An evolutionary fusion two-stage hybrid-based crowd sensing collaborative optimization method and system

Abstract

The application relates to the technical field of path planning, in particular to a crowd-sensing cooperative optimization method and system based on evolutionary fusion two-stage mixing. The method comprises the following steps: constructing a multi-agent cooperative optimization model of a heterogeneous space based on a mobile crowd-sensing operation scene; adopting a stage-type evolutionary fusion strategy to deeply fuse MOPSO and NSGA-II, and constructing an EF-DH algorithm; using the EF-DH algorithm to perform unmanned aerial vehicle multi-target path planning based on the constructed multi-agent cooperative optimization model, including first-stage unmanned aerial vehicle cluster path optimization and second-stage ground operation personnel task optimization; and performing air-ground cooperative execution and dynamic re-optimization based on the path planning. The evolutionary fusion two-stage optimization algorithm fusing MOPSO and NSGA-II is constructed, and the global search capability and local convergence performance of the multi-target optimization problem are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a collaborative optimization method and system based on evolutionary fusion and two-stage hybrid swarm intelligence perception. Background Technology

[0002] Mobile Crowd Sensing (MCS), as a novel distributed data acquisition paradigm, relies on the sensing capabilities of smart terminals such as smartphones and wearable devices, combined with the characteristics of group participation. It can complete tasks in large-scale spatial environmental perception, traffic monitoring, and public safety at low cost and high efficiency, and has become one of the core technologies for the construction of the Internet of Things and smart cities. However, the operational efficiency of existing mobile crowd sensing systems is limited by two core problems: unreasonable task allocation and inefficient resource utilization. The root cause is that the system lacks the perception and utilization of real-time environmental information and cannot adaptively adjust the task allocation strategy according to dynamic environmental changes (such as road congestion, physical obstacles, and regional environmental characteristics). This leads to redundant task execution paths for ground personnel, excessive energy consumption, and even task failures.

[0003] To compensate for the environmental perception shortcomings of mobile swarm sensing systems, unmanned aerial vehicles (UAVs) have been introduced into the MCS (Multi-Site Controller) system as aerial sensing nodes. Leveraging their high mobility, flexible deployment, and wide-area perception capabilities, they can collect information in areas difficult to cover on the ground. However, the current application of UAVs in MCS still has significant limitations: on the one hand, the function of UAVs is limited to sparse ground truth data collection or static monitoring, and their powerful dynamic environmental perception capabilities have not been fully utilized, failing to provide effective real-time environmental intelligence for ground task allocation; on the other hand, UAV path planning often adopts a single-objective optimization strategy, focusing only on task point coverage or minimizing flight distance, ignoring the dual objectives of task truth data collection and environmental information acquisition, resulting in low overall utility of UAV flight paths.

[0004] Meanwhile, in existing UAV-ground personnel collaborative MCS systems, there is an information gap between aerial perception and ground execution. UAV perception data is not effectively integrated into the ground task allocation decision-making process. Ground task allocation remains based on pre-set static environmental information, failing to utilize real-time road congestion and environmental feature data acquired by the UAV to optimize operational paths, thus creating a technical bottleneck of "disconnect between aerial perception and ground execution." Furthermore, existing multi-objective optimization algorithms applied to UAV path planning and ground task allocation often employ single algorithms (such as particle swarm optimization and genetic algorithms), resulting in insufficient global exploration capabilities or poor local convergence, making it difficult to obtain optimal Pareto solutions in complex scenarios with multiple objectives and constraints. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a collaborative optimization method and system for swarm intelligence perception based on evolutionary fusion and two-stage hybridization.

[0006] In a first aspect, the present invention provides a collaborative optimization method for collective intelligence perception based on evolutionary fusion and two-stage hybridization, which adopts the following technical solution:

[0007] A collaborative optimization method for swarm intelligence perception based on evolutionary fusion and a two-stage hybrid approach includes:

[0008] Based on mobile swarm sensing operation scenarios, a multi-agent collaborative optimization model for heterogeneous spaces is constructed.

[0009] The EF-DH algorithm is constructed by deeply fusing MOPSO and NSGA-II using a staged evolutionary fusion strategy.

[0010] Based on the constructed multi-agent collaborative optimization model, the EF-DH algorithm is used for UAV multi-objective path planning, including the first stage of UAV swarm path optimization and the second stage of ground operation personnel task optimization.

[0011] Air-to-ground collaborative execution and dynamic re-optimization based on path planning.

[0012] Secondly, a swarm intelligence perception and collaborative optimization system based on evolutionary fusion and a two-stage hybrid approach includes:

[0013] The data acquisition module is configured to construct a multi-agent collaborative optimization model in a heterogeneous space based on the mobile swarm perception operation scenario.

[0014] The algorithm fusion module is configured to use a staged evolutionary fusion strategy to deeply fuse MOPSO and NSGA-II to construct the EF-DH algorithm.

[0015] The model optimization module is configured to perform multi-objective path planning for UAVs based on the constructed multi-agent collaborative optimization model and the EF-DH algorithm, including the first stage of UAV swarm path optimization and the second stage of ground operation personnel task optimization.

[0016] The re-optimization module is configured to perform air-to-ground collaborative execution and dynamic re-optimization based on path planning.

[0017] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned method for collaborative optimization based on evolutionary fusion dual-stage hybrid sensing.

[0018] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide the aforementioned method for collaborative optimization based on evolutionary fusion dual-stage hybrid crowd sensing.

[0019] In summary, the present invention has the following beneficial technical effects:

[0020] (1) This invention constructs an evolutionary fusion two-stage optimization algorithm that integrates MOPSO and NSGA-II, which effectively improves the global search capability and local convergence performance of multi-objective optimization problems. Compared with a single optimization algorithm, it can obtain a better Pareto solution set, thereby significantly improving the overall optimization quality of UAV path planning and task allocation.

[0021] (2) By introducing real-time environmental perception information from UAVs and dynamically feeding back the aerial perception data to the ground task allocation process, this invention realizes the information loop of air-ground collaboration, effectively solves the problem of "disconnection between environmental perception and task execution" in traditional mobile swarm perception systems, reduces path redundancy and energy consumption, and improves the system's ability to adapt to dynamic environmental changes.

[0022] (3) The present invention adopts a two-stage collaborative optimization mechanism, which jointly models and solves the UAV path planning and the task allocation of ground operators in stages. While ensuring the efficiency of task completion, it realizes the collaborative scheduling of resources of multiple subjects, significantly improving the task completion rate, the overall system benefits and resource utilization efficiency. It is suitable for large-scale collective intelligent perception task optimization in complex dynamic scenarios. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of a crowd intelligence perception collaborative optimization method based on evolutionary fusion dual-stage hybridization according to Embodiment 1 of the present invention;

[0024] Figure 2 This is a topological diagram of the internal structure of the EF-DH algorithm;

[0025] Figure 3 It is an air-to-ground collaborative execution and dynamic re-optimization topology graph;

[0026] Figure 4 This is a graph showing the total revenue of the experimental drone in Embodiment 1 of the present invention;

[0027] Figure 5 This is a diagram of the maximum working time of a worker in the experiment of Embodiment 1 of the present invention;

[0028] Figure 6 This is a diagram of the total work time of the experiment and the workers in Embodiment 1 of the present invention;

[0029] Figure 7 This is the experimental-total worker income diagram of Embodiment 1 of the present invention. Detailed Implementation

[0030] The present invention will be further described in detail below with reference to the accompanying drawings.

