedy reinsertion strategy for time window violation tasks improves the quality of the feasible solution of the initial scheme. These improvements significantly improve the global optimality of the task allocation scheme and the convergence speed of the algorithm. The genetic algorithm evolution weight further enhances the intelligence of the algorithm, and adjusts the weight coefficients of the cost function according to different task scenarios, so that the allocation result is more optimal under multi-objective trade-off.

[0034] (4) Good constraint satisfaction and cost control effect: the application regards the task time window as a hard constraint and strictly satisfies it, and through a high penalty coefficient, it ensures that any allocation scheme that violates the time window is minimized, thus meeting the strict requirements of the task on timeliness. At the same time, the task execution subject change cost is particularly introduced in the redistribution link, effectively controlling the frequent task reassignment, and only when the benefit is greater than the cost, the task is transferred, thereby balancing the task completion speed and the stability of the scheme, avoiding the low efficiency problem caused by blind redistribution in the prior art.

[0035] In summary, the heterogeneous unmanned ship task dynamic optimization method provided by the application significantly improves the cooperative working ability of the unmanned ship cluster in a complex task environment, enabling the heterogeneous unmanned ship cluster to achieve intelligent, efficient and reliable allocation and execution of multiple tasks while ensuring task timeliness and success rate. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The method flowchart of the application;

[0037] Figure 2 A specific implementation process diagram of the method of the present application is shown in the figure;

[0038] Figure 3 A flowchart of the HDBSCAN adaptive density clustering algorithm is shown in the figure;

[0039] Figure 4 A flowchart of the improved genetic algorithm is shown in the figure. DETAILED DESCRIPTION

[0040] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments are implemented on the premise of the technical solutions of the present application, and detailed implementation modes and specific operation processes are given, but the protection scope of the present application is not limited to the following embodiments.

[0041] The present embodiment provides a heterogeneous unmanned ship task dynamic optimization method based on intelligent distributed CBBA, as shown in Figure 1 and Figure 2 The method comprises the following steps:

[0042] S1, initializing data and parameters: obtaining the state data and task environment data of each unmanned ship in the heterogeneous unmanned ship cluster, preprocessing the data, and issuing the processed task parameters to each unmanned ship through encrypted communication.

[0043] In the present embodiment, the unmanned ship state data includes the current position, navigation speed, remaining power and completed task quantity of the unmanned ship; the environment data includes the temperature, humidity, wind speed and wind direction of the working sea area.

[0044] The preprocessing includes filling of missing data, elimination of abnormal data and cleaning of repeated data, and standardization or normalization conversion of the unmanned ship state data and environment data.

[0045] In the specific implementation process, the state information of each unmanned ship in the unmanned ship cluster and the information of the task set to be executed are obtained, and the simulation environment parameters are initialized; the cost function weight coefficients in the pre-set genetic algorithm gene coding are read, which are used to guide the optimization search of the task allocation process; the current time is set as the task start time, and all unmanned ships are in a standby executable state.

[0046] S2, task pre-allocation: defining a cost function, clustering and grouping all tasks to be executed according to the number of unmanned ships based on the intelligent distributed CBBA algorithm, and the corresponding unmanned ship preferentially bidding for the tasks in each subgroup; calculating the cost of each unmanned ship executing the candidate task based on the cost function, and initially selecting the task combination with the lowest cost to construct a task package; the unmanned ships interact with each other through communication for task bidding, and iterate until an initial task allocation scheme without conflict is obtained.

[0047] S21, define a cost function.

[0048] The cost function is defined by considering multiple factors, including the distance cost required for task execution, the time cost required for task execution, and the penalty term for task time window violation, and the relative importance of each cost is balanced by a weight coefficient. That is, the cost function is a weighted linear combination of distance cost, time cost and time window penalty, wherein the distance cost is determined based on the time / energy cost corresponding to the navigation distance of the unmanned ship from the current location to the task execution location, the time cost is determined based on the execution time and waiting time consumed by the task itself, and the time window penalty is defined as: if the required completion time of the task exceeds the specified time window, the penalty value is obtained by multiplying the exceeding time by a large penalty coefficient. In a preferred embodiment, other factors such as task priority and unmanned ship task switching loss can also be considered and added to the cost function.

