A crowd-sensing task scheduling method and system based on multi-space modeling and fairness reinforcement learning

Abstract

The present invention relates to the intersection of crowd sensing technology and artificial intelligence technology, and in particular to a crowd sensing task scheduling method and system based on multi-space modeling and fairness reinforcement learning. The method comprises constructing a multi-layer space task structure for ground and air collaboration based on acquired tasks; establishing a topological association model between tasks based on spatial region division results, and clarifying the priority sequence of task scheduling through a critical path extraction mechanism; constructing a differentiated intelligent agent matching strategy, and rationally configuring critical path tasks based on the priority sequence; and dynamically allocating non-critical path tasks using a fairness-aware multi-objective scheduling algorithm. The present invention reduces the search space through space division and path screening mechanisms, improves the task completion rate and the overall scheduling performance of the system, and is suitable for the efficient resource allocation requirements in multi-space collaborative sensing task scenarios.

Description

Technical Field

[0001] The present invention relates to the intersection of crowd sensing and artificial intelligence optimization technology, and in particular to a crowd sensing task scheduling method and system based on multi-space modeling and fairness reinforcement learning. Background Art

[0002] With the rapid development of information technology and the widespread use of intelligent devices, crowd sensing has become an important means of large-scale data collection and intelligent analysis, with widespread applications in environmental monitoring, urban management, public safety, and traffic perception. Traditional crowd sensing task allocation, often based on two-dimensional planar models, struggles to meet the demands of three-dimensional urban space, where tasks are highly heterogeneous and resources are highly dynamic. This is particularly true in coupled multi-space environments, such as those on the ground, in the air, and underground, where tasks are diverse and prioritized, and where sensing device capabilities, network bandwidth, and energy constraints vary significantly. This places even higher demands on task allocation.

[0003] Existing task allocation methods mostly focus on a single spatial level or static environment setting, ignoring the coupling relationship between multi-spatial tasks and the need for coordinated scheduling between tasks. This can easily lead to problems such as uneven resource allocation, low task completion rates, and reduced overall system efficiency. Furthermore, in multi-spatial perception environments, task priority scheduling and fairness assurance also face significant challenges. Traditional methods often fail to balance task urgency with agent resource load, resulting in overuse of high-load agents and wasted resources for low-load agents.

[0004] On the other hand, although some studies have attempted to introduce multi-objective optimization algorithms, reinforcement learning and graph structure modeling to improve task allocation efficiency, in the face of multi-space and multi-level dynamic task flow scenarios, there is still a lack of systematic modeling and optimization of spatial hierarchy division, task group collaborative scheduling, critical path identification and long-term fairness constraints.

[0005] Therefore, there is an urgent need for a fair multi-space crowd sensing task allocation method and system that can integrate spatial clustering, reinforcement learning and multi-objective optimization mechanisms, which can not only achieve high efficiency in task allocation, but also ensure the fairness of task execution and the rationality of resource utilization, so as to adapt to the increasingly complex crowd sensing application scenarios. Summary of the Invention

[0006] In order to solve the problems existing in the existing multi-space crowd perception task allocation process, such as complex task space structure, frequent dynamic changes in resources, strong task heterogeneity, low task scheduling efficiency and poor fairness in intelligent agent allocation, the present invention provides a crowd perception task scheduling method and system based on multi-space modeling and fairness reinforcement learning, which can efficiently realize task division, critical path identification and fairness-aware scheduling in a multi-layer spatial environment, thereby improving the task completion rate, scheduling efficiency and agent participation fairness of the crowd perception system.

[0007] In a first aspect, the present invention provides a method for group intelligence perception task scheduling based on multi-space modeling and fairness reinforcement learning, which adopts the following technical solutions:

[0008] A crowd-sensing task scheduling method based on multi-space modeling and fairness reinforcement learning includes:

[0009] In the multi-space region division phase, a multi-layer space task structure with ground and air coordination is constructed based on the acquired tasks;

[0010] In the task graph construction and critical path identification phase, a topological association model between tasks is established based on the spatial region division results, and the priority sequence of task scheduling is clarified through the critical path extraction mechanism;

[0011] In the task allocation phase, a differentiated agent matching strategy is constructed to rationally configure key path tasks based on priority sequences;

[0012] In the reinforcement learning phase, a fairness-aware multi-objective scheduling algorithm is used to dynamically allocate non-critical path tasks;

[0013] A dynamic fairness evaluation mechanism is used to evaluate and optimize task allocation results.

[0014] Furthermore, a multi-layer spatial task structure for ground and air collaboration is constructed based on the acquired tasks, including density clustering of the ground task area, identifying task hotspots using the DBSCAN algorithm, and dividing the ground perception tasks into several spatially autonomous task clusters; then, for the air task area, spatial stratification is performed according to the altitude information of the tasks, and the airspace is divided into multiple levels of low altitude, medium altitude and high altitude. A quadtree structure is introduced in each layer to recursively divide the plane area, thereby realizing hierarchical management and refined expression of air tasks.