[0031] Example 1

[0032] Reference Figure 1 This embodiment presents a collaborative optimization method for swarm perception based on evolutionary fusion dual-stage hybrid. Addressing the multi-objective optimization challenge of air-to-ground coordination in mobile swarm perception systems, it constructs a two-stage hierarchical coupled collaborative optimization architecture and designs an evolutionary fusion dual-stage hybrid (EF-DH) multi-objective optimization algorithm. Through a full-link technology system encompassing modal feature decoupling, multi-objective collaborative optimization, environmental information feedback, and ground task readjustment, it achieves deep collaborative optimization of UAV path planning and ground personnel task allocation. This scheme breaks through the optimization bottleneck of traditional single algorithms, integrating the global exploration capability of Multi-Objective Particle Swarm Optimization (MOPSO) with the local mining and Pareto solution convergence capability of Non-Dominated Sorting Genetic Algorithm II (NSGA-II). It establishes an information closed loop between UAV environmental perception and ground mission execution. At the same time, through solution space encoding optimization, dynamic weight iteration strategy, Pareto solution optimization mechanism and constraint condition adaptive verification, it achieves high accuracy and high robustness of multi-objective optimization under complex constraints. The specific technical solution is divided into five core parts: system modeling and problem bounding, EF-DH algorithm core design, UAV multi-objective path planning (first stage optimization), ground personnel multi-objective task allocation (second stage optimization), air-ground collaborative execution and dynamic re-optimization. The technical details of each part are coupled layer by layer and progressively advance to form a complete technical implementation system.

[0033] (1) System modeling and problem delimitation

[0034] For practical operational scenarios of mobile swarm perception, a multi-agent collaborative optimization model in a heterogeneous space is constructed. The operational environment, perception tasks, UAV swarms, and ground personnel are accurately mathematically modeled. Based on the principles of hierarchical optimization and information coupling, the overall collaborative optimization problem is defined as two interrelated sub-problems: UAV multi-objective path planning (MOPP) and ground personnel multi-objective task allocation (MOTA). The optimization objectives, constraints, and information interaction relationships of the sub-problems are clarified, providing a rigorous mathematical foundation and problem boundary for subsequent algorithm design.

[0035] 1) Work Environment Model

[0036] Define a two-dimensional square work area The region is an orthogonal road network containing N longitudinal roads and M transverse roads. The roads are discretized into a finite number of road segments at intersections, and a set of road segments is constructed. For each road segment Two core attributes are assigned: congestion level The value is obtained by normalizing the real-time traffic frequency of the road using the min-max method, representing the driving efficiency of ground personnel on that road segment; environmental value. The weighted average of spatial coverage weight, information entropy, and task relevance of a road segment represents the reference value of the road segment's environmental information for ground task allocation. A road map is constructed based on the road segment set. Where V is the set of nodes at road intersections and E is the set of edges for road segments, the movement paths of ground personnel strictly follow the topological structure of the road graph.

[0037] 2) Perception Task Model

[0038] Define a set of perception tasks All tasks are distributed within the road network neighborhood (distance from road segments not exceeding δ), meeting the practical operational requirements of mobile group sensing. For each task... Assigning a four-dimensional attribute vector ,in The two-dimensional spatial coordinates of the task. As the basic reward for the task, The task difficulty level. The base execution time of the task, and ( (The noise is random, simulating time fluctuations in actual operations). Task execution satisfies a uniqueness constraint, meaning each task can only be executed by one ground personnel, and the execution value is positively correlated with the personnel's ability level.

[0039] 3) Unmanned Aerial Vehicle (UAV) Swarm Model

[0040] Define a drone swarm set All drones departed from designated bases. The drone departs, performs its flight mission, and returns to base, satisfying the closed-loop trajectory constraint. Each drone... With a fixed sensing radius When the task spatial coordinates Euclidean distance from the drone's flight path At that time, it is determined that the task has been perceived by the drone, and the task truth data is collected; simultaneously, the road sections covered by the drone's flight path are... It was identified as an environmental information sensing device, collecting data on the congestion level of that road section. Environmental value The drone's flight path is a polygonal path within a two-dimensional plane. ,in The path length is the sum of the Euclidean distances between the waypoints of the UAV, satisfying the energy consumption constraint (the total flight distance does not exceed the maximum range of the UAV). ).

[0041] 4) Ground worker model

[0042] Define the set of ground workers All personnel initially occupy node set V of road graph G, and their movement paths strictly follow the topological structure of the road graph, satisfying spatial movement constraints. For each personnel... Assigning three-dimensional attribute vectors ,in The initial spatial location of the personnel, Personnel capability levels determine the value gain and time efficiency of task execution (the higher the capability level, the greater the value gain and the shorter the execution time). This is the base speed for pedestrian movement; actual movement speed depends on the degree of road congestion. Revised to This enables dynamic constraints on personnel movement based on environmental information.

[0043] 5) Problem layering and information coupling

[0044] The air-to-ground cooperative optimization problem is decomposed into two sub-problems: MOPP (Moving Upward Optimization Problem) and MOTA (Moving Downward Optimization Problem). MOPP is the upstream optimization sub-problem, and MOTA is the downstream optimization sub-problem. These two sub-problems are unidirectionally strongly coupled through UAV environmental perception information: the output of MOPP is the real-time environmental state matrix of the road network perceived by the UAV. (Each row corresponds to the congestion level and environmental value of a road segment). This matrix serves as the core input of MOTA, directly correcting the movement speed and mission execution value of ground personnel, and realizing the information loop of "airborne perception - ground optimization". At the same time, both sub-problems are multi-objective integer optimization problems and belong to NP-hard problems, which cannot be solved in polynomial time by exact algorithms. Therefore, the EF-DH heuristic algorithm is designed to achieve efficient solution of high-quality suboptimal solutions.