[0049] In particular, the time window constraint of the task is set as a hard constraint condition, that is, once it is calculated that the estimated completion time of a task on a certain unmanned ship exceeds its allowed time window, it is considered that the allocation of the task is infeasible, and the penalty term is set as an extreme value much larger than the normal cost, so as to force the avoidance of schemes violating the time window in optimization. In addition, a cost threshold screening mechanism is introduced in the initial allocation: a task cost threshold is set for each unmanned ship, and only when the cost of a task relative to the unmanned ship is lower than the threshold, the task is considered to be included in its execution sequence, so as to filter out task allocation options with too high cost (low execution cost or unrealistic), reduce the search space of the algorithm and the interference of inferior solutions.

[0050] Specifically, the cost function is determined as follows:

[0051] Total navigation distance minimization G1: The navigation distance cost is determined based on the time / energy cost corresponding to the navigation distance of the unmanned ship from the current location to the task execution location. This cost aims to minimize the total navigation distance of the unmanned ship cluster, and its calculation formula is:

[0052]

[0053] wherein U num represents the total number of unmanned ships in the unmanned ship cluster, i represents the serial number of the unmanned ship, represents the navigation distance of the i-th unmanned ship from the current location to the first task point ID i,1 in its task sequence, represents the total number of tasks allocated to the i-th unmanned ship U i , and represents the navigation distance from the j-th task point ID i,j to the j+1-th task point ID i,j+1 .

[0054] Maximal task completion time minimization G2: The time cost is determined based on the execution time and waiting time consumed by the task itself. This cost aims to minimize the maximum time required for all tasks to be completed, and its calculation formula is:

[0055]

[0056] Where speed i represents the sailing speed of the ith unmanned ship, TS i,j represents the task execution time required by the ith unmanned ship to execute the jth task in its task sequence.

[0057] Minimization of the number of unfinished tasks G3: This cost aims to minimize the total number of unfinished tasks to improve the task completion rate. Its calculation formula is:

[0058]

[0059] Where T num represents the total number of tasks to be executed.

[0060] Total time for all unmanned ships to complete the execution of tasks G4: This cost aims to minimize the total time for all unmanned ships to execute tasks, and its calculation formula is:

[0061]

[0062] The final comprehensive objective function is obtained by weighted linear combination of the above costs:

[0063] min G = w1G1 + w2G2 + w3G3 + w4G4

[0064] Where w1, w2, w3, w4 are the weight factors of each cost.

[0065] S22, all tasks to be executed are clustered and grouped according to the number of unmanned ships.

[0066] First, multi-modal feature engineering and embedding are performed on all tasks to be executed and unmanned ships, and heterogeneous information such as the geographical location, type, time window, priority of the task, and the ability of the unmanned ship is fused to generate task feature representation.

[0067] As shown in Figure 3 , multi-modal feature extraction mainly includes four aspects, and these features together constitute the comprehensive attributes of the task for subsequent similarity calculation:

[0068] Geographical features: refer to the location information of the task in space, usually latitude and longitude coordinates. The extraction process involves obtaining the geographical coordinate data of each task from the pool of tasks to be assigned.

[0069] Time features: refer to information related to the time of a task. This includes but is not limited to the time window of a task (the earliest start and latest end time), the estimated execution time of a task, and the waiting time of a task.

[0070] Functional features: refer to the category or functional type of a task, such as reconnaissance, supply, patrol, etc. These discrete functional types need to be converted into numerical or vector representations for quantitative comparison.

[0071] Cost features: refer to the cost related to the execution of a task, such as the estimated energy consumption or resources consumed to complete the task, which is usually directly related to the time and distance of the task.

