A Task Allocation Method Based on Dynamic Multi-Objective Optimization in Time-Varying Mobile Crowd Sensing
By constructing a dynamic multi-objective optimization model and the DPPS-EC algorithm, the problem of balancing travel costs and worker convenience in task allocation under time-varying environments in mobile swarm intelligence sensing was solved, achieving rapid response and efficient task allocation, thereby improving system efficiency and worker experience.
Patent Information
- Application Number
- CN202611283611.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
Existing mobile swarm intelligence sensing task allocation methods struggle to effectively balance travel costs and worker convenience in time-varying environments. Furthermore, traditional algorithms exhibit slow convergence speeds in dynamic environments, difficulty in tracking Pareto fronts, and insufficient constraint handling capabilities.
A task allocation method based on dynamic multi-objective optimization is adopted. By constructing a dynamic multi-objective task allocation model that considers potential workers and time-varying environment, the dynamic push-pull search evolutionary algorithm (DPPS-EC) is used to solve the problem. Combined with dual-population co-evolution and selection strategy, the Pareto front is quickly tracked to generate non-dominated solution set.
It enables rapid response to changes in worker status in time-varying environments, reduces task interruption rates, optimizes travel costs and worker convenience, increases worker participation, and efficiently searches for Pareto fronts under complex constraints.
Smart Images

Figure CN123117635A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mobile swarm intelligence sensing technology, and in particular to a task allocation method based on dynamic multi-objective optimization in time-varying mobile swarm intelligence sensing. Background Technology
[0002] Mobile Crowd Sensing (MCS) utilizes smart devices carried by the public (workers) to complete large-scale urban data collection tasks, and has wide applications in traffic monitoring, environmental monitoring, and edge computing. Task allocation is the core of the MCS system, directly affecting task completion quality, worker experience, and platform revenue.
[0003] Existing research on task allocation mainly falls into two categories: static scheduling and dynamic scheduling. Static scheduling methods assume a stable environment and cannot cope with time-varying factors such as sudden worker offlineness and dynamic changes in tasks. While a few dynamic scheduling methods consider changes in worker online status, they only focus on the reallocation of idle workers, neglecting the inconvenience caused by disruptions to the task plans of "potential workers" (i.e., workers already performing tasks). If the reallocation scheme alters the original routes of potential workers, it increases their travel costs and reduces their willingness to participate, thus affecting the overall efficiency of the system. Furthermore, existing research lacks an effective balance between multiple conflicting objectives in task reallocation (such as minimizing travel costs and maximizing worker convenience), typically optimizing only a single objective, leading to the loss of control over other objectives.
[0004] At the technical level, traditional genetic algorithms converge slowly in dynamic environments and have difficulty tracking the Pareto front; evolutionary algorithms based on single populations are prone to getting trapped in local optima; existing constraint handling methods are insufficient in exploring feasible regions and have difficulty finding high-quality non-dominated solutions under complex constraints. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of existing methods and propose a task allocation method based on dynamic multi-objective optimization in time-varying mobile swarm intelligence sensing. This invention constructs a dynamic multi-objective task allocation model considering potential workers and time-varying environments, designs a worker convenience measurement method based on angle and distance, and for the first time comprehensively evaluates the dual impact of task reassignment on worker travel direction and distance. Simultaneously, environmental changes (worker offline or task completed) are modeled as a dynamically constrained multi-objective optimization problem, and solved using a dynamic push-pull search evolutionary algorithm (DPPS-EC) based on relaxation and individual contributions. This algorithm, through dual-population co-evolution, dynamic change response, and two novel selection strategies, achieves rapid tracking of the Pareto front in time-varying environments, obtaining an optimal task allocation scheme that balances multiple conflicting objectives.
[0006] The technical solution adopted in this invention is as follows:
[0007] In a first aspect, this invention proposes a task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing, comprising the following steps:
[0008] S1. Construct a mobile swarm intelligence sensing system, the system including a task requester, a scheduler, multiple worker terminals and multiple sensing tasks; each sensing task includes geographical coordinates and allocation status; each worker terminal includes current location, original task plan and newly allocated task set; the scheduler maintains the set of tasks to be allocated and the set of online worker terminals, and monitors task completion status and worker terminal online status;
[0009] S2. Detect environmental change events, including worker terminal offline or sensing task completion;
[0010] S3. When environmental changes are detected, a dynamic multi-objective task allocation model is established with the objectives of minimizing the average travel cost of worker terminals and minimizing the loss of convenience of worker terminals.