[0015] Furthermore, the construction of a multi-layer spatial task structure for ground and air collaboration based on the acquired tasks also includes introducing a Voronoi diagram construction mechanism, generating a spatial control area based on the spatial centroid of the divided task group, and automatically mapping unattributed tasks to corresponding spatial task units based on the nearest centroid principle; for collaborative tasks involving both ground and air execution requirements, cross-layer coupling mapping is achieved by constructing a hybrid Voronoi diagram to ensure the consistency and integrity of the task space organization.

[0016] Furthermore, a topological association model between tasks is established based on the results of spatial area division, and the priority sequence of task scheduling is clarified through the critical path extraction mechanism, including constructing a task area graph based on the spatial task groups formed by the division, and the task area graph adopts a weighted undirected graph, in which nodes correspond to task groups, and the weights of edges are determined by the time urgency difference and spatial distance between task groups; an urgency score is calculated for each task, and the urgency score is defined based on the time window of the task; and a group urgency index is constructed for each task group, which comprehensively reflects the scheduling priority of the task group by weighted aggregation of the urgency of tasks within the group.

[0017] Furthermore, a topological association model between tasks is established based on the spatial area division results, and the priority sequence of task scheduling is clarified through the critical path extraction mechanism. It also includes dynamically generating the association weights of the edges in the graph by combining the earliest start time difference between task groups and the cross-space movement cost, which is used to characterize the dependency relationship and cost information between tasks; wherein, the nodes of the Voronoi diagram are used as the vertices of the graph to construct a task group network, in which each seed node corresponds to a task group. When determining the edge weights, all critical paths in the task group network are found by dynamically adjusting the weights and combining with the NetworkX framework.

[0018] Furthermore, the differentiated agent matching strategy is constructed to rationally configure critical path tasks based on priority sequences, including the use of a greedy matching algorithm and a fairness-aware multi-objective reinforcement learning algorithm FA-PMOAQL as complementary scheduling mechanisms. For tasks in the critical path, based on a pre-built task scheduling priority sequence, a greedy strategy is used to prioritize matching the most suitable agent node. During the matching process, the optimization goals are to minimize the spatial distance between tasks and agents and maximize the matching degree of equipment capabilities.

[0019] Furthermore, the fairness-aware multi-objective scheduling algorithm is used to dynamically allocate non-critical path tasks, including by constructing a state-action mapping, combining pheromone concentration and value function estimation, and continuously iteratively updating the task-agent allocation strategy to ensure the dynamic convergence of multi-objective optimization goals. Among them, task allocation is modeled as a Markov decision process, and an ant colony optimization mechanism is introduced to enhance local search and global exploration capabilities. For Q learning, task allocation is modeled as a state-action process, and the Bellman equation is used to update the Q table.

[0020] Furthermore, the use of a fairness-aware multi-objective scheduling algorithm to dynamically allocate non-critical path tasks also includes dynamically adjusting the pheromone concentration for ant colony optimization and using the Pareto pheromone update criterion: only when the path belongs to the Pareto optimal solution set, its pheromone concentration is increased and pheromone volatilization is performed at the same time; finally, the task allocation probability is calculated based on the pheromone concentration and Q value.

[0021] Furthermore, the dynamic fairness evaluation mechanism is used to evaluate and optimize the task allocation results, including calculating the task load of the agent after each round of allocation is completed, calculating the Gini coefficient and load variance based on the task load and using them as fairness indicators to evaluate scheduling fairness; and embedding the fairness indicator as part of the reward function into the reinforcement learning model to maintain task completion efficiency while achieving task allocation fairness.

[0022] The second aspect is a crowd-aware task scheduling system based on multi-space modeling and fairness reinforcement learning, including:

[0023] The multi-space region division module is configured to, during the multi-space region division phase, construct a multi-layer space task structure for ground and air coordination based on the acquired tasks;

[0024] The critical path identification module is configured to,in the task graph construction and critical path identification phases,build a topological association model between tasks based on the,spatial region division results, and clarify the priority sequence of task scheduling,through the critical path extraction mechanism;

[0025] The task allocation module is configured to,build differentiated agent matching strategies in the task allocation,and reasonably allocate critical path tasks based on the,priority sequence;

[0026] The reinforcement learning module is configured to,in the reinforcement learning phase, dynamically allocate non-critical path tasks,using the fairness-aware multi-objective scheduling algorithm;

[0027] The evaluation module is configured to evaluate and optimize the task allocation results using a dynamic fairness evaluation mechanism.

[0028] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device. The method for scheduling group intelligence tasks based on multi-space modeling and fairness reinforcement learning is described above.

[0029] In a fourth aspect, the present invention provides a terminal device comprising 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, and the instructions are suitable for being loaded and executed by the processor to perform the said fairness-guaranteed multi-space group intelligence perception task allocation dynamic collaborative optimization method.

[0030] In summary, the present invention has the following beneficial technical effects:

[0031] 1. The present invention constructs a spatial task area division mechanism for ground and air collaboration through the MCQVP algorithm, and adopts DBSCAN clustering and quadtree structure to finely model the multi-space task environment. It can effectively identify task hotspots and spatial density differences, realize spatial autonomy and regional separation of task grouping, and provide basic support for subsequent task graph modeling and collaborative scheduling.