[0072] Then, these multi-source heterogeneous indicators are directly measured by defining a composite kernel function (Composite Kernel Function) to measure the comprehensive similarity between tasks. The calculation formula of the kernel function is as follows:

[0073] K(x i ,x j )=w1K type (x i ,x j )+w2K time_window (x i ,x j )+w3K geo (x i ,x j )+w4K duration (x i ,x j )

[0074] Where K type is used to measure the similarity of task types, K time_window is used to measure the overlap or proximity of task time windows, K geo is used to measure the proximity of task geographic location (Haversine distance can be used), K duration is used to measure the similarity of task time consumption, w m , m = 1, 2, 3, 4 are the weights corresponding to each indicator, which will be adaptively adjusted through subsequent optimization iterations. In this way, a high-dimensional task feature vector is generated that can fully represent the inherent association of tasks.

[0075] Based on the task feature representation, the HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) adaptive density clustering algorithm is used to cluster the tasks. The HDBSCAN adaptive density clustering algorithm first constructs a minimum spanning tree (MST) of the data by analyzing the core distance and mutual reachable distance of the data points, wherein the weight of the edge is the mutual reachable distance between the task points based on the fusion feature, and the specific formula is as follows:

[0076] MRD k (a,b)=max(core_dist k (a),core_dist k (b),dist(a,b))

[0077] Wherein, MRD k (a,b) represents the mutual reachable distance between task points a and b; k is an integer parameter, usually related to the minimum cluster size, which defines the number of nearest neighbors considered when calculating the core distance; core_dist k (x) represents the core distance of task point x, that is, the distance to its kth nearest neighbor; dist(a,b) represents the original distance between task points a and b based on the composite kernel function.

[0078] Then a series of interrelated cluster hierarchy structures are created by deleting the edges with relatively higher weights in the minimum spanning tree, and different shape and density task clusters are adaptively identified based on cluster stability analysis, thereby overcoming the preset limitations of cluster shape and number of traditional K-means.

[0079] After generating the task clusters, the task clusters are constrainedly matched and optimized according to the task categories and the performance of the unmanned ships (task ability, power endurance, etc.), that is, it is ensured that the tasks in each task cluster can be executed by the unmanned ships with the corresponding functions or abilities, and the performance of the unmanned ships such as endurance and load is comprehensively considered. Specifically, the first stage is capability matching (hard constraint): this part is a functional preliminary screening, which first checks each task cluster generated by HDBSCAN, identifies the types of tasks inside (for example, reconnaissance, supply), and then iterates through all available unmanned ships, and directly excludes the unmanned ships that do not have the core functions required to execute these tasks. The second stage is performance evaluation and optimization (soft constraint): after passing the screening of the first stage, the specific performance indicators of each qualified unmanned ship executing the task cluster are evaluated. This includes the navigation distance of the unmanned ship from the current location to the task cluster, the total time required to complete the task, the power consumption, and whether the load is matched, etc. Then, a comprehensive cost or benefit score is calculated, which takes all these performance indicators into account. Finally, through an optimization process (for example, a simple greedy algorithm or a more complex matching algorithm), each task cluster is assigned to the unmanned ship with the lowest cost and the best performance, thereby generating an optimized "priority task set" for each unmanned ship that best matches its capabilities.

[0080] Finally, by multi-objective dynamic optimization of the assignment relationship of the task clusters to the unmanned ships, one or more task clusters are determined for each unmanned ship as its priority task set until the assignment result is stable. Specifically,

[0081] If the number of task clusters is equal to the number of unmanned ships: convert the cost matrix into a assignment problem, and use the Hungarian algorithm or other effective graph matching algorithm to find the optimal one-to-one matching to minimize the total cost.

[0082] If the number of task clusters is not equal to the number of unmanned ships: if the number of task clusters is more than the number of unmanned ships, a greedy strategy is adopted: the smallest element in the cost matrix is selected, the corresponding task cluster is assigned to the unmanned ship, and then the state is updated and the process is repeated until all task clusters are assigned.

[0083] If the number of task clusters is less than the number of unmanned ships, a many-to-one matching strategy is adopted: one task cluster can be assigned to multiple unmanned ships to be completed cooperatively (if feasible), or part of the unmanned ships are marked as "idle" for subsequent dynamic scheduling.