[0011] S4. The model is solved using a dynamic push-pull search evolution algorithm, which includes dual-population co-evolution, environmental selection based on population constraint index, and pair selection based on individual contribution, to generate a non-dominated solution set.
[0012] S5. Select a task allocation scheme from the non-dominated solution set and send the allocation instruction to the corresponding worker terminal for execution.
[0013] Furthermore, the dynamic multi-objective task allocation model includes:
[0014] Objective function:
[0015] Minimize average travel cost : Where n is the total number of online worker terminals, For online worker terminals, For worker terminals The total travel distance for executing the newly assigned task set;
[0016] Minimize worker convenience loss : ;in, For worker terminals The ease of executing newly assigned task sets;
[0017] The constraints include: each task can only be assigned to one worker terminal, each worker terminal must be assigned at least one task, and the number of tasks assigned to each worker terminal cannot exceed its maximum carrying capacity.
[0018] Furthermore, the convenience loss is calculated based on the angle between the displacement vectors of the worker's terminal and the change in path length between the original task plan and the newly assigned task, using the following formula:
[0019] ;
[0020] in, For worker terminals Total travel distance before redistribution For worker terminals Total travel distance after redistribution The angle between the displacement vectors before and after the redistribution.
[0021] Furthermore, S4 includes:
[0022] S41: Initialize the first subpopulation and the second subpopulation. Each individual in the population is represented by a label encoding method to indicate the task allocation scheme. The task allocation scheme is represented by a two-dimensional matrix, where the rows of the matrix correspond to the worker terminals and the columns correspond to the task sequences assigned to the worker terminals. The first subpopulation ignores the constraints and searches only with the objective function minimization as the guide. The second subpopulation searches the feasible region under the constraints.
[0023] S42: Determine whether the first subpopulation converges to the unconstrained Pareto front;
[0024] S43: Perform push-pull cooperative search through information exchange between two subpopulations, including:
[0025] When the first subpopulation fails to converge, select solutions from the second subpopulation that are not dominated by the first subpopulation and merge them with it. After non-dominated sorting and selection, offspring are generated, and feasible and non-dominated solutions from the offspring are added to the second subpopulation.
[0026] When the first subpopulation has converged, extract feasible and non-dominated solutions from the first subpopulation and add them to the second subpopulation.
[0027] S44: Perform pairing selection based on individual contribution on the second subpopulation to generate offspring;
[0028] S45: Perform environmental selection based on the population constraint index on the second subpopulation, eliminating individuals and reducing the population size;
[0029] S46: Use eliminated individuals to update the first subpopulation;
[0030] S47: Repeat steps S42 to S46 until the termination condition is met, and output the non-dominated solution set in the second subpopulation.
[0031] Furthermore, in S43, when the first subpopulation fails to converge, the specific steps include: selecting solutions from the second subpopulation that are not dominated by any individual in the first subpopulation, merging them with the first subpopulation to form a temporary population; performing a non-dominated sort on the temporary population to obtain the dominance level of each individual; using a binary tournament selection method to select parents based on the dominance level, and generating offspring through genetic operators; extracting feasible and non-dominated solutions from the offspring and adding them to the second subpopulation.
[0032] Furthermore, in S44, the pairing selection based on individual contribution includes: generating a reference vector set to divide the target space, associating each non-dominated solution with the nearest reference vector; assigning higher selection weights to non-dominated solutions in sparse regions according to the number of associated solutions, generating a mating pool by selecting non-dominated solutions through roulette wheel selection, and then generating offspring.
[0033] Further, in S45, the environment selection based on the population constraint index includes: calculating the population constraint index pci, where pci is the ratio of the number of feasible individuals in the current population to the population size; generating a random number, and if the random number is less than pci, then the individual with the highest dominance level is selected as the first candidate A; otherwise, the individual with the greatest constraint violation is selected as the first candidate A; finding the second candidate B in the target space that is closest to the first candidate in terms of angular distance, and eliminating one of them according to the constraint satisfaction and dominance level, repeating until the population size is reduced to a preset size.