[0032] 2. The GUMSP critical path identification algorithm proposed in this invention comprehensively considers task priority and time urgency, and can efficiently identify multiple critical scheduling paths in the constructed task area graph, ensuring that the system prioritizes the completion of important tasks under resource-constrained conditions, significantly improving the overall task completion rate and scheduling timeliness.

[0033] 3. This paper uses a fairness-aware multi-objective scheduling algorithm, FA-PMOAQL, which integrates reinforcement learning and ant colony optimization. This algorithm constructs a multi-objective model centered around device matching, distance optimization, and fairness threshold assurance, enabling dynamic allocation of tasks and agents. This method effectively improves agent resource utilization, optimizes task execution paths, and enhances the stability and scalability of the scheduling system.

[0034] 4. The present invention embeds fairness indicators such as the Gini coefficient and agent load variance in the task scheduling process, constructs a fairness reward function and embeds it into the learning process, so that task allocation can ensure efficiency while taking into account the fair participation of multiple agents, effectively preventing the occurrence of agent overload or idleness, and improving the overall sustainable operation capability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of the overall system framework of a fairness-guaranteed multi-space crowd sensing task allocation method based on multi-objective collaborative optimization in an embodiment of the present invention;

[0036] Figure 2 Schematic diagram of task area spatial division and multi-layer spatial modeling in an embodiment of the present invention;

[0037] Figure 3 Schematic diagram of the process of constructing a task area map and identifying a critical path in an embodiment of the present invention;

[0038] Figure 4 This is a schematic diagram of multi-space task distribution in an embodiment of the present invention;

[0039] Figure 5 This is a graph showing comparative experimental results of different algorithms under the task completion rate indicator as the number of agents changes in an embodiment of the present invention;

[0040] Figure 6 This is a graph showing comparative experimental results of different algorithms under the task completion rate indicator as the number of tasks changes in an embodiment of the present invention;

[0041] Figure 7 This is a graph showing comparative experimental results of different algorithms under the runtime index in an embodiment of the present invention;

[0042] Figure 8 This is a graph showing comparative experimental results of different algorithms under resource utilization indicators in an embodiment of the present invention;

[0043] Figure 9 This is a graph showing comparative experimental results of different algorithms under the fairness index indicator in an embodiment of the present invention;

[0044] Figure 10 This is a graph showing comparative experimental results of different algorithms under the cross-layer task completion rate indicator in an embodiment of the present invention;

[0045] Figure 11 This is a comparative experimental result diagram of different algorithms under the average task delay indicator in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described in detail below with reference to the accompanying drawings.

[0047] Example 1

[0048] Reference Figure 1 In this embodiment, a method for group intelligence sensing task scheduling based on multi-space modeling and fairness reinforcement learning includes:

[0049] In the multi-space region division phase, a multi-layer space task structure with ground and air coordination is constructed based on the acquired tasks;

[0050] During the task graph construction and critical path identification phase, a topological association model between tasks is established based on the spatial region division results, and the priority sequence of task scheduling is clarified through the critical path extraction mechanism;

[0051] In the task allocation phase, a differentiated agent matching strategy is constructed to rationally configure key path tasks based on priority sequences;

[0052] In the reinforcement learning phase, a fairness-aware multi-objective scheduling algorithm is used to dynamically allocate non-critical path tasks;

[0053] A dynamic fairness evaluation mechanism is used to evaluate and optimize task allocation results.

[0054] like Figure 1 As shown, the present invention designs a crowd-sensing task scheduling framework based on multi-space modeling and fairness reinforcement learning. The system introduces three core links in the scheduling framework: spatial modeling, path identification, and intelligent scheduling, which correspond to the spatial organization, priority sorting, and dynamic resource allocation of tasks respectively. In the spatial organization stage, the platform uses the DBSCAN clustering algorithm to identify ground hotspots, and combines the three-layer structure division of aerial tasks to construct a multi-space collaborative perception environment; in the path identification stage, the platform uses the GUMSP algorithm to construct a task area map based on the earliest start time and priority of the task, and identifies critical path tasks, giving priority to emergency scheduling.

[0055] In the intelligent scheduling phase, the platform assigns critical path tasks to the agent node with the strongest equipment capability and the best distance, while non-critical tasks are assigned using the fairness-aware multi-objective optimization model FA-PMOAQL that integrates Q-Learning and ant colony optimization algorithms. In this optimization process, the system takes the minimum task assignment distance, maximum equipment matching, and minimum agent load variance as objective functions, and comprehensively considers factors such as task location, equipment requirements, and agent resource status to construct a task assignment candidate set. Its task-agent equipment matching is calculated by The platform performs calculations and constructs the allocation probability based on the Q value between the task and the agent and the pheromone concentration, and dynamically generates the task allocation results.

[0056] To ensure overall system fairness during scheduling, the platform incorporates statistical metrics such as the Gini coefficient and load variance to construct a fairness reward function. This function is then embedded into a reinforcement learning model to achieve a dynamic balance between task efficiency and fair agent participation. Ultimately, the platform outputs a set of multi-objective optimal solutions as the task-agent allocation strategy. It also adjusts subsequent scheduling strategies based on historical task status and feedback, maintaining system stability, efficiency, and fairness in multi-space, highly dynamic perception scenarios.