[0084] S23, execute initial task assignment

[0085] Each unmanned ship evaluates the candidate tasks based on the cost function and the clustering results, respectively, and preferentially selects tasks with a cost lower than a preset threshold to join its execution sequence, thereby gradually constructing an initial task bundle with the maximum revenue (equivalent to the minimum total cost) for each unmanned ship. The task allocation is implemented using a distributed consensus bundling algorithm (CBBA) mechanism: each unmanned ship locally calculates the optimal task bundle and performs task bidding, and the temporary winners of the tasks are updated through multiple rounds of iterative communication until convergence.

[0086] Specifically, the intelligent distributed CBBA algorithm performs the following steps to implement the initial task allocation: each unmanned ship selects a set of tasks with the lowest cost and that meet the threshold condition to join its task execution sequence, thereby forming a preliminary task bundle. In this process, the unmanned ship preferentially selects tasks within its priority task cluster to reduce potential execution costs and task conflicts. Then, each unmanned ship broadcasts the selected tasks and the corresponding bidding revenue (negative cost) to other unmanned ships in the cluster. After one round of bidding, each unmanned ship updates the task selection set and enters the next round of bidding. The above steps are repeated iteratively until all tasks have a clear allocation and no longer conflict, i.e., all unmanned ships agree on the task allocation scheme. At this point, the task sequence set held by each unmanned ship constitutes a task allocation scheme.

[0087] S3, Time window adjustment: verify the time window constraints of each task in the initial task allocation scheme, and use a greedy strategy to redistribute tasks that exceed the time window requirements: remove tasks that cannot be executed on time from the original execution unmanned ship's task sequence, try to insert them into other unmanned ships' task sequences, and select the insertion scheme that minimizes the total cost increase to meet the time window constraints of all tasks.

[0088] Specifically, the simulation verification is performed on the tasks in the initial execution sequence of each unmanned ship according to a time axis, to determine whether the start / end time of each task meets the time window constraint thereof. The expected start and end time of each task is calculated according to the execution sequence, and compared with the upper and lower limits of the time window thereof. When it is found that the planned start time of a task is later than the latest allowed start time thereof (or the end time exceeds the deadline), it is determined that the task cannot meet the time window constraint in the current sequence. If it is found that a task cannot be executed within the specified time window under the current allocation, that is, the task cannot meet the time window constraint, the task is removed from the sequence of the unmanned ship. Subsequently, a greedy insertion strategy is used to attempt to re-allocate the task to other unmanned ships: the task is simulated to be inserted into the execution sequence of other unmanned ships at appropriate positions, the increased cost (including the increased sailing distance and waiting time of the unmanned ship, etc.) is calculated on the premise of meeting the time window requirement of the task, after traversing all insertion possibilities of other unmanned ships, an insertion scheme with the minimum increased cost and meeting the time window is selected, and the task is allocated to the corresponding unmanned ship for execution. If there are multiple unmanned ships that cannot meet the time window requirement of the task in their sequences, the task cannot be allocated temporarily and will be further processed in the subsequent re-allocation stage. Through the greedy insertion process, the time window conflict is repaired as much as possible without disturbing the overall situation, the satisfaction degree of the initial task plan to the time constraint is improved, and the feasibility and quality of the task allocation scheme are improved.

[0089] S4, conflict resolution: when a conflict occurs in which multiple unmanned ships bid for the same task in the initial task allocation scheme, the conflict is resolved according to the conflict resolution rule: the task is preferentially allocated to the unmanned ship with fewer allocated tasks, and if the number of allocated tasks is the same, the unmanned ship with lower execution cost is allocated; the bidding and conflict resolution are repeated until all tasks are uniquely attributed, a task allocation scheme that meets the constraint condition is obtained, and the global objective function corresponding to the scheme is calculated.