[0034] Furthermore, the degree of constraint violation is defined as the sum of the degree to which an individual violates all constraints:
[0035] ;
[0036] in, Let s be the degree of constraint violation for individual s. Let K be the violation amount of the k-th constraint. If the constraint is satisfied, it is 0; otherwise, the absolute value of the violation amount is taken. K is the number of constraints.
[0037] Furthermore, the elimination is based on either constraint satisfaction or dominance level, with the specific rules executed according to the following priority:
[0038] If both A and B violate the constraint, the one with the greater constraint violation will be eliminated.
[0039] If A and B have different levels of dominance, the one with the higher level will be eliminated.
[0040] Otherwise, one will be randomly eliminated.
[0041] Secondly, this invention proposes a task allocation system based on dynamic multi-objective optimization in time-varying mobile swarm intelligence sensing, which is used to implement the above-mentioned task allocation method based on dynamic multi-objective optimization in time-varying mobile swarm intelligence sensing.
[0042] The beneficial effects of this invention are:
[0043] This invention can detect changes in worker terminal status in real time, including offline status or task completion, and quickly trigger re-optimization to output a non-dominated allocation scheme adapted to the current environment, significantly reducing task interruption rate. Simultaneously, it optimizes the conflicting objectives of travel cost and worker convenience by providing a series of non-dominated solutions, allowing the scheduler to flexibly choose according to actual needs, avoiding the loss of control of one objective due to single-objective optimization. Worker convenience measurement relies only on the displacement vectors before and after reassignment, without requiring worker destination information, effectively protecting worker privacy and increasing worker participation willingness. Furthermore, the combination of dual-population push-pull collaboration, gradual selection, and contribution selection strategies enables the algorithm to efficiently search the Pareto front even in complex constraints and time-varying environments, outperforming existing mainstream algorithms in multiple performance metrics. Attached Figure Description
[0044] Figure 1 This is a schematic flowchart of the method of the present invention;
[0045] Figure 2 A schematic diagram of the mobile crowd intelligence sensing system model constructed in this invention;
[0046] Figure 3 This is a schematic diagram of the worker convenience measurement of the present invention (path changes and included angles before and after redistribution).
[0047] Figure 4 This is a schematic diagram of the population coding of the present invention. Detailed Implementation
[0048] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.
[0049] Example 1: This example uses city-level mobile air quality monitoring as an application scenario. In this scenario, the municipal government's environmental protection department, as the task requester, publishes air quality data collection tasks for multiple monitoring locations through a crowdsourcing platform. A large number of volunteers (i.e., workers) use their smartphones or sensors to complete these tasks during their commutes or outings. The scheduler, acting as a cloud-based intelligent decision-making center, is responsible for dynamically allocating the continuously increasing monitoring tasks to suitable volunteers in real time, maximizing their participation experience (i.e., convenience) while reducing their detour costs (i.e., travel costs). This scenario has typical time-varying characteristics; for example, volunteers may log off at any time or complete a monitoring task along their original route.
[0050] Reference Figure 1This invention provides a time-varying mobile swarm intelligence sensing task allocation method based on a dynamic multi-objective evolutionary algorithm. The method includes:
[0051] S1: Constructing a Mobile Crowd Sensing System Model
[0052] like Figure 2 As shown, this embodiment considers an MCS system that includes a task requester, a sensed task, a worker, and a scheduler.
[0053] The task requester is the initiator of the task on the crowdsourcing platform, that is, a user entity that publishes task requirements through the platform, expecting workers to go to a designated location to complete specific work. Each task requester corresponds to at least one task to be executed, which has clear geographical location attributes and work content requirements. After the task requester submits the task through the crowdsourcing platform, the scheduler allocates the task based on the status information of the currently available workers. When a task is successfully executed by a worker, the task is removed from the set of tasks to be assigned; when a worker goes offline or abnormally exits during execution, the assigned but uncompleted tasks will be reassigned by the scheduler triggering an environmental change response mechanism.
[0054] All unfinished tasks will be included in the perception task set. Each task Includes geographic coordinates and allocation status .
[0055] Online worker aggregation each worker Includes current location Original mission plan (Ordered set) and newly assigned task set In this embodiment, workers already have task plans before being assigned new tasks. The scheduler periodically monitors the completion status of all tasks and the online status of workers.