[0057] The specific steps are as follows:

[0058] Embodiment 1 of the present invention provides a crowd-sensing task scheduling method based on multi-space modeling and fairness reinforcement learning, including a multi-space area division stage, a task graph construction and critical path identification stage, a task allocation stage, reinforcement learning and optimization algorithm implementation, and a fairness evaluation mechanism.

[0059] Multi-spatial region division phase: The multi-spatial region division phase uses the MCQVP algorithm to construct a multi-layered spatial task structure for ground and air collaboration, enabling orderly spatial organization and efficient management of crowd-sensing tasks. Specifically, the ground task region is first subjected to density clustering, and the DBSCAN algorithm is used to identify task hotspots, dividing the ground perception tasks into several spatially autonomous task clusters. Subsequently, for the air task region, spatial stratification is performed based on task altitude information, dividing the airspace into low-altitude, mid-altitude, and high-altitude levels. A quadtree structure is then introduced within each level to recursively partition the planar regions, enabling hierarchical management and refined representation of air tasks. To further improve the accuracy of the mapping between tasks and spatial units, the system incorporates a Voronoi diagram construction mechanism. Spatial control regions are generated based on the spatial centroids of the divided task groups, and unassigned tasks are automatically mapped to corresponding spatial task units based on the nearest centroid principle. For collaborative tasks involving both ground and air execution requirements, a hybrid Voronoi diagram is constructed to achieve cross-layer coupled mapping, ensuring the consistency and integrity of the spatial organization of tasks.

[0060] Step 1: For the ground area: Assume that the historical mission dataset is , where each historical task Contains three-dimensional space coordinates , but its is 0. The distance between task points is measured by Euclidean distance, which is defined as: , and then determine any point The neighborhood of If you click The neighborhood of (minimum number of samples), then The cluster is generated from the core point, and all points in its ε-neighborhood (including boundary points) are assigned to the same cluster, eventually forming a dense hotspot area. The task set of the hotspot area is defined as: ,in, It represents the total number of clusters identified by DBSCAN, and each cluster represents a ground hotspot area.

[0061] Step 2: For the aerial area: stratify based on the mission’s altitude information. Let the three-dimensional coordinates of each aerial mission be ,in Indicates the mission altitude. The airspace is divided into low altitude layers ( ), hollow layer ( ) and upper air layer ( ). In each layer, the quadtree structure is used to further divide the two-dimensional plane area. Let the bounding box of area A in a certain layer be , when the number of tasks in the area And the current partition depth , divide B evenly into four sub-areas For each sub-region, recursively apply the same partitioning rule until .

[0062] Step 3: Current task allocation and Voronoi diagram construction: Let the current batch task set be , where each task For each task in the divided area, , if a region is satisfied The boundary conditions are , it will be directly assigned to the area. For tasks that cannot be assigned to the predefined area, the Voronoi diagram method based on the centroid of the task group is used. , calculate the center of mass Then construct the Voronoi diagram , divide the entire space into several areas, each area Defined as: For each unassigned task ,make , and Assigned to in the corresponding task group.

[0063] Step 4: For collaborative tasks involving both ground and aerial operations (such as drone-vehicle relay delivery): let the collaborative task set be , a cross-layer task processing method is used to construct a Hybrid Voronoi Diagram (HVD). , if there are coordination constraints, they are enforced to meet: .

[0064] Task Graph Construction and Critical Path Identification: This phase uses the GUMSP algorithm to establish a topological association model between tasks based on the spatial partitioning results. A critical path extraction mechanism is then used to clarify the priority sequence for task scheduling, thereby ensuring the urgency responsiveness and global coordination efficiency of the overall system scheduling. Specifically, a task region graph is first constructed based on the spatial task groups formed by the aforementioned partitioning. This task region graph is a weighted undirected graph, in which each node corresponds to a task group, and the edge weights are determined by the temporal urgency difference and spatial distance between task groups. Furthermore, the system calculates an urgency score (US) for each task, defined based on the task's time window. A group urgency (GU) metric is then constructed for each task group. This metric comprehensively reflects the scheduling priority of the task group by weightedly aggregating the urgency of tasks within the group. Furthermore, the association weights for the edges in the graph are dynamically generated by combining the earliest start time difference between task groups and the cross-spatial movement cost to characterize the dependencies and costs between tasks.

[0065] Step 5: Calculate the urgency score (US) of each task. The calculation formula is: ,in, Indicates the end time of the task. Indicates the start time of the task.

[0066] Step 6: Calculate the group urgency (GU) by weighted aggregation of the urgency scores of all tasks in the task group. The calculation formula is: ,in, The weight is based on the number of tasks in the task group, ensuring that the situation where there are more tasks in the task group has a higher weight on the impact of urgency.