[0090] Specifically, when two or more unmanned ships compete for the same task, the final execution attribution of the task is determined according to the number of allocated tasks and the increased cost between the conflicting unmanned ships. The specific conflict resolution rule is: the task is preferentially allocated to the unmanned ship with fewer allocated tasks; if the number of allocated tasks of the competing parties is the same, the cost of adding the task to the sequence of each party is compared, and the party with smaller cost is selected to execute the task.

[0091] Through the conflict resolution rule, the task is preferentially directed to the unmanned platform that is more idle or more efficient in executing the task, thereby improving the rationality and efficiency of the overall task allocation. After the attribution of the conflicting task is determined, the unmanned ship that does not obtain the task removes the task from its task package, and considers other alternative tasks in subsequent iterations.

[0092] S5, scheme optimization iteration: the weight coefficients of the cost function are evolved and adjusted by using a genetic algorithm, and steps S2-S4 are repeatedly executed to iteratively optimize the task allocation scheme until a preset iteration number is reached or the global objective function converges, and an optimized optimal task allocation scheme is output.

[0093] The initial scheme is often not a globally optimal scheme. To further optimize the task allocation result, the present application uses a genetic algorithm to optimize the weight coefficients of the cost function. Specifically, as shown in Figure 4 the global objective function is minimized as the optimization target, the weight coefficients are encoded and evolved as individual chromosomes in the genetic algorithm, and under the condition that the task set and the initial state of the unmanned ship are unchanged, new weight combinations are continuously generated through selection, crossover, mutation and other operations to find better weight combinations. After each weight update, a new task allocation scheme is generated by re-executing steps S2-S4, and its global objective function value is calculated. This iteration continues until the genetic algorithm reaches a preset number of iterations or a convergence condition, and finally outputs the optimal task allocation scheme in the task pre-allocation stage. Optionally, Monte Carlo simulation is introduced during the genetic algorithm iteration process to sample and simulate evaluate some candidate allocation schemes, from which stable and excellent schemes are selected for the next generation iteration to improve the robustness and global optimality of the algorithm result. When the genetic algorithm iteration reaches the preset number of iterations or the global fitness is no longer significantly improved, the task scheme corresponding to the highest fitness (lowest objective function value) is taken as the optimal scheme output. This optimal scheme is the final unmanned ship task execution plan determined in the task pre-allocation stage, which will guide the actual execution of subsequent tasks when there are no sudden events.

[0094] S6, task execution monitoring: control each unmanned ship to execute tasks according to the optimal task allocation scheme instructions, and monitor the task execution state in real time, detect sudden events, and execute step S7 when any of the following sudden events is detected: unmanned ship failure, new unmanned ship, task cancellation or task insertion.

[0095] Specifically, during task execution, the state of the unmanned ship cluster and external environmental changes are monitored in real time to determine whether a sudden event that affects task execution has occurred. When no sudden event occurs, each unmanned ship will travel and execute tasks in accordance with the established sequence order, and after completing the task, it will enter standby or return. All possible sudden events and their occurrence times are recorded in advance, and when the current time reaches the trigger point of a sudden event, the execution of the current task scheme is immediately suspended, the current time and scene state are recorded, and the state updating and task pool reconstruction step is entered.

[0096] S7, state update and task pool reconstruction: update the available list of unmanned ships and the task pool according to the type of emergency event: if an unmanned ship fails, its unfinished tasks are returned to the task pool and the unmanned ship is marked as unavailable, if a new unmanned ship is added, it is added to the available resource list, if a task is canceled, the task is removed from the task pool, if a new task is added, the new task is added to the task pool.

[0097] Specifically, when an unmanned ship fails, i.e. a certain unmanned ship fails or loses contact, it cannot continue to perform its unfinished tasks, the task currently being performed by the failed unmanned ship is terminated, the task (if not completed) and all tasks not yet performed by the failed unmanned ship are returned to the task pool, and the unmanned ship is marked as unavailable and no longer participates in subsequent task allocation;

[0098] When a new unmanned ship is added, i.e. a new unmanned ship is added to the task formation as an available resource, the execution state of the executed part of the task is maintained unchanged, all unfinished tasks are put into the task pool to wait for allocation again, and the unmanned ship is included in the available execution unit, and its initial state is regarded as idle and available, and it participates in allocation calculation from the next round of task allocation;