[0056] S2: Initialize the optimization algorithm and its parameters
[0057] Set the total population size proportional parameters Specifically, there are two populations: PopUncpf (Unconstrained Pareto Front Population) and PopCpf (Constrained Pareto Front Population), each with 50 individuals. PopUncpf is used to explore infeasible regions, while PopCpf is used to focus on feasible regions. The size ratio of the two populations is determined by a parameter. Task allocation. The maximum number of function evaluations is set to 40,000, and the crossover probability is 0.1. The scheduler acts as the decision center, generating the task allocation matrix. A label encoding method is used, with each matrix representing a task allocation scheme, such as... Figure 4 As shown, each row represents an ordered sequence of tasks assigned to a worker. The task assignment matrix is a Boolean matrix. The element is located in the i-th row and j-th column. ,in Indicates task Assigned to workers For example, a row in a matrix This indicates that the corresponding workers must complete the tasks in this order. , and .
[0058] During the initialization phase, each worker's original task plan will be read and populated. For unassigned workers and tasks, the tasks are randomly shuffled and then assigned sequentially to ensure that the assignment results meet the constraints.
[0059] S3: Real-time monitoring of worker status and task completion, establishing a dynamic multi-objective task allocation model.
[0060] The main detection is for two types of events: (1) a worker goes offline; (2) a worker completes a task.
[0061] Upon detecting an environmental change, the scheduler collects the status information of all currently online workers and incomplete tasks, and establishes the following dynamic multi-objective optimization model:
[0062] The objective function includes:
[0063] Minimize average travel cost : ;
[0064] Minimize worker convenience loss : ;
[0065] The constraints include:
[0066] Each task can only be assigned to one worker: ;
[0067] Each worker must be assigned at least one task: ;
[0068] The number of tasks assigned to each worker shall not exceed their maximum carrying capacity. : ;
[0069] in, , Indicates task Assigned to workers , Indicates task Not allocated to workers n is the total number of workers. For workers The total travel distance of the new task set after the reassignment is calculated using Euclidean distance after WGS84 coordinate to ENU coordinate conversion, so as to approximate the arc distance on the Earth's surface as the straight distance in the local plane, thus simplifying the calculation. For workers The ease of implementing the new task set after reallocation is discussed; the calculation process is described in [link to calculation]. Figure 3 Specifically, the process involves first obtaining the worker's displacement vector before the redistribution. and the displacement vector of the workers after redistribution , Indicates workers The current position coordinates, Indicates workers The position of the last task completed in the original task plan before the reallocation. Indicates workers The last task completed in the newly reassigned task list. The function name is used to calculate the two-dimensional displacement vector from the starting point to the ending point; then the included angle is calculated. Finally, the Law of Cosines was used to calculate... .here, For workers The total travel distance to complete the original mission plan.
[0070] S4: Solve the model using a dynamic push-pull search evolutionary algorithm.
[0071] This step uses the DPPS-EC algorithm to solve the problem. The specific sub-steps are as follows:
[0072] S4.1: Two-population co-evolution. PopUncpf ignores constraints and searches solely with the objective function minimization in mind; PopCpf searches for feasible solutions while satisfying constraints.
[0073] S4.2: Calculate whether the distribution of PopUncpf has changed between the two generations. Specifically, use the distribution of the population in the target space as evidence to determine whether PopUncpf has converged to the unconstrained Pareto front. If the population has changed compared to the previous generation, it is determined that it has not converged, and S4.3 is executed to continue evolving PopUncpf. Otherwise, it is determined that it has converged, and jumps to S4.4.
[0074] S4.3: Select all solutions from PopCpf that are not dominated by any individual in PopUncpf, and merge them with PopUncpf to form a temporary population. Sort this temporary population by non-dominance, obtaining the dominance level FrontNo for each individual. Use a binary tournament (based on FrontNo) to select parents, and generate a progeny population Offu using genetic operators. Extract feasible and non-dominant solutions from Offu and add them to PopCpf.
[0075] S4.4: Extract feasible and non-dominated solutions from PopUncpf and add them to PopCpf.
[0076] S4.5: Implement a pairing selection strategy based on individual contributions for PopCpf.
[0077] First, a set of reference vectors uniformly distributed in the target space is generated. Specifically, the systematic method of Das and Dennis is adopted, in a... dimension( In this embodiment, the target number is... On the unit hyperplane, by setting the step size To generate A set of reference vectors is formed by drawing rays from the origin to each reference point, where the reference points are uniformly distributed. These reference vectors divide the target space into Each sector. For example, when At that time, five reference vectors can be generated, dividing the two-dimensional target space into five angular regions.