[0045] (2) Core design of EF-DH algorithm

[0046] The EF-DH algorithm is the core technological innovation of this invention. Addressing the shortcomings of traditional single multi-objective optimization algorithms—such as difficulty in simultaneously addressing global exploration and local mining, uneven distribution of Pareto solutions, and poor convergence under complex constraints—it employs a staged evolutionary fusion strategy, deeply integrating MOPSO and NSGA-II. This is not a simple algorithm patchwork; rather, it achieves complementary advantages of the two algorithms through four key technologies: unified solution space encoding, seamless iteration state transitions, collaborative mapping of optimization objectives, and global constraint verification. Furthermore, it designs a dynamic inertia weight decay strategy, an adaptive archive set update mechanism, an improved congestion distance calculation method, and an ideal point Pareto solution selection criterion to enhance the algorithm's global optimization capability, local convergence speed, and Pareto solution quality. The core design principles and technical details of the algorithm are as follows:

[0047] 1) Algorithm Fusion Design Principles

[0048] ① Stage division of labor principle: MOPSO is responsible for the global exploration of the first 50% of the algorithm iterations, quickly traversing the solution space, mining potential high-quality solution regions, and generating a preliminary set of non-dominated solutions; NSGA-II is responsible for the local mining of the last 50% of the algorithm iterations, performing a refined search on the solution regions generated by MOPSO, optimizing the distribution uniformity and convergence of Pareto solutions, and finally obtaining a high-quality set of Pareto optimal solutions.

[0049] ② Seamless state transition principle: The particle swarm population after the MOPSO iteration is directly used as the initial parent population of NSGA-II, and the non-dominated solutions of the external archive set of MOPSO are integrated into the initial population of NSGA-II to ensure the continuity of the algorithm iteration state, avoid repeated search of the solution space, and improve the optimization efficiency of the algorithm.

[0050] ③ Global Target Collaboration Principle: For the different optimization objectives of MOPP and MOTA, a target function normalization mapping strategy is designed to uniformly map target functions with different dimensions and optimization directions to the [0,1] interval, eliminating the difference in dimensions between targets. At the same time, each target is assigned a dynamic weight coefficient, and the weight is adaptively adjusted according to the solution space distribution during the iteration process to achieve collaborative optimization of multiple objectives.

[0051] ④ Constraint global verification principle: In each iteration of the algorithm, all generated solutions are fully constrained, including UAV endurance constraints, mission uniqueness constraints, personnel movement space constraints, etc. Solutions that do not meet the constraints are adaptively corrected (by generating neighboring solutions that meet the constraints through local search of the solution space), avoiding the generation of invalid solutions and improving the robustness of the algorithm.

[0052] 2) Desolution space unified coding strategy process

[0053] To address the issues of heterogeneous optimization objects and inconsistent decision variable dimensions in multi-objective path planning (MOPP) and multi-objective task assignment (MOTA) problems, this invention designs a unified solution space encoding and decoding mechanism to achieve unified representation and efficient computation at different optimization stages.

[0054] Specifically, the task set is first standardized and modeled. Let the task set be...

[0055] ,in The total number of tasks; the drone collection is Ground workers assembled .

[0056] Based on this, each candidate solution is represented as an integer vector of length m:

[0057] ,

[0058] Among them, the element Indicates task The assigned execution entity is determined according to the following rules:

[0059] During the MOPP phase:

[0060] , indicating task Assigned to the A drone;

[0061] In the MOTA phase:

[0062] , indicating task Assigned to the One ground worker.

[0063] To ensure a computable mapping between the encoding and the actual problem, this invention further defines the following decoding process:

[0064] ①Task grouping and processing:

[0065] According to the encoding vector Divide the task set into several subsets, such as Indicates assignment to the first A set of tasks for each entity (drone or operator);

[0066] ② Path generation process:

[0067] For each entity, its execution path is generated by sorting the paths of the corresponding task subset. The sorting method is based on the nearest neighbor algorithm, generating a task access sequence:

[0068] ,

[0069] ③ Calculation of the objective function:

[0070] The path length is defined as the sum of the Euclidean distances between the executing entity and each task point in sequence from the initial position:

[0071] ,

[0072] The total path length of the system is .

[0073] Assume the movement speed of the executing entity is constant. ,Task The execution time is Then the first The total completion time for each entity is The total system task completion time is defined as the maximum completion time among all entities (i.e., the system completion time). .

[0074] Define the energy consumption coefficient per unit distance as The energy consumption per unit task execution is , No. The energy consumption of each entity is The total energy consumption of the system is .

[0075] Set each task It has weights (i.e., priorities). Define an indicator function to indicate whether the task is completed:

[0076]

[0077] The task priority satisfaction is defined as follows: .

[0078] To unify the optimization direction, it is transformed into a minimization objective. ,

[0079] ④ Fitness evaluation mapping:

[0080] The above multi-objective indicators are combined to form an objective vector:

[0081] ,

[0082] Used for non-dominated sorting and selection operations in subsequent multi-objective optimization algorithms.

[0083] By adopting the unified encoding strategy described above, the mapping from the "discrete task allocation problem" to the "integer vector optimization problem" avoids the errors caused by continuous space discretization, facilitates unified processing by the algorithm, supports fast fitness calculation, has good engineering feasibility, and improves overall computational efficiency.

[0084] 3) Core optimization strategies for the MOPSO stage

[0085] In the first stage of UAV path planning optimization, this invention is based on the multi-objective particle swarm optimization algorithm (MOPSO) and improves the particle update mechanism, archive management strategy and search mechanism to improve the optimization performance of the algorithm in complex multi-objective scenarios, taking into account the characteristics of discrete task allocation problems.

[0086] First, the particles are represented as the integer encoded vectors described above. And define its velocity vector as a real vector of the same dimension:

[0087] ,

[0088] in This indicates the update trend of the task assignment variable.

[0089] The particle update process includes the following steps:

[0090] ① Dynamic inertia weight decay strategy

[0091] To balance global exploration and local development capabilities, inertia weights are defined. Dynamically changes with the number of iterations:

[0092] ,

[0093] in:

[0094] , ,

[0095] This represents the current iteration number;

[0096] This represents the maximum number of iterations.

[0097] In each iteration, based on the current... Update particle velocity:

[0098] ,

[0099] The continuous velocity is then mapped to the discrete solution space, and rounded down.

[0100] ,

[0101] ② Adaptive external archive set update mechanism

[0102] Let the current external archive set be The newly generated solution is The objective function is defined as follows: ,

[0103] The Pareto dominance relation is defined as follows:

[0104] ,

[0105] Construct the deleted set as ,

[0106] Next, update the archive set. ,

[0107] The non-dominated addition determination is defined as follows: if the following conditions are met... Then execute the join operation. Otherwise, discard. .

[0108] To perform capacity control, the first step is to define the crowding distance. At that time, all solutions in the archive set are processed as follows:

[0109] Sort: ,

[0110] Boundary solution assignment: ,

[0111] Internal solution calculation: ,

[0112] Accumulation: ,

[0113] Then we can remove the solution with the smallest crowding distance:

[0114] ,

[0115] ,

[0116] Repeat the above process until: ,

[0117] ③ Leader selection and guidance strategies

[0118] Let the archive set be Each solution corresponds to a congestion distance. .

[0119] First, we need to construct the probability distribution, defining the selection probability as... To avoid extreme cases, a smoothing term is introduced. .

[0120] Then sampling is performed based on the probability distribution. ,in .

[0121] Finally, a leader guidance mechanism can be constructed, specifically through particle velocity updates. middle, It changes dynamically according to the above probability mechanism.