[0099] When a task is canceled, i.e. a certain task is canceled due to external reasons and no longer needs to be executed, if the task is being executed by a certain unmanned ship, the unmanned ship is immediately notified to stop the task and marked as completed; the task is removed from the task pool and the execution sequence of all unmanned ships. If the task has not yet started execution, it only needs to be removed from the task pool;

[0100] When a new task is inserted, i.e. a new emergency task appears and needs to be temporarily added to the existing task set for execution, the new task is added to the task pool; other tasks that have not yet been executed are also left in the task pool to wait for re-allocation.

[0101] S8, task reallocation: based on the updated task pool and the state of the unmanned ship, the intelligent distributed CBBA algorithm is re-executed to allocate all unfinished tasks, and a new task allocation scheme is obtained, wherein a task whose execution unmanned ship changes is given an additional reallocation penalty to reduce the loss caused by replacing the execution subject; the new task allocation scheme is sent to the corresponding unmanned ship to continue executing the remaining tasks.

[0102] Specifically, taking the system state at the time of the current event (including the real-time position of each unmanned ship, the idle / execution state, and the updated task pool) as the new initial condition, the same distributed CBBA algorithm as in the task pre-allocation stage is used to re-allocate the tasks in the task pool, and an adjusted task execution scheme is generated. In this re-allocation process, for tasks whose execution subjects change compared with the previous round of scheme, a re-allocation cost (such as a fixed switching cost) is introduced and added to the value of the task, to explicitly measure the performance loss brought by task reassignment.

[0103] wherein the re-allocation cost function is:

[0104]

[0105] In the above formula, C is a fixed switching cost constant, which represents the average loss brought by each task reassignment; fixed is the set of all unfinished tasks that need to be re-allocated; and δ is a binary indicator variable, which takes the value of 1 when task j is assigned to a different unmanned ship in this re-allocation than in the last time, and 0 otherwise. j

[0106] The weight coefficient of the re-allocation cost is also adaptively adjusted and optimized by a genetic algorithm, and the optimization goal remains the same as in the pre-allocation stage.

[0107] The re-allocation algorithm comprehensively considers the sailing path cost, the remaining execution time, and the above-mentioned re-allocation penalty, while ensuring the successful completion of the task, and minimizing the scheme shock. After completing a new round of task re-allocation, the updated task instructions are issued to the relevant unmanned ships, and each unmanned ship adjusts its own task execution plan accordingly.

[0108] S9, loop until completion: real-time monitoring of task execution state, repeated execution of the steps of sudden event detection, state update and task pool reconstruction and task re-allocation, and a plurality of sudden events occurring in time sequence are processed in turn, until all tasks are completed.

[0109] Specifically, the newly updated task allocation scheme is applied to the subsequent task execution process, and the simulation time is continued to advance. If there are still sudden events that have not been processed on the subsequent time axis, steps S7-S8 are repeated when the time advances to the occurrence of the next conflict event, and dynamic re-allocation is performed again; after each re-planning, the unmanned ship continues to execute the remaining tasks according to the latest scheme. This cycle continues until all scheduled tasks are completed and no new sudden events occur, and the entire heterogeneous unmanned ship cluster task dynamic execution optimization process ends. Finally, the method outputs the task execution sequence and path planning of each unmanned ship, ensuring that all tasks are efficiently completed in a dynamic environment. ​​

[0110] Through the above steps, the application realizes task optimization allocation and execution control of the heterogeneous unmanned ship cluster in a dynamic environment. Even if unpredictable situations such as damage of the unmanned ship, insertion of a new task, etc. occur during task execution, the system can timely adjust the task scheme, re-optimize the allocation of the remaining tasks, and ensure the smooth completion and optimal performance of the overall task.