[0078] Next, the population objective value is normalized, PopCpf is non-dominated and sorted, non-dominated solutions are extracted, and the angular distance between each non-dominated solution and the reference vector is calculated. Each solution is then associated with the nearest reference vector. For solutions associated with the same reference vector, their selection weight is reduced; for solutions in sparse sectors, their selection weight is increased. The weight calculation formula is as follows: ,in The number of non-dominated solutions associated with the same reference vector. This is the total number of reference vectors associated with at least one solution among all current non-dominated solutions (i.e., the number of reference vectors with solutions). The weight represents the probability that an individual will be selected as a parent in pairing selection. Non-dominated solutions are selected using a roulette wheel, adding the individual to the mating pool with a probability of 0.5; otherwise, adding its parent to the mating pool. This process is repeated until the mating pool size reaches [a certain value]. Offspring Offc are generated through genetic operations.
[0079] S4.6: Implement a lenient environment selection strategy for PopCpf.
[0080] First, merge the offspring Offc with PopCpf, and normalize the population target value. Then calculate... The value range is [0, 1]. Generate random numbers. ,like Then, perform a non-dominated sort on the current population PopCpf, obtain the dominance level FrontNo for each individual, and select the individual with the highest dominance level as a candidate. Otherwise, select the individual with the highest constraint violation as... Specifically, individuals Constraint violation degree Defined as the sum of the degree to which the individual violates all constraints (C1, C2, C3), i.e. ,in Indicates the first The violation amount of a constraint is 0 if the constraint is satisfied, otherwise the absolute value of the violation amount is taken. The larger the value, the more severely the individual deviates from the feasible region. Then find the... The individual with the closest angular distance in the target space. One of them will be eliminated according to the following rules:
[0081] like and All violated the rules; those with higher violations were eliminated.
[0082] like and The hierarchy of dominance differs, while the hierarchy of elimination is higher;
[0083] Otherwise, one will be randomly eliminated.
[0084] Repeat the above process until the population size is reduced to The eliminated individuals are denoted as... .
[0085] S4.7: Will It merges with PopUncpf and adopts a comparison method of first dominance level and then crowding distance (the distance between individuals under the same dominance level in the target space) for environment selection, and updates PopUncpf.
[0086] Throughout the evolutionary process, PopUncpf is updated through the following means: 1) being generated during initialization; 2) merging with superior individuals of PopCpf in S4.3 to produce offspring; and 3) absorbing individuals eliminated by PopCpf in S4.7.
[0087] S4.8: Repeat the above steps until the maximum number of function evaluations is reached.
[0088] S5: Output the non-dominated solution set as a task allocation scheme.
[0089] When the algorithm terminates, it outputs all non-dominated solutions in PopCpf. Each solution corresponds to a feasible task assignment scheme (i.e., the assignment matrix). The scheduler selects a solution from the solution set based on actual needs (such as a preference for lower travel costs, higher worker convenience, or a normalized objective weighting based on actual conditions) and sends the allocation instructions to the corresponding workers for execution.
[0090] This embodiment applies the method of the present invention to a specific mobile crowdsourcing sensing scenario: urban environmental monitoring. It assumes there are 30 workers (20 of whom are potential line workers) and 45 sensing tasks (such as air quality detection, noise monitoring, etc.) within a city area. After initial task allocation, a potential line worker... The sudden offline status caused the original task schedule to be interrupted. After detecting this change, the scheduler executes the following process:
[0091] (1) Collect the current location, task status, and original task plan of all workers.
[0092] (2) A dynamic multi-objective model is established with the goal of minimizing average travel costs and minimizing worker convenience loss.
[0093] (3) Call the DPPS-EC algorithm, setting N=100, R=0.5, and a maximum number of evaluations of 40,000. After running the algorithm, a set of non-dominated solutions is obtained, for example:
[0094] Option A: Lower average travel costs, but significant loss of convenience for some workers;
[0095] Option B: Less loss of worker convenience, but higher average travel costs;
[0096] Option C: A compromise between the two.