[0122] 4) Core optimization strategy for NSGA-II stage

[0123] To improve the local mining capability and Pareto solution convergence performance of the NSGA-II algorithm during the second-stage optimization of ground operation personnel task allocation, this invention improves the crowding distance calculation and population selection mechanism. The complete data calculation process is given below.

[0124] ① Improved method for calculating congestion distance

[0125] After completing the fast nondominated sort, for the same nondominated level Let the set of solutions be defined as containing There are 10 solutions, each corresponding to a multi-objective function vector:

[0126] ,

[0127] First, a target-by-target sorting process is performed, that is, for each objective function... For sets All solutions are sorted in ascending order according to the objective function value, resulting in the sorted sequence:

[0128] ,

[0129] Next, the boundary solutions are processed. For the solutions located at both ends after sorting (i.e., the 1st and the 2nd solutions),... (one solution), directly assigning its crowding distance to infinity. This ensures that the boundary solutions are not deleted in subsequent selection processes, thereby maintaining the coverage of the Pareto front.

[0130] Then, the objective function is normalized to eliminate the differences in dimensions between different objective functions. First, the maximum and minimum values ​​of the objective function are calculated. Then, the internal de-crowding distance is calculated. Specifically, for the solution at the middle position... In its first The local congestion distance on each target is ,in To prevent extremely small positive numbers with a denominator of zero.

[0131] The final congestion distance is obtained by summing up all the targets. .

[0132] Through the above improvements, the crowding distance calculation can simultaneously possess the following capabilities: boundary protection (maintaining the range of the solution set), multi-objective scale consistency (normalization processing), and distribution uniformity measurement (neighborhood difference).

[0133] ② Elite population stratification selection strategy

[0134] To ensure the preservation of high-quality solutions and population diversity, this invention employs an elite preservation mechanism to jointly screen the parent and offspring populations.

[0135] Let the parent population be The offspring population is Population size is .

[0136] First, the parent and offspring generations are merged into a mixed population. in .

[0137] Then, for mixed populations Perform non-dominated sorting to obtain multiple non-dominated levels. in This is the first non-dominated layer (optimal layer). For the first The layering was before Layer solution dominance.

[0138] Then, the new population is gradually filled in layer by layer. Specifically, the new population is first initialized. Add them sequentially according to their hierarchical order. There are two cases when adding them: Then the entire Joining a new population Otherwise, execute the truncation selection process.

[0139] The truncation selection based on crowding distance will affect the current layer. All solutions are based on crowding distance. Sort in descending order and select the top few. A solution is added to the new population.

[0140] Finally, the next generation population can be generated, resulting in a population of size [missing information]. The new generation of population .

[0141] Through the above mechanisms, the superiority of solutions is guaranteed by non-dominated sorting, the uniformity of solution distribution is guaranteed by squeezing distance, the loss of high-quality solutions is avoided by elite preservation, and the fine search capability of local solution space is improved.

[0142] 5) Pareto optimality mechanism

[0143] After completing the two-stage optimization, the algorithm outputs a set of Pareto optimal solutions. Since the Pareto solution is a set of non-dominated solutions, in order to obtain a practically executable solution, this invention designs a multi-objective optimization mechanism based on ideal points, the specific process of which is as follows.

[0144] ① Construct the objective function matrix

[0145] For each solution in the Pareto solution set Calculate its objective function vector:

[0146] ,

[0147] Construct the objective function matrix:

[0148] ,

[0149] ② Calculation of ideal points

[0150] For each objective function Calculate their optimal values ​​respectively. If the objective is minimization, then... If the goal is to maximize, then Then construct the ideal point vector: .

[0151] ③ Target normalization processing

[0152] Normalization is performed for each objective, and the calculation range is:

[0153] ,

[0154] Normalize each objective for each solution:

[0155] Minimize objective: ,

[0156] Maximize the goal: ,

[0157] At this point, all objectives are transformed into "the smaller the better".

[0158] ④ Distance calculation

[0159] For each solution Calculate its Euclidean distance to the ideal point. The ideal point, after normalization, is: Therefore, it can be simplified to: .

[0160] ⑤ Optimal solution selection

[0161] After calculating the distance for all Pareto solutions, select the one that satisfies the condition. The solution is taken as the final execution plan.

[0162] Through the above optimization mechanism, multi-objective unified measurement, global optimal trend guidance, and executable single solution output are achieved, avoiding subjective bias caused by human weight setting.

[0163] (3) Multi-target path planning for UAVs (first stage optimization)

[0164] The first stage of optimization focuses on a swarm of drones, with three main objectives: maximizing mission reward, maximizing environmental value, and minimizing total flight distance. It utilizes the EF-DH algorithm to achieve multi-objective optimization of the MOPP problem, while simultaneously collecting real-time environmental status information of the road network to provide input for the second stage of optimization. The specific implementation steps include five steps: objective function construction, constraint definition, EF-DH algorithm iterative optimization, optimal path generation, and environmental information collection. The technical details of each step are as follows:

[0165] 1) Construction of multi-objective functions

[0166] Three mutually constrained optimization objective functions are constructed, all of which are computed based on the solution encoding array X, achieving a direct mapping between the objective function and the solution space:

[0167] ①Task Reward Maximization Objective: Calculate the sum of the base rewards for all unique tasks perceived by the drone swarm, representing the drones' mission truth acquisition capability. The formula is:

[0168] ,

[0169] in The unique set of tasks perceived by the drone swarm is represented by the decoded array. With the sensing radius of drones To be determined jointly;

[0170] ② Environmental Value Maximization Objective: Calculate the sum of the environmental values ​​of all unique road segments covered by the flight path of the UAV swarm, representing the environmental information perception capability of the UAVs. The formula is:

[0171] ,

[0172] in It is the unique set of road segments covered by the drone swarm, determined by the drone's flight path and the spatial location of the road segment;

[0173] ③ Minimize total flight distance: Calculate the sum of the Euclidean lengths of all flight paths of the UAV swarm, representing the energy consumption and operational efficiency of the UAVs. The formula is:

[0174] ,

[0175] in For decoding array The generated first The flight path of the drone This is a function for calculating the Euclidean length of a path.

[0176] 2) Definition of Constraints

[0177] To address the practical operational requirements of UAV path planning, three hard constraints are defined. All solutions must satisfy these constraints; otherwise, they are considered invalid:

[0178] ① Flight path closed-loop constraint: The flight path of each UAV must start from the base. Departure, and eventual return to base. ,Right now Both the starting point and the ending point are ;

[0179] ② Flight range constraint: The flight path length of each drone must not exceed its maximum flight range, i.e. , ;

[0180] ③Perception uniqueness constraint: The perception results for the same task and the same road segment are calculated only once to avoid duplicate calculation of task rewards and environmental value.