[0067] Step 7: Calculate the weights in the task group network. Edge weights The definition of is based on the time difference and urgency between task groups. Its calculation formula is: ,in, It represents the time difference between the earliest starting tasks of the two task groups. is the cost of cross-layer movement (such as the energy consumption of drone lifting). Source Task Group and target task force Average urgency level: .

[0068] Step 8: Find the Start Regions and End Regions. Start Regions: The set of task groups with the highest urgency, defined as: End Regions: The set of task groups with the lowest urgency, which is defined as: .

[0069] Step 9: Using the nodes of the Voronoi diagram as vertices, construct a task group network, where each seed node corresponds to a task group. When determining edge weights, dynamically adjust the weights using the above formula and, combined with the NetworkX framework, identify all critical paths in the task group network.

[0070] Task Allocation Phase: This phase constructs differentiated agent matching strategies for identified critical-path tasks and common tasks, comprehensively considering multiple factors such as geographic distance, equipment capability matching, and scheduling fairness to achieve efficient and rational allocation of task resources. This phase distinguishes between critical-path and non-critical-path tasks and employs two complementary scheduling mechanisms: a greedy matching algorithm and a fairness-aware multi-objective reinforcement learning algorithm (FA-PMOAQL). Specifically, for tasks on the critical path, the system employs a greedy strategy to prioritize matching with the most suitable agent nodes based on a pre-established task scheduling priority sequence. The matching process optimizes by minimizing the spatial distance between tasks and agents and maximizing equipment capability matching. For non-critical-path tasks, the proposed fairness-aware multi-objective reinforcement learning scheduling algorithm (FA-PMOAQL) performs dynamic matching. This algorithm, with the joint optimization objectives of minimizing the total distance cost of task allocation, maximizing equipment matching, and satisfying a preset fairness threshold, constructs a task-agent allocation probability model and iteratively generates the optimal task allocation strategy based on a reinforcement learning framework.

[0071] Step 10: The platform calculates the following indicators for the task group on the critical path and selects the optimal agent: Geographic distance calculation: ,in, The location information of the agent and the task group. Device matching calculation: , where K is the number of sensors possessed by the agent, p is the position of the sensor required by the task in the agent's sensor list (1 indicates the best match for the agent), α is a positive number used to penalize tasks that do not match the sensor list, HeightMatch is the degree of match between the aerial agent's height capability and the task's height, and β is the height matching weight coefficient.

[0072] Step 11: Perform priority task allocation, prioritize the tasks to the agents that best meet the equipment requirements, and ensure that the agents To the task The priority of task assignment is determined by the optimization function: ,in, Representing an agent To the task distance, For intelligent agents For task type Device matching score.

[0073] Step 12: For non-critical path tasks, use the fairness-aware multi-objective reinforcement learning scheduling algorithm FA-PMOAQL for dynamic scheduling. This algorithm jointly optimizes three objectives: minimizing the task distance cost ; Maximize device matching ; Ensure that the fairness threshold is met ,in represents the total travel distance of the agent, represents the device matching score between the agent and the task, represents the fairness index, is the fairness threshold.

[0074] Reinforcement learning and optimization algorithm implementation: The reinforcement learning and optimization algorithm implementation stage is mainly aimed at the dynamic allocation problem of non-critical path tasks, and a fairness-aware multi-objective scheduling algorithm, FA-PMOAQL, which integrates ant colony optimization and Q-learning mechanism is proposed. The algorithm continuously iterates and updates the task-agent allocation strategy by constructing state-action mapping, combining pheromone concentration and value function estimation, to ensure the dynamic convergence of multi-objective optimization goals. Specifically, the system models task allocation as a Markov decision process, where state s represents the current task-agent allocation state, action a represents the allocation of task t to agent a, and reward function The task relevance (matching degree), distance cost and fairness performance are measured respectively. At the same time, the system introduces the ant colony optimization mechanism to enhance local search and global exploration capabilities. In each round of allocation, the pheromone concentration is dynamically adjusted based on the historical interaction results between task t and agent a. In addition, the system continuously maintains the Pareto frontier solution set, performs non-dominated sorting on the task allocation results of each round, eliminates inferior solutions, and retains only the solutions that perform best in terms of the three objectives (distance, matching, and fairness) as the target objects for subsequent pheromone enhancement.

[0075] Step 13: For Q-learning, task assignment is modeled as a state-action process, where state s is the current task assignment state, which can be represented as the set of tasks S = {S1, S2, ..., Sn} to which the agent has been assigned, and action a is to assign task t to agent a, i.e. , the reward R is used to measure the quality of the task allocation plan, combining relevance, distance and fairness, and is defined as follows: Where, R1: normalized value of total correlation , R2: Normalized value of the total distance and negative , R3: Fairness Reward At the same time, the Q table is updated using the Bellman equation: , where s is the current state, a is the selected action, a′ is the next possible action, R i is the i-th component of the reward vector, is the learning rate and γ is the discount factor.