[0111] Finally, the application tests and verifies the algorithm for small-scale and large-scale random cases, and obtains the following comparative effects:

[0112] Table 1 shows the comparative effect of the algorithm

[0113]

[0114] The results show that for the two stages of pre-allocation and re-allocation in large-scale and small-scale scenarios, the improved algorithm proposed by the application can be superior to the traditional CBBA algorithm in performance.

[0115] In summary, the embodiments provided by the application effectively verify the feasibility and superiority of the heterogeneous unmanned ship task dynamic optimization method based on intelligent distributed CBBA in a complex environment. The proposed method can significantly improve the efficiency and reliability of multi-unmanned ship cooperative task execution, and has important significance for improving the autonomous cooperative combat capability of the marine unmanned equipment.

[0116] The above describes the preferred embodiments of the application. It should be understood that those skilled in the art can make many modifications and changes without creative labor based on the concept of the application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment based on the existing technology according to the concept of the application should be within the protection scope determined by the claims.

Claims

1. A method for dynamic optimization of heterogeneous unmanned surface vessel missions based on intelligent distributed CBBA, characterized in that, Includes the following steps: S1. Initialize data and parameters: Obtain the status data and mission environment data of each unmanned vessel in the heterogeneous unmanned vessel cluster, preprocess the data, and send the processed mission parameters to each unmanned vessel through encrypted communication; S2. Task pre-allocation: Define a cost function, based on the intelligent distributed CBBA algorithm, cluster all tasks to be executed according to the number of unmanned vessels, and let the corresponding unmanned vessels bid for the tasks in their respective groups first. The cost of each unmanned vessel executing a candidate task is calculated based on the cost function, and the task combination with the lowest cost is initially selected to construct a task package. The unmanned vessels communicate with each other to bid for tasks, and the process is iterated until a conflict-free initial task allocation scheme is obtained. S3. Time window adjustment: Check the time window constraints of each task in the initial task allocation scheme, and use a greedy strategy to reallocate tasks that exceed the time window requirements to satisfy the time window constraints of all tasks. S4. Conflict Resolution: When multiple unmanned vessels compete for the same task in the initial task allocation scheme, conflict resolution is carried out according to the conflict resolution rules; the bidding and conflict resolution are repeated until all tasks are uniquely assigned, a task allocation scheme that satisfies the constraints is obtained, and the global objective function corresponding to the scheme is calculated. S5. Scheme optimization iteration: Use a genetic algorithm to evolve and adjust the weight coefficients of the cost function, repeat steps S2 to S4 to iteratively optimize the task allocation scheme until the preset number of iterations is reached or the global objective function converges, and output the optimized optimal task allocation scheme. S6. Task execution monitoring: Control each unmanned vessel to execute tasks according to the instructions of the optimal task allocation scheme, monitor the task execution status in real time, and detect emergencies. When any emergency is detected, execute step S7. S7. Status Update and Task Pool Reconstruction: Update the list of available unmanned vessels and the task pool according to the type of emergency. S8. Task Reassignment: Based on the updated task pool and the unmanned vessel status, the intelligent distributed CBBA algorithm is re-executed to reassign all unfinished tasks, resulting in a new task allocation scheme. Additional reassignment cost penalties are imposed on tasks whose execution unmanned vessels have changed to reduce the losses caused by the change of task execution entities. The new task allocation scheme is then distributed to the corresponding unmanned vessels to continue executing the remaining tasks. S9. Loop until completion: Monitor the task execution status in real time, repeatedly execute the steps of emergency event detection, status update and task pool reconstruction and task reallocation, and process multiple emergencies that occur in time sequence until all tasks are completed.

2. The method for dynamic optimization of heterogeneous unmanned surface vessels based on intelligent distributed CBBA according to claim 1, characterized in that, The unmanned vessel status data includes the unmanned vessel's current location, speed, remaining battery power, and number of completed tasks.

3. The method for dynamic optimization of heterogeneous unmanned surface vessels based on intelligent distributed CBBA according to claim 1, characterized in that, The environmental data includes the temperature, humidity, wind speed, and wind direction of the operating sea area.