[0181] 3) Iterative optimization of the EF-DH algorithm

[0182] Based on the designed EF-DH algorithm, the MOPP problem is iteratively optimized. The specific steps are as follows:

[0183] ① Initialization: Generation scale is The initial particle swarm (decoded array) has particle velocities initialized to 0, and the construction size is... external archive set Set the total number of iterations MOPSO and NSGA-II each performed 100 iterations;

[0184] ②MOPSO stage (iterations 1-100): Decode each particle, calculate three objective function values, complete fitness evaluation; update the individual optimal position of the particle. With the global leader; update particle velocity and position using a dynamic inertia weight decay strategy; perform constraint verification on newly generated particles and adaptively correct invalid solutions; update the external archive set. This ensures the uniformity of the distribution of non-dominated solutions;

[0185] ③NSGA-II phase (101-200 iterations): The particle swarm after the MOPSO phase ends is used as the initial parent population and integrated into the external archive set. The nondominated solution generates a size of The initial population; through single-point crossover (crossover probability) ) and random mutation (probability of mutation) Generate a child population; merge the parent and child populations, perform fast non-dominated sorting and improved crowding distance calculation; generate the next generation population through an elite population hierarchical selection strategy; verify the constraints of the newly generated population and adaptively correct invalid solutions;

[0186] ④ Generation of Pareto solution set: After the iteration is completed, all non-dominated solutions in the last generation population are used as the Pareto optimal solution set of the MOPP problem. .

[0187] 4) Optimal path generation

[0188] Based on the ideal point Pareto solution optimization mechanism of the design, from the Pareto optimal solution set Choose the unique optimal solution ;right Decode the data to generate the optimal flight path for each drone. Simultaneously determine the task perception set of the drone swarm. With road segment coverage set .

[0189] 5) Environmental Information Collection

[0190] Optimal flight path based on UAV With road segment coverage set Collect data for each covered road segment Real-time congestion level Environmental value

[0191] Construct a real-time environmental status matrix for the road network. The uncovered road segments are filled with historical statistical values ​​to ensure the integrity of the matrix; As the core input for the second phase of optimization, it enables information coupling between airborne environmental perception and ground mission optimization.

[0192] (4) Multi-target task allocation for ground personnel (second phase optimization)

[0193] The second phase of optimization focuses on ground-based personnel and utilizes real-time environmental state matrices collected by drones. With the core constraint of minimizing the maximum completion time (Makespan) and maximizing the total task value as the two major optimization objectives, this paper implements multi-objective optimization of the MOTA problem based on the EF-DH algorithm, while satisfying constraints such as task uniqueness and personnel spatial movement. The specific implementation steps include four steps: objective function construction, constraint condition update, EF-DH algorithm iterative optimization, and optimal task allocation scheme generation. The technical details of each step are deeply coupled with the first-stage optimization, and targeted optimizations are made according to the characteristics of ground task allocation.

[0194] 1) Construction of multi-objective functions

[0195] Construct two mutually constrained optimization objective functions, both based on the encoding array of the solutions. With the environmental state matrix Calculations are performed to dynamically constrain the optimization of ground missions based on environmental information:

[0196] ① Minimize maximum completion time objective: Calculate the maximum total operation time for all ground personnel to complete their assigned tasks, representing the overall efficiency of ground task execution. The formula is:

[0197] ,

[0198] in For decoding array Assigned to personnel The task set, For personnel From current location to task Real-time driving time (from the environment state matrix) congestion level correction), For personnel Execute the task Actual execution time (based on personnel competency level) (Note: The higher the ability level, the shorter the execution time).

[0199] ② Maximizing Total Mission Value: Calculate the sum of the total value of all ground personnel performing their assigned tasks, representing the overall benefit of ground mission execution. The formula is:

[0200] ,

[0201] in Value gain coefficient for personnel competency level ( The higher, The larger ( Information gain coefficient for the environmental value of road segments ( The higher, (The larger the value), the more accurate it is to make adjustments to the mission value based on both personnel capabilities and environmental information.

[0202] 2) Constraint update

[0203] Based on the constraints optimized in the first phase, and taking into account the characteristics of ground task allocation, four new hard constraints have been added to ensure the practical feasibility of the task allocation scheme:

[0204] ①Task uniqueness constraint: Each task can only be assigned to one ground personnel, i.e. ( Assign matrices to 0-1. Personnel Execute the task );

[0205] ② Spatial movement constraints: The movement paths of personnel on the ground must strictly follow the road map. The topology, the travel time is determined by the environment state matrix. congestion level Dynamic correction;

[0206] ③ Competency matching constraints: personnel competency levels The difficulty level must not be lower than the assigned task. ,Right now To avoid mission failure due to insufficient personnel capabilities;

[0207] ④ Time Feasibility Constraint: The total working time for personnel to perform all assigned tasks shall not exceed the preset maximum working time. ,Right now .

[0208] 3) Iterative optimization of the EF-DH algorithm

[0209] Based on the EF-DH algorithm designed above, the MOTA problem is iteratively optimized. The core parameters of the algorithm remain the same as in the first stage (population size 50, total number of iterations 200, crossover probability 0.9, mutation probability 0.05). Only adaptive adjustments are made according to the optimization objective and constraints of MOTA. The specific execution steps are basically the same as the first stage optimization:

[0210] ① Initialization: Generate an initial particle swarm of size 50, initialize particle velocities to 0, and construct an external archive set of size 100;

[0211] ②MOPSO phase (iterations 1-100): Combining the environment state matrix Complete the fitness assessment of particles, update the individual optimal and global leader, dynamically adjust particle velocity and position, verify and correct invalid solutions, and update the external archive set;

[0212] ③NSGA-II stage (101-200 iterations): The particle swarm from the MOPSO stage is used as the initial parent population to generate the offspring population. Fast non-dominated sorting and improved crowding distance calculation are performed. The next generation population is generated through elite hierarchical selection. Invalid solutions are verified and corrected.

[0213] ④ Generation of Pareto solution set: After the iteration is completed, the Pareto optimal solution set of the MOTA problem is generated. .

[0214] 4) Generation of optimal task allocation scheme

[0215] Based on the ideal point Pareto solution optimization mechanism, from the Pareto optimal solution set Choose the unique optimal solution ;right Decode the data to generate the optimal task allocation matrix. Clearly define the personnel responsible for each task; simultaneously, combine this with the environmental state matrix. With road map It generates the optimal task execution sequence and real-time optimal driving path for each ground personnel, enabling refined and personalized allocation of ground tasks.

[0216] (5) Air-Ground Cooperative Execution and Dynamic Re-optimization

[0217] After completing the UAV path planning (MOPP) and ground task assignment (MOTA), this invention further constructs an air-ground collaborative execution and dynamic re-optimization mechanism to realize the system's adaptive optimization capability in dynamic environments.

[0218] ① Air-Ground Coordinated Execution Mechanism

[0219] During the execution phase, the system comprises two main entities: drone swarms and ground personnel. The drones follow the path sets generated in the first phase. Perform environmental perception tasks. Additionally, ground operations personnel will follow the task allocation results from the second phase. To carry out specific tasks.