[0097] (4) The scheduler selects option B based on the current strategy (e.g., prioritizing worker experience) and sends the reassigned tasks to the relevant workers. and The task plan was adjusted, but the adjustment was minor, and the loss of convenience was within an acceptable range.
[0098] Through the above process, the method of the present invention successfully addresses the problem of task interruption caused by workers suddenly going offline, and achieves a reasonable balance between travel costs and worker convenience.
[0099] Example 2
[0100] The embodiment also provides a task allocation system based on dynamic multi-objective optimization in time-varying mobile swarm sensing, corresponding to the task allocation method based on dynamic multi-objective optimization in time-varying mobile swarm sensing provided in the above embodiments, which includes:
[0101] The system modeling module is used to construct a mobile swarm intelligence sensing system. The system includes a task requester, a scheduler, multiple worker terminals, and multiple sensing tasks. Each sensing task includes geographic location coordinates and allocation status. Each worker terminal includes its current location, original task plan, and newly allocated task set. The scheduler maintains the set of tasks to be allocated and the set of online worker terminals, and monitors the task completion status and the online status of worker terminals.
[0102] An environmental detection unit is used to detect environmental change events, including worker terminal offline or sensing task completion.
[0103] The model building unit is used to establish a dynamic multi-objective task allocation model with the objectives of minimizing the average travel cost of worker terminals and minimizing the convenience loss of worker terminals.
[0104] An optimization solution unit is used to solve the model using a dynamic push-pull search evolution algorithm. The algorithm includes dual-population co-evolution, environment selection based on population constraint index, and pair selection based on individual contribution, generating a non-dominated solution set.
[0105] The allocation execution unit is used to select a task allocation scheme from the non-dominated solution set and send the allocation instruction to the corresponding worker terminal for execution.
[0106] It should also be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the embodiments provided in this application, the division of steps or modules in the system and method is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or steps may be combined or integrated together, and a module or step may also be split.
[0107] The system embodiments of the present invention can be applied to any device with data processing capabilities, such as a computer or other similar device. The system embodiments can be implemented in software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution.
[0108] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.
Claims
1. A task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing, characterized in that, Includes the following steps: S1. Construct a mobile swarm intelligence sensing system, the system including a task requester, a scheduler, multiple worker terminals and multiple sensing tasks; each sensing task includes geographical coordinates and allocation status; each worker terminal includes current location, original task plan and newly allocated task set; the scheduler maintains the set of tasks to be allocated and the set of online worker terminals, and monitors task completion status and worker terminal online status; S2. Detect environmental change events, including worker terminal offline or sensing task completion; S3. When environmental changes are detected, a dynamic multi-objective task allocation model is established with the objectives of minimizing the average travel cost of worker terminals and minimizing the loss of convenience of worker terminals. S4. The model is solved using a dynamic push-pull search evolution algorithm, which includes dual-population co-evolution, environmental selection based on population constraint index, and pair selection based on individual contribution, to generate a non-dominated solution set. S5. Select a task allocation scheme from the non-dominated solution set and send the allocation instruction to the corresponding worker terminal for execution.
2. The task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing according to claim 1, characterized in that, The dynamic multi-objective task allocation model includes: Objective function: Minimize average travel cost : Where n is the total number of online worker terminals, For online worker terminals, For worker terminals The total travel distance for executing the newly assigned task set; Minimize worker convenience loss : ;in, For worker terminals The ease of executing newly assigned task sets; The constraints include: each task can only be assigned to one worker terminal, each worker terminal must be assigned at least one task, and the number of tasks assigned to each worker terminal cannot exceed its maximum carrying capacity.
3. As described in claim 2, characterized in that, The convenience loss is calculated based on the angle between the displacement vectors of the worker's terminal and the change in path length between the original task plan and the newly assigned task. The specific formula is as follows: ; in, For worker terminals Total travel distance before redistribution For worker terminals Total travel distance after redistribution The angle between the displacement vectors before and after the redistribution.