[0076] Step 14: For Ant Colony Optimization, dynamically adjust the pheromone concentration , and its update formula is: ,in, is the pheromone volatilization rate, which controls the rate at which pheromones decay over time. To increase the probability of selecting a high-quality solution, the pheromone increment is added. To ensure that pheromones only enhance Pareto optimal solutions, the Pareto pheromone update criterion is used: the pheromone concentration of a path (t, a) is increased only when it belongs to the Pareto optimal solution set. The calculation method is: , where Q is the pheromone enhancement factor and f(D,S) is the objective function that measures the quality of the current solution. .

[0077] Step 15: Calculate the probability of task assignment , which is determined by the pheromone concentration and Q value, and the formula is as follows: ,in, is the pheromone concentration, reflecting the historical performance of agent a assigned to task t, is the estimated value of assigning task t to agent a, reflecting the contribution of agent relevance and fairness. α and β control the importance of pheromone and Q value. After each ant completes the assignment, the Pareto frontier is updated according to its reward vector R as follows: , where, when maintaining the Pareto frontier, the dominance relationship between the two reward vectors R1 and R2 is determined , , if R1 dominates R2, then R2 can be removed from the Pareto frontier, and if R1 is dominated by R2, then R1 cannot be added to the Pareto frontier.

[0078] Fairness Assessment Mechanism: To ensure balanced utilization of agent resources and the long-term sustainability of task allocation, the system designs and introduces a dynamic fairness assessment mechanism. This mechanism combines two statistical indicators, the Gini coefficient and load variance, to comprehensively measure the fairness of task scheduling from both individual and overall perspectives. This mechanism is embedded in the reinforcement learning reward model to guide the agent selection strategy to continuously converge toward fair allocation. This fairness assessment mechanism not only participates in real-time reinforcement learning feedback during the model training phase, but also serves as a posterior evaluation metric for multiple rounds of task allocation evaluation and model improvement. This allows the scheduling system to achieve coordinated optimization of efficiency while maintaining long-term stability and fairness in agent participation.

[0079] Step 16: After each round of allocation is completed, the platform calculates the agent's task load and evaluates the scheduling fairness based on the following indicators:

[0080]

[0081]

[0082]

[0083]

[0084] in, is the average load of the agent, , n is the number of agents, For intelligent agents The task load, In non-decreasing order, is the load variance of the ground agent, is the load variance of the aerial agent, is the Gini coefficient of the ground agent, is the Gini coefficient of the aerial agent, and λ and μ are dynamically adjusted according to the task distribution (e.g., they are reduced when the aerial tasks are dense).

[0085] Step 17: Embed the fairness indicator as part of the reward function into the reinforcement learning model to promote the system to achieve fair task allocation while maintaining task completion efficiency. The calculation formula of the fairness reward function is: When Var+G→0 (i.e., both the load variance and the Gini coefficient are close to 0), R3→1. At this point, the system load is balanced, resource allocation is fair, and the reward function reaches its maximum value. When Var+G→∞, R3→0. At this point, the system load varies greatly, resource allocation is extremely unfair, and the reward function approaches its minimum value.

[0086] like Figure 2As shown, the present invention first divides the sensing area into ground mission areas and aerial mission areas. The ground mission areas are clustered using the DBSCAN algorithm to identify mission hotspots. Aerial missions are divided into three levels based on altitude: low altitude, medium altitude, and high altitude. A quadtree structure is used for fine-grained spatial partitioning. Each spatial area is ultimately constructed into a multi-layer spatial region structure diagram for subsequent scheduling modeling.

[0087] like Figure 3 As shown in the figure, the platform constructs a task area graph for all spatial task groups. The nodes in the graph are spatial task groups, and the edges represent the dependencies between tasks.

[0088] like Figure 4 As shown in Figure 2, the platform maps tasks to multiple spatial distribution scenarios, including ground hotspots, airspace monitoring zones, and collaborative coverage areas. For tasks on the critical path, the platform uses a greedy strategy based on the geographical distance between the task and the agent and the similarity of the equipment.

[0089] Example 2

[0090] The following example provides a detailed explanation of the fairness-guaranteed multi-space crowd sensing task scheduling problem. Consider a smart city platform receiving sensing tasks from 10 different regions in an urban monitoring application, including six ground tasks and four aerial tasks. The platform first spatially clusters the ground tasks based on their location, identifying three ground hotspots. It then categorizes the aerial tasks into low-altitude and mid-altitude tiers based on their altitude. The platform then broadcasts the tasks to candidate agents within a radius of 1 kilometer (on the ground) and 500 meters (in the air) around each task point. These candidate agents include unmanned vehicles, ground-based cameras, and aerial drone nodes.

[0091] After receiving task information, each agent calculates its device compatibility and spatial distance to the task based on its own device capabilities, historical task completion records, and task location, and generates a willingness to respond to the task. The platform aggregates the candidate agent information and generates a list of 10 candidate agents based on task-agent compatibility. This list is then fed into the fairness-aware multi-objective optimization scheduling model, FA-PMOAQL, as input.

[0092] During the optimization process, the model iterates using an ant colony search mechanism within a Q-learning-driven strategy framework, aiming to minimize the total task allocation distance, maximize device similarity, and ensure that a fairness threshold is met. After multiple rounds of strategy convergence, the system outputs an optimal allocation of 10 tasks and 10 agents, with each task matched to the optimal agent. The system outperforms traditional static allocation strategies in key metrics such as task completion rate, fairness, and resource utilization.