[0220] During the execution process, the drone collects environmental data in real time to form an environmental state vector. This includes the degree of traffic congestion on certain road sections. This includes regional toll costs, dynamic obstacle information, and meteorological or environmental risk indicators. The environmental data is then mapped to the ground road network model via a communication module, updating the edge weight functions. ,

[0221] ② State awareness and trigger determination mechanism

[0222] To avoid the computational overhead of frequent global re-optimization, this invention designs an event-driven re-optimization triggering mechanism. First, a system state deviation function is defined. This includes path deviation. Time deviation Intensity of environmental change When the following conditions are met Time-triggered re-optimization, among which This is a preset threshold.

[0223] ③ Modeling of dynamic re-optimization problems

[0224] When the triggering condition is met, only the "affected task subset" is locally re-optimized, rather than globally recalculated. First, the set of affected tasks is defined. .

[0225] Then, construct sub-problems and extract corresponding sub-encoding vectors. Update the road network, time, and location constraints.

[0226] ④ Incremental two-stage re-optimization algorithm

[0227] During the re-optimization phase, an "incremental initialization" strategy is adopted, which preserves the unaffected parts of the original solution during population initialization. A new solution is generated by perturbing the affected part. .

[0228] Local MOPSO optimization is performed on the UAV side, which means updating the path weights, objective function, and local optimum only for the affected area.

[0229] On the ground side, local NSGA-II optimization is performed, which involves reassigning affected tasks based on the updated environmental information of the UAV and then updating personnel paths.

[0230] Then the local optimization results are merged with the unchanged parts. .

[0231] Finally, a feedback loop mechanism is implemented. Specifically, the updated solution is redistributed to the execution layer, while environmental perception (UAV), status monitoring (system), and dynamic evaluation (trigger function) are continuously performed, forming a complete closed loop from perception, evaluation, re-optimization to execution.

[0232] This technical solution achieves end-to-end collaborative optimization of UAV path planning and ground task allocation through a layered and coupled collaborative optimization architecture and a deeply integrated EF-DH algorithm. It is not a simple combination of algorithms and stage division. From the mathematical accuracy of system modeling to the innovative integration of algorithm design, and then to the dynamic execution of air-ground collaboration, it forms a complete technical system that is theoretically rigorous, technologically advanced, and highly operable. It breaks through the technical bottlenecks of air-ground information fragmentation and low efficiency of multi-objective optimization in traditional mobile swarm perception systems, and achieves a significant improvement in the overall performance of the system.

[0233] Experimental verification

[0234] To verify the performance advantages of the proposed UAV path planning and ground task allocation method based on collaborative optimization (Co-UPP framework and EF-DH algorithm), this invention selects three representative real taxi trajectory datasets for experimental verification: the Beijing T-Drive taxi trajectory dataset (Beijing, China), the Porto taxi trajectory dataset (Porto, Portugal), and the Rome taxi trajectory dataset (Rome, Italy). All of these datasets contain vehicle GPS trajectory information (longitude, latitude, and timestamp), which can realistically reflect the spatial distribution and dynamic changes of urban road traffic.

[0235] In the experiment, the original trajectory data was first preprocessed uniformly, including map matching, road segment division, and spatial gridding, discretizing the experimental area into a 100×100 spatial grid. The road segment congestion level was calculated by normalizing the historical trajectory traffic frequency, and the environmental value was modeled by combining spatial coverage weights and uniformly normalized to the [0,1] interval to ensure the comparability of experiments between different datasets.

[0236] Regarding the setup of tasks and execution entities, the experiment generated perception tasks of varying sizes, ranging from 10 to 300. The task locations were sampled and distributed along road segments according to their environmental value. In the experiment, the drone swarm size was set at 5 drones, and the number of ground personnel was set at 8. The drones departed from a unified base location to perform environmental perception tasks and return.

[0237] To comprehensively evaluate the performance of the method of this invention, several representative algorithms were selected as comparison benchmarks, including:

[0238] (1) GREEDY: A greedy heuristic method based on task value ranking;

[0239] (2) MOPSO-DH: A two-stage method based on multi-objective particle swarm optimization;

[0240] (3) NSGA-II-DH: A two-stage optimization method based on non-dominated sorting genetic algorithm;

[0241] (4) EF: Evolutionary fusion method that does not include air-ground cooperation mechanism;

[0242] (5) EHTA (SA): Heterogeneous task allocation method based on simulated annealing.

[0243] (6) MOPSO: A single-stage method based on multi-objective particle swarm optimization;

[0244] (7) NSGA-II: A single-stage optimization method based on non-dominated sorting genetic algorithm;

[0245] The experiment was conducted under the same computing environment, and the average value of each group of experiments was taken as the final result.

[0246] To comprehensively evaluate the performance of different methods, this invention uses six metrics for evaluation, including:

[0247] (1) Total UAV revenue: The total reward obtained by the UAV in performing environmental perception tasks, used to measure aerial perception capability;

[0248] (2) Total distance of UAVs: The total distance of all UAV flight paths, used to measure flight energy consumption;

[0249] (3) Total algorithm running time: The overall running time of the algorithm, used to reflect computational efficiency;

[0250] (4) Maximum working time of workers: The longest time required for all ground workers to complete the task, used to reflect the overall efficiency bottleneck of the system;

[0251] (5) Total worker working time: The sum of the time all ground workers spend performing tasks, used to measure the consumption of ground resources;

[0252] (6) Total worker income: The total value obtained by ground staff in completing tasks, used to measure the overall income of the system.

[0253] Table 1 shows the experimental results of different methods on the three datasets. From the perspective of UAV execution performance, the method of this invention achieves superior results in terms of total UAV revenue on all three datasets. Figure 4 Meanwhile, the invention maintains a low level in the total flight distance of the UAV, indicating that it can effectively control the flight cost of the UAV while improving mission benefits, thereby achieving a good balance between benefits and energy consumption.

[0254] In terms of efficiency in executing ground tasks, the maximum working time of workers ( Figure 5Environmental awareness (UAV) is a crucial indicator for measuring the overall efficiency of a system. The method described in this invention significantly reduces this indicator across three datasets. Compared to the greedy algorithm, the maximum operation time is reduced by approximately 83%; compared to the MOPSO method, it is reduced by approximately 32%–44%; compared to the NSGA-II method, it is reduced by approximately 29%–42%; and compared to the EF method without UAV cooperative mechanisms, it is reduced by at least 29%. Furthermore, compared to the simulated annealing method EHTA (SA), the method of this invention still achieves a time reduction of approximately 7%–19%. This indicates that using UAV environmental awareness information to assist ground task allocation can effectively avoid bottlenecks in ground personnel task scheduling, thereby significantly improving overall system efficiency.