4. The task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing according to claim 1, characterized in that, S4 include: S41: Initialize the first subpopulation and the second subpopulation. Each individual in the population is represented by a label encoding method to indicate the task allocation scheme. The task allocation scheme is represented by a two-dimensional matrix, where the rows of the matrix correspond to the worker terminals and the columns correspond to the task sequences assigned to the worker terminals. The first subpopulation ignores the constraints and searches only with the objective function minimization as the guide. The second subpopulation searches the feasible region under the constraints. S42: Determine whether the first subpopulation converges to the unconstrained Pareto front; S43: Perform push-pull cooperative search through information exchange between two subpopulations, including: When the first subpopulation fails to converge, select solutions from the second subpopulation that are not dominated by the first subpopulation and merge them with it. After non-dominated sorting and selection, offspring are generated, and feasible and non-dominated solutions from the offspring are added to the second subpopulation. When the first subpopulation has converged, extract feasible and non-dominated solutions from the first subpopulation and add them to the second subpopulation. S44: Perform pairing selection based on individual contribution on the second subpopulation to generate offspring; S45: Perform environmental selection based on the population constraint index on the second subpopulation, eliminating individuals and reducing the population size; S46: Use eliminated individuals to update the first subpopulation; S47: Repeat steps S42 to S46 until the termination condition is met, and output the non-dominated solution set in the second subpopulation.
5. The task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing according to claim 4, characterized in that, In S43, when the first subpopulation does not converge, the specific steps include: selecting solutions from the second subpopulation that are not dominated by any individual in the first subpopulation and merging them with the first subpopulation to form a temporary population; performing a non-dominated sort on the temporary population to obtain the dominance level of each individual; using a binary tournament selection method to select the parent generation based on the dominance level and generating offspring through genetic operators; extracting feasible and non-dominated solutions from the offspring and adding them to the second subpopulation.
6. The task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing according to claim 4, characterized in that, In S44, the pairing selection based on individual contribution includes: generating a reference vector set to divide the target space, associating each non-dominated solution with the nearest reference vector; assigning higher selection weights to non-dominated solutions in sparse regions according to the number of associated solutions, generating a mating pool by selecting non-dominated solutions through roulette wheel selection, and then generating offspring.
7. The task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing according to claim 4, characterized in that, In S45, the environment selection based on the population constraint index includes: calculating the population constraint index pci, where pci is the ratio of the number of feasible individuals in the current population to the population size; generating a random number, and if the random number is less than pci, then the individual with the highest dominance level is selected as the first candidate A; otherwise, the individual with the greatest constraint violation is selected as the first candidate A; finding the second candidate B in the target space that is closest to the first candidate in terms of angular distance, and eliminating one of them according to the constraint satisfaction and dominance level, repeating until the population size is reduced to a preset size.
8. The task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing according to claim 7, characterized in that, Constraint violation degree is defined as the sum of the degree to which an individual violates all constraints: ; in, Let s be the degree of constraint violation for individual s. Let K be the violation amount of the k-th constraint. If the constraint is satisfied, it is 0; otherwise, the absolute value of the violation amount is taken. K is the number of constraints.
9. The task allocation method based on dynamic multi-objective optimization in time-varying mobile crowd sensing according to claim 7, characterized in that, The elimination is based on either constraint satisfaction or dominance level, with specific rules executed according to the following priority: If both A and B violate the constraint, the one with the greater constraint violation will be eliminated. If A and B have different levels of dominance, the one with the higher level will be eliminated. Otherwise, one will be randomly eliminated.
10. A task allocation system based on dynamic multi-objective optimization in time-varying mobile swarm sensing, used to implement the task allocation method based on dynamic multi-objective optimization in time-varying mobile swarm sensing as described in claim 1, characterized in that, The system includes: The system modeling module is used to construct a mobile swarm intelligence sensing system. The system includes a task requester, a scheduler, multiple worker terminals, and multiple sensing tasks. Each sensing task includes geographic location coordinates and allocation status. Each worker terminal includes its current location, original task plan, and newly allocated task set. The scheduler maintains the set of tasks to be allocated and the set of online worker terminals, and monitors the task completion status and the online status of worker terminals. An environmental detection unit is used to detect environmental change events, including worker terminal offline or sensing task completion. The model building unit is used to establish a dynamic multi-objective task allocation model with the objectives of minimizing the average travel cost of worker terminals and minimizing the convenience loss of worker terminals. An optimization solution unit is used to solve the model using a dynamic push-pull search evolution algorithm. The algorithm includes dual-population co-evolution, environment selection based on population constraint index, and pair selection based on individual contribution, generating a non-dominated solution set. The allocation execution unit is used to select a task allocation scheme from the non-dominated solution set and send the allocation instruction to the corresponding worker terminal for execution.