[0093] Experimental verification:

[0094] In order to verify the effectiveness of the method in this embodiment, the present invention conducts experiments on multiple real trajectory data sets, and the results are as follows: Figures 5 to 10 The above real data sets are derived from existing public data and have been approved by the information owners.

[0095] Among them, MSFTA represents the framework algorithm proposed in this invention, and the other algorithms are comparison algorithms, namely improved greedy algorithm (IGA), ant colony algorithm (ACO), FGT algorithm, GT+RL algorithm, etc.

[0096] (1) Task completion rate,

[0097] like Figure 5 As shown in Figure 2, as the number of agents increases, the task completion rate of all algorithms increases significantly. The MSFTA algorithm consistently outperforms the other algorithms, with its task completion rate increasing by 12.8%. This is mainly due to the fact that the increase in the number of agents expands the choice space for task allocation, thereby effectively reducing the possibility of task failure due to insufficient resources. Figure 6 As shown in the figure, the task completion rate of all algorithms decreases as the number of tasks increases, indicating that the increased number of tasks leads to greater processing difficulty. However, the MSFTA algorithm maintains the best performance under high load conditions. This excellence is attributed to its multi-objective optimization and critical path priority allocation strategy, which effectively matches tasks with agents and minimizes the decrease in completion rate caused by task backlogs.

[0098] (2) Running time,

[0099] like Figure 7 As shown in the figure, the runtime of each algorithm increases linearly with the number of agents, but there are significant differences in the growth rates. It is worth noting that the MSFTA algorithm maintains a consistently low runtime, further verifying the model's efficiency and applicability.

[0100] (3) Resource utilization,

[0101] like Figure 8 As shown in the figure, the resource utilization of each algorithm decreases as the number of agents increases, but the MSFTA algorithm performs the best. This is because the algorithm fully considers device relevance during task allocation and prioritizes tasks to the most suitable agents, effectively reducing resource idleness.

[0102] (4) Fairness index,

[0103] like Figure 9As shown in the figure, the fairness index of each algorithm decreases slightly as the number of agents increases, but the MSFTA algorithm still maintains a high level of fairness. This is because high-urgency tasks may be concentrated in a small number of efficient agents, resulting in uneven load. The MSFTA algorithm effectively alleviates this problem by introducing load variance and the Gini coefficient into the reward function, avoiding excessive sacrifice of fairness.

[0104] (5) Cross-layer task completion rate,

[0105] like Figure 10 As shown in the figure, the cross-layer task completion rates of all algorithms increase with the number of agents. This is because a small number of agents may lead to insufficient task allocation coverage and limited cross-layer collaboration efficiency. Increasing the number of agents can improve task scheduling flexibility and resource utilization through dynamic load balancing, thereby improving the completion rate.

[0106] (6) Average task delay,

[0107] like Figure 11 As shown in the figure, the average task latency of almost all algorithms decreases as the time window increases. This is because the expansion of the time window provides the algorithm with a longer global optimization scheduling cycle, allowing the task allocation strategy to more fully coordinate resources, reduce task waiting conflicts, and improve overall scheduling efficiency.

[0108] Extensive experiments were conducted on two real-world datasets to validate the performance of the proposed fair-guaranteed multi-space crowdsensing task allocation method based on multi-objective collaborative optimization. The results show that the fair-guaranteed multi-space crowdsensing task allocation method based on multi-objective collaborative optimization outperforms other baseline methods.

[0109] Example 3

[0110] This embodiment provides a group intelligence perception task scheduling system based on multi-space modeling and fairness reinforcement learning. The system structure is as follows: Figure 1 As shown, it includes the following functional modules:

[0111] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device. The method for dynamic collaborative optimization of fairness-guaranteed multi-space crowd intelligence perception task allocation is disclosed.

[0112] A terminal device includes 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, wherein the instructions are suitable for being loaded and executed by the processor. The method for dynamic collaborative optimization of fairness-guaranteed multi-space crowd intelligence perception task allocation is described.