[0255] Regarding the total working time index for workers ( Figure 6 The method of this invention also outperforms most of the comparative algorithms. Experimental results show that compared with the NSGA-II method, the method of this invention can reduce the total job time by more than 39%; compared with the MOPSO and EF methods, the reduction is more than 33%; and compared with the EHTA (SA) method, it reduces the job time by an average of about 6.6%. The only exception is the greedy algorithm, whose total job time is slightly lower than that of the method of this invention (about 9%). This is because the greedy algorithm tends to use the fewest number of people to complete the task, thereby reducing the total accumulated time. However, this strategy can lead to excessive task load for individual personnel, which significantly increases the maximum job time of the system and is detrimental to the overall task efficiency.

[0256] Regarding the total worker benefits indicator ( Figure 7 The method of this invention achieves the highest results on all three datasets. Compared with the EF and MOPSO methods, the total revenue is improved by more than 43%; compared with the NSGA-II method, it is improved by more than 42%; and compared with the greedy algorithm, it is improved by about 10% to 18%, and compared with the EHTA (SA) method, it is improved by about 10% to 20%. The above results show that through the collaborative optimization of UAV environmental perception and ground task allocation, the method of this invention can significantly improve the overall system revenue and achieve a better comprehensive performance between task efficiency and revenue.

[0257] The experimental results above show that the collaborative optimization method proposed in this invention can outperform existing methods in many aspects, such as total UAV revenue, total UAV distance, maximum worker working time, and total worker revenue, thus verifying the effectiveness and practical value of the method in mobile swarm perception tasks.

[0258] Table 1. Comparison of data from different methods under six major indicators.

[0259]

[0260] Example 2

[0261] This embodiment provides a swarm intelligence perception collaborative optimization system based on evolutionary fusion dual-stage hybridization.

[0262] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned method for collaborative optimization based on evolutionary fusion dual-stage hybrid sensing.

[0263] A terminal device includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide a crowd-based intelligent perception collaborative optimization method based on evolutionary fusion dual-stage hybrid.

[0264] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A collaborative optimization method for swarm intelligence perception based on evolutionary fusion and a two-stage hybrid approach, characterized in that, include: Based on mobile swarm sensing operation scenarios, a multi-agent collaborative optimization model for heterogeneous spaces is constructed. The EF-DH algorithm is constructed by deeply fusing MOPSO and NSGA-II using a staged evolutionary fusion strategy. Based on the constructed multi-agent collaborative optimization model, the EF-DH algorithm is used for UAV multi-objective path planning, including the first stage of UAV swarm path optimization and the second stage of ground operation personnel task optimization. Air-to-ground collaborative execution and dynamic re-optimization based on path planning; The aforementioned multi-agent collaborative optimization model for heterogeneous spaces, based on mobile swarm sensing operation scenarios, includes defining a two-dimensional square operation area. The region is an orthogonal road network, containing N longitudinal roads and M transverse roads. The roads are discretized into a finite number of road segments at intersections. Construct a set of road segments Constructing a road map based on a set of road segments Where V is the set of nodes at road intersections and E is the set of edges for road segments; define the perception task set. For each task Assigning a four-dimensional attribute vector ,in The two-dimensional spatial coordinates of the task. As the basic reward for the task, The task difficulty level. The base execution time of the task, and ( () represents random noise; define the drone swarm set. All drones departed from designated bases. Departure, execution of flight mission, and return to base; each drone With a fixed sensing radius When the task spatial coordinates Euclidean distance from the drone's flight path At that time, it is determined that the task has been perceived by the drone, and the task truth data is collected; simultaneously, the road sections covered by the drone's flight path are... It was identified as an environmental information sensing device, collecting data on the congestion level of that road section. Environmental value The flight path of the drone is a polygonal path in a two-dimensional plane. ,in Define the drone's waypoints; define the set of ground personnel. For each person Assigning three-dimensional attribute vectors ,in The initial spatial location of the personnel, For personnel competency levels, This refers to the basic movement speed of personnel. The proposed method employs a phased evolutionary fusion strategy to deeply integrate MOPSO and NSGA-II, constructing the EF-DH algorithm. This algorithm includes optimization objects for UAV multi-objective path planning (MOPP) and ground personnel multi-objective task allocation (MOTA). It constructs a unified integer array encoding method based on task allocation, mapping the solution space to a finite integer space to avoid discretization errors in continuous solution spaces. Simultaneously, it ensures a one-to-one correspondence between the encoding and the actual optimization problem. The encoding rule is: the solution is encoded as an integer array of length m. ,in Assigned to task The main ID; in MOTA, it is the ground personnel ID, with a value range of [1, |W|]. To improve the global exploration capability and non-dominated solution mining efficiency of MOPSO, three optimization strategies are constructed: dynamic inertia weight decay strategy, adaptive archive set update mechanism, and leader random selection strategy. To improve the local mining capability and Pareto solution convergence of NSGA-II, two optimization strategies are designed: an improved crowding distance calculation method and an elite population hierarchical selection strategy. Among them, the improved crowding distance calculation method addresses the problem that boundary solutions are easily deleted in traditional crowding distance calculation by assigning an infinitely large crowding distance to the boundary solutions of the Pareto solution set, ensuring that the boundary solutions are not pruned and improving the coverage of the Pareto solution set. The elite population hierarchical selection strategy merges the parent and offspring populations into a mixed population and divides the mixed population into different non-dominated layers through fast non-dominated sorting. , The optimal non-dominated layer is selected. Solutions are selected from high to low priority of the non-dominated layer to generate the next generation population. If the number of solutions in the last selected non-dominated layer exceeds the population size, solutions are selected in descending order of crowding distance to ensure that the next generation population contains both high-quality non-dominated solutions and has good distribution uniformity.

2. The method for collaborative optimization of swarm intelligence perception based on evolutionary fusion and dual-stage hybridization as described in claim 1, characterized in that, The method employs a staged evolutionary fusion strategy to deeply integrate MOPSO and NSGA-II to construct the EF-DH algorithm. It also includes using the Pareto optimal solution set generated after the algorithm iterations. To select a unique optimal actual execution solution from the set, an optimization criterion based on the Euclidean distance of the ideal point is designed: First, the ideal point of the multi-objective optimization is calculated. ,in For the first The optimal value of an objective function is determined by maximizing the objective function to find its maximum value and minimizing the objective function to find its minimum value. Then calculate each solution in the Pareto solution set. To the ideal point The standardized Euclidean distance is expressed as: ， in To solve The The objective function value, and The first The maximum and minimum values ​​of each objective function in the Pareto solution set are determined. Finally, the solution with the smallest standardized Euclidean distance is selected as the optimal actual execution solution, ensuring that the selected solution is close to the optimal value for all objectives, thus achieving a comprehensive optimal balance of multiple objectives.