4. The method for dynamic optimization of heterogeneous unmanned surface vessel missions based on intelligent distributed CBBA as described in claim 1, characterized in that, The preprocessing includes: filling in missing data, removing abnormal data, and cleaning up duplicate data, as well as standardizing or normalizing the unmanned vessel status data and environmental data.

5. The method for dynamic optimization of heterogeneous unmanned surface vessels based on intelligent distributed CBBA according to claim 1, characterized in that, The cost function includes distance cost, time cost, and time window penalty. The distance cost is determined based on the time / energy cost corresponding to the travel distance from the unmanned vessel's current location to the mission execution location. The time cost is determined based on the execution time consumed by the mission itself and the waiting time. The time window penalty is defined as follows: if the required completion time of the mission exceeds its specified time window, the penalty value is obtained by multiplying the excess time by a maximum penalty coefficient. The distance cost, time cost, and time window penalty are combined linearly with weights to form the total cost function.

6. The method for dynamic optimization of heterogeneous unmanned surface vessels based on intelligent distributed CBBA according to claim 1, characterized in that, The process of clustering all tasks to be performed according to the number of unmanned vessels includes the following steps: Multimodal feature engineering and embedding are performed on all tasks to be executed and unmanned vessels, and heterogeneous information is fused to generate task feature representations; Based on task feature representation, the HDBSCAN adaptive density clustering algorithm is used to cluster tasks. The HDBSCAN algorithm first constructs a minimum spanning tree of the data, then creates a series of interconnected cluster hierarchies by deleting edges with relatively higher weights from the minimum spanning tree. Based on cluster stability analysis, it adaptively identifies task clusters of different shapes and densities. After generating task clusters, constrained matching and optimization are performed on the task clusters according to task category and unmanned surface vessel (USV) performance. Through multi-objective dynamic optimization of the assignment relationship between task clusters and USVs, one or more task clusters are assigned as the priority task set for each USV until the allocation result stabilizes.

7. The method for dynamic optimization of heterogeneous unmanned surface vessel missions based on intelligent distributed CBBA according to claim 1, characterized in that, The intelligent distributed CBBA algorithm performs the following steps to achieve initial task allocation: Each unmanned surface vessel (USV), acting as an agent, selects a set of tasks with the lowest cost and meeting a threshold condition to add to its own task execution sequence, forming a preliminary task package. During this process, USVs prioritize tasks within their priority task clusters to reduce potential execution costs and task conflicts. Then, each USV broadcasts its selected tasks and corresponding bidding rewards to other USVs in the cluster. After one round of bidding, each USV updates its task selection set and enters the next round of bidding. The above steps are repeated iteratively until all tasks have a clear allocation and no longer conflict, which is considered as all USVs reaching a consensus on the task allocation scheme. At this point, the task sequence set held by each USV constitutes a task allocation scheme.

8. The method for dynamic optimization of heterogeneous unmanned surface vessel missions based on intelligent distributed CBBA according to claim 1, characterized in that, The greedy strategy for reallocation specifically involves removing tasks that cannot be executed on time from the original task sequence of the unmanned vessel, attempting to insert them into the task sequences of other unmanned vessels, and selecting the insertion scheme that minimizes the total cost.

9. The method for dynamic optimization of heterogeneous unmanned surface vessels based on intelligent distributed CBBA according to claim 1, characterized in that, The specific conflict resolution rule is as follows: the task is preferentially assigned to the unmanned vessel with the fewer assigned tasks, and if the number of tasks is the same, it is assigned to the unmanned vessel with the lower execution cost.

10. The method for dynamic optimization of heterogeneous unmanned surface vessel missions based on intelligent distributed CBBA according to claim 1, characterized in that, The types of emergencies include unmanned vessel malfunction, addition of unmanned vessels, mission cancellation, and mission insertion. The update of the unmanned surface vessel (USV) availability list and task pool based on the type of emergency is as follows: if a USV malfunctions, its unfinished tasks are re-added to the task pool and the USV is marked as unavailable; if a new USV is added, it is added to the available resource list; if a task is canceled, it is removed from the task pool; if a new task is added, it is added to the task pool.

Images