[0113] 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, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A crowd-sensing task scheduling method based on multi-space modeling and fairness reinforcement learning, characterized by: include: In the multi-space region division phase, a multi-layer space task structure with ground and air coordination is constructed based on the acquired tasks; In the task graph construction and critical path identification phase, a topological association model between tasks is established based on the spatial region division results, and the priority sequence of task scheduling is clarified through the critical path extraction mechanism; In the task allocation phase, a differentiated agent matching strategy is constructed to rationally configure key path tasks based on priority sequences; In the reinforcement learning phase, a fairness-aware multi-objective scheduling algorithm is used to dynamically allocate non-critical path tasks; Use dynamic fairness evaluation mechanism to evaluate and optimize task allocation results; The method constructs a multi-layer spatial task structure for ground and air collaboration based on the acquired tasks, including density clustering of the ground task area, identifying task hotspots using the DBSCAN algorithm, and dividing the ground perception tasks into several spatially autonomous task clusters. Subsequently, for the air task area, spatial stratification is performed based on the altitude information of the tasks, dividing the airspace into multiple levels of low altitude, medium altitude, and high altitude. A quadtree structure is introduced in each layer to recursively divide the plane area, thereby realizing hierarchical management and refined expression of air tasks. The method of constructing a multi-layer spatial task structure for ground and air collaboration based on the acquired tasks also includes introducing a Voronoi diagram construction mechanism, generating a spatial control area based on the spatial centroid of the divided task group, and automatically mapping unassigned tasks to corresponding spatial task units based on the nearest centroid principle; for collaborative tasks involving both ground and air execution requirements, cross-layer coupling mapping is achieved by constructing a hybrid Voronoi diagram to ensure the consistency and integrity of the task space organization; The method establishes a topological association model between tasks based on the spatial region division results, and clarifies the priority sequence of task scheduling through a critical path extraction mechanism, including constructing a task region graph based on the spatial task groups formed by the division. The task region graph adopts a weighted undirected graph, in which nodes correspond to task groups and the weights of edges are determined by the time urgency difference and spatial distance between task groups; calculates an urgency score for each task, and the urgency score is defined based on the time window of the task; and constructs a group urgency index for each task group, which comprehensively reflects the scheduling priority of the task group by weighted aggregation of the urgency of tasks within the group. The method establishes a topological association model between tasks based on the spatial area division results, and clarifies the priority sequence of task scheduling through a critical path extraction mechanism. It also includes dynamically generating the association weights of the edges in the graph by combining the earliest start time difference between task groups and the cross-space movement cost, which is used to characterize the dependency and cost information between tasks. Among them, the nodes of the Voronoi diagram are used as the vertices of the graph to construct a task group network, in which each seed node corresponds to a task group. When determining the edge weights, the weights are dynamically adjusted and combined with the NetworkX framework to find all critical paths in the task group network.

2. The method for group intelligence sensing task scheduling based on multi-space modeling and fairness reinforcement learning according to claim 1 is characterized in that: The differentiated agent matching strategy is constructed to rationally configure critical path tasks based on priority sequences, including the use of a greedy matching algorithm and a fairness-aware multi-objective reinforcement learning algorithm FA-PMOAQL as complementary scheduling mechanisms. For tasks in the critical path, a greedy strategy is used to preferentially match them with the most suitable agent nodes based on a pre-built task scheduling priority sequence. During the matching process, the optimization goals are to minimize the spatial distance between tasks and agents and maximize the matching degree of equipment capabilities.

3. The method for group intelligence sensing task scheduling based on multi-space modeling and fairness reinforcement learning according to claim 2 is characterized in that: The fairness-aware multi-objective scheduling algorithm is used to dynamically allocate non-critical path tasks, including constructing a state-action mapping, combining pheromone concentration and value function estimation, and continuously iteratively updating the task-agent allocation strategy to ensure the dynamic convergence of multi-objective optimization goals. Task allocation is modeled as a Markov decision process, and an ant colony optimization mechanism is introduced to enhance local search and global exploration capabilities. For Q-learning, task allocation is modeled as a state-action process, and the Bellman equation is used to update the Q table.

4. The method for group intelligence sensing task scheduling based on multi-space modeling and fairness reinforcement learning according to claim 3 is characterized in that: The method utilizes a fairness-aware multi-objective scheduling algorithm to dynamically allocate non-critical path tasks, and also includes dynamically adjusting the pheromone concentration for ant colony optimization and using the Pareto pheromone update criterion: only when the path belongs to the Pareto optimal solution set, its pheromone concentration is increased and pheromone volatilization is performed at the same time; finally, the task allocation probability is calculated based on the pheromone concentration and Q value.

5. The method for group intelligence sensing task scheduling based on multi-space modeling and fairness reinforcement learning according to claim 4 is characterized in that: The dynamic fairness evaluation mechanism is used to evaluate and optimize the task allocation results. This includes calculating the task load of the agent after each round of allocation, calculating the Gini coefficient and load variance based on the task load, and using them as fairness indicators to evaluate scheduling fairness. The fairness indicator is embedded in the reinforcement learning model as part of the reward function to maintain task completion efficiency while achieving task allocation fairness.

6. A crowd-sensing task scheduling system based on multi-space modeling and fairness reinforcement learning, which executes the crowd-sensing task scheduling method based on multi-space modeling and fairness reinforcement learning as described in claim 1, characterized in that: include: The multi-space region division module is configured to, during the multi-space region division phase, construct a multi-layer space task structure for ground and air coordination based on the acquired tasks; The critical path identification module is configured to,in the task graph construction and critical path identification phases,build a topological association model between tasks based on the,spatial region division results, and clarify the priority sequence of task scheduling,through the critical path extraction mechanism; The task allocation module is configured to,build differentiated agent matching strategies in the task allocation,and reasonably allocate critical path tasks based on the,priority sequence; The reinforcement learning module is configured to,in the reinforcement learning phase, dynamically allocate non-critical path tasks,using the fairness-aware multi-objective scheduling algorithm; The evaluation module is configured to evaluate and optimize the task allocation results using a dynamic fairness evaluation mechanism.

Images