3. The method for collaborative optimization of swarm intelligence perception based on evolutionary fusion and dual-stage hybridization as described in claim 2, characterized in that, The first stage of UAV swarm path optimization includes constructing mutually constrained optimization objective functions: a task reward maximization objective, an environmental value maximization objective, and a total flight distance minimization objective. These are calculated based on the solution encoding array X, achieving a direct mapping between the objective functions and the solution space. The task reward maximization objective involves calculating the sum of the basic rewards for all unique tasks perceived by the UAV swarm, representing the UAVs' mission truth acquisition capability. The formula is: ,in The unique set of tasks perceived by the drone swarm is represented by the decoded array. With the sensing radius of drones The objective of maximizing environmental value, jointly determined, involves calculating the sum of the environmental values ​​of all unique road segments covered by the flight paths of the UAV swarm, representing the environmental information perception capability of the UAVs. The formula is as follows: ,in The set of unique road segments covered by the drone swarm; the objective of minimizing the total flight distance includes calculating the sum of the Euclidean lengths of all flight paths of the drone swarm, characterizing the energy consumption and operational efficiency of the drones, as shown in the formula: ,in For decoding array The generated first The flight path of the drone The Euclidean length of the path is calculated using a function. Then, considering the operational requirements of UAV path planning, trajectory closure constraints, endurance constraints, and perception uniqueness constraints are defined as hard constraints. Finally, based on the designed EF-DH algorithm, the MOPP problem is iteratively optimized. The velocity and position of particles are updated through a dynamic inertial weight decay strategy. Constraint conditions are verified for newly generated particles, and invalid solutions are adaptively corrected. The particle swarm after the MOPSO stage is used as the initial parent population, and offspring populations are generated through single-point crossover and random mutation. The parent and offspring populations are merged, and fast non-dominated sorting and improved crowding distance calculation are performed. The next generation population is generated through an elite population hierarchical selection strategy. After iteration, all non-dominated solutions in the last generation population are used as the Pareto optimal solution set for the MOPP problem. .

4. The crowd intelligence perception collaborative optimization method based on evolutionary fusion dual-stage hybridization according to claim 3, characterized in that, The first stage of drone swarm path optimization also includes a Pareto optimal solution selection mechanism based on the ideal point of the design, from the Pareto optimal solution set. Choose the unique optimal solution ;right Decode the data to generate the optimal flight path for each drone. Simultaneously determine the task perception set of the drone swarm. With road segment coverage set Optimal flight path based on UAVs With road segment coverage set Collect data for each covered road segment Real-time congestion level Environmental value Construct a real-time environmental status matrix for the road network. The uncovered road segments are filled with historical statistical values ​​to ensure the integrity of the matrix; As the core input for the second phase of optimization, it enables information coupling between airborne environmental perception and ground mission optimization.

5. The crowd intelligence perception collaborative optimization method based on evolutionary fusion dual-stage hybridization according to claim 4, characterized in that, The second stage, task optimization for ground personnel, involves optimizing the ground personnel as the primary focus. Two mutually constrained objective functions are constructed, including a maximum completion time minimization objective: calculating the maximum total time for all ground personnel to complete their assigned tasks, representing the overall efficiency of ground task execution. The formula is as follows: ,in For decoding array Assigned to personnel The task set, For personnel From current location to task Real-time driving time For personnel Execute the task The actual execution time; The objective of maximizing total mission value is to calculate the sum of the total value of all ground personnel performing their assigned tasks, representing the overall benefit of ground mission execution. The formula is as follows: ,in The value gain coefficient for personnel ability level. This is the information gain coefficient for the environmental value of a road segment.

6. The crowd intelligence perception collaborative optimization method based on evolutionary fusion dual-stage hybridization according to claim 5, characterized in that, The second phase of ground personnel task optimization also includes updating and adding four hard constraints based on the constraints of the first phase optimization and considering the characteristics of ground task allocation, to ensure the practical feasibility of the task allocation scheme: Task uniqueness constraint: Each task can only be assigned to one ground personnel, i.e. , Assign matrices to 0-1. Personnel Execute the task Spatial movement constraints: The movement paths of personnel on the ground must strictly follow the road map. The topology, the travel time is determined by the environment state matrix. congestion level Dynamic adjustment; capability matching constraints: personnel capability levels The difficulty level must be no lower than the assigned task. ,Right now To avoid task failure due to insufficient personnel capabilities; time feasibility constraint: the total work time for personnel to perform all assigned tasks must not exceed the preset maximum work time. ,Right now Finally, based on the EF-DH algorithm, the MOTA problem is iteratively optimized. The core parameters of the algorithm remain consistent with those in the first stage, and only adaptive adjustments are made according to the optimization objective and constraints of MOTA. Based on the ideal point Pareto solution optimization mechanism, the Pareto optimal solution set is selected. Choose the unique optimal solution ;right Decode the data to generate the optimal task allocation matrix. Clearly define the personnel responsible for each task; simultaneously, combine this with the environmental state matrix. With road map It generates the optimal task execution sequence and real-time optimal driving path for each ground personnel, enabling refined and personalized allocation of ground tasks.

7. The crowd intelligence perception collaborative optimization method based on evolutionary fusion dual-stage hybridization according to claim 6, characterized in that, The path planning-based air-to-ground collaborative execution and dynamic re-optimization includes constructing an air-to-ground collaborative execution system and a dynamic re-optimization mechanism to ensure dynamic adaptability and robustness in actual operational scenarios. This enables real-time adjustment of UAV flight paths and ground task allocation schemes. Specifically, based on the optimal UAV flight paths and optimal ground task allocation schemes generated in the first and second stages of optimization, a trigger-based dynamic re-optimization strategy is designed to achieve real-time adjustment of the optimization schemes. This includes re-optimization trigger conditions: when the system detects any of the following situations, dynamic re-optimization is triggered, and local re-optimization of the MOPP or MOTA problem is performed based on the current operational status and the latest environmental information: if only environmental information changes, only the MOTA problem is locally re-optimized, and the ground task allocation scheme is quickly adjusted; if the task set or equipment status changes, the MOPP problem is first locally re-optimized, and the UAV flight path and environmental state matrix are updated. Then, a local re-optimization is performed on the MOTA problem. During the local re-optimization, the current optimal solution is used as the initial population of the EF-DH algorithm, which greatly reduces the number of iterations of the algorithm, realizes the rapid generation of the optimization scheme, ensures the continuity of the operation process, and meets the near real-time requirements of actual operations.

8. A swarm intelligence perception collaborative optimization system based on evolutionary fusion two-stage hybrid approach, executing the swarm intelligence perception collaborative optimization method based on evolutionary fusion two-stage hybrid approach as described in claim 1, characterized in that, include: The data acquisition module is configured to construct a multi-agent collaborative optimization model in a heterogeneous space based on the mobile swarm perception operation scenario. The algorithm fusion module is configured to use a staged evolutionary fusion strategy to deeply fuse MOPSO and NSGA-II to construct the EF-DH algorithm. The model optimization module is configured to perform multi-objective path planning for UAVs based on the constructed multi-agent collaborative optimization model and the EF-DH algorithm, including the first stage of UAV swarm path optimization and the second stage of ground operation personnel task optimization. The re-optimization module is configured to perform air-to-ground collaborative execution and dynamic re-optimization based on path planning.

Images