Community management method and system based on unmanned aerial vehicle, electronic equipment and storage medium

Abstract

The invention provides a community management method and system based on an unmanned aerial vehicle, electronic equipment and a storage medium, and relates to the field of community management. The method comprises the steps of obtaining an initial inspection task of an unmanned aerial vehicle cluster and planning an inspection route, and controlling the unmanned aerial vehicle cluster to execute inspection operation to obtain multi-source inspection data. And constructing a community portrait based on the inspection data, identifying community management hidden dangers through a classification reasoning model, and generating an early warning event and a processing task. A processing task is decomposed into hierarchical sub-tasks by utilizing an ontology reasoning framework, a task execution main body is selected, and a task set is generated. And a multi-Agent reinforcement learning framework is adopted to carry out space-time collaborative arrangement on the task set to form an overall collaborative scheme, the overall collaborative scheme is converted into an interpretable scheduling decision instruction, and finally the unmanned aerial vehicle is controlled to execute the task. According to the technical scheme, the emergency response speed of security and protection and the capability of coping with complex conditions are improved.

Description

Technical Field

[0001] The present application relates to the field of community management, and in particular to a community management method, system, electronic device and storage medium based on drones. Background Art

[0002] With the acceleration of urbanization, the demand for community security management is growing. Traditional security systems rely on manual patrols, which are inefficient and unable to respond quickly to emergencies, especially at night or in bad weather conditions.

[0003] Currently, most smart security systems on the market use fixed cameras and sensor networks, which lack mobility and flexibility and are difficult to cover complex terrain or dynamically changing areas. The limitations of fixed security facilities lead to the existence of monitoring blind spots, slow emergency response speed, low intelligence level, and inability to actively identify and respond to complex situations. Summary of the invention

[0004] The present application provides a community management method, system, electronic device and storage medium based on drones, which are used to improve the emergency response speed of security and the ability to cope with complex situations.

[0005] In a first aspect, the present application provides a community management method based on drones, comprising: Obtain the initial inspection task of the drone cluster, and plan the initial inspection route of the drone cluster based on the initial inspection task; Control the drone cluster to perform inspection operations based on the initial inspection route, and obtain multi-source inspection data of the inspection area through the drone cluster during the inspection operation; Build community portraits based on multi-source inspection data, and input the community portraits into the preset classification reasoning model to output community management risks; Generate structured warning events based on the categories and levels of community management risks, and determine corresponding disposal tasks based on the warning events; Decompose the disposal task based on the preset ontology reasoning framework to obtain multiple hierarchical subtask descriptions, and select the corresponding execution task subject based on the subtask descriptions; Generate a task set corresponding to the execution task subject according to the execution task subject and the corresponding subtask description; For the task execution subject, a preset multi-agent reinforcement learning framework is used to perform spatiotemporal collaborative arrangement of the task set to form an overall collaborative plan, and the overall collaborative plan is expressed as an interpretable scheduling decision instruction; Control the drone cluster to perform tasks according to the scheduling instructions.

[0006] In the above technical scheme, the present application firstly performs initial inspection route planning for the drone cluster based on the initial inspection task, so that the drone cluster can perform inspection operations according to the optimized route, thereby improving the inspection efficiency; during the inspection process, the drone cluster obtains multi-source inspection data and constructs a community portrait, and combines the classification reasoning model to realize the intelligent identification of community management risks, thereby enhancing the system's perception ability; by converting community management risks into structured early warning events and disposal tasks, and using the ontology reasoning framework to decompose the tasks, a hierarchical sub-task description is obtained, thereby realizing the intelligent transformation from risk discovery to task execution; then, by selecting the appropriate task execution subject according to the sub-task description and generating the corresponding task set, the rationality of task allocation is ensured; finally, a multi-agent reinforcement learning framework is used to perform spatiotemporal collaborative arrangement of the task set, forming an overall collaborative plan and converting it into an interpretable scheduling decision instruction, thereby realizing the intelligent collaborative scheduling of the drone cluster. This technical route based on multi-source data perception, ontology reasoning and reinforcement learning not only improves the automation and intelligence level of community management, but also enhances the reliability and credibility of the system through an explainable scheduling decision-making mechanism. At the same time, the introduction of a multi-agent collaborative framework also significantly improves the system's adaptability to complex scenarios and task execution efficiency.

[0007] In a second aspect of the present application, a community management system based on a drone is provided, the system comprising: The route planning module is used to obtain the initial inspection task of the drone cluster and plan the initial inspection route of the drone cluster based on the initial inspection task; The data acquisition module is used to control the drone cluster to perform inspection operations based on the initial inspection route, and to obtain multi-source inspection data of the inspection area through the drone cluster during the inspection operation; The hidden danger identification module is used to build a community portrait based on multi-source inspection data, and input the community portrait into a preset classification reasoning model to output community management hidden dangers; The disposal task generation module is used to generate structured warning events and corresponding disposal tasks based on the categories and levels of community management risks; The task decomposition module is used to decompose the disposal task based on the preset ontology reasoning framework, obtain multiple hierarchical subtask descriptions, and select the corresponding execution task subject based on the subtask descriptions; A task set generation module is used to generate a task set corresponding to the execution task subject according to the execution task subject and the corresponding subtask description; The scheduling decision instruction module is used to perform spatiotemporal collaborative arrangement of task sets for task execution entities using a preset multi-agent reinforcement learning framework to form an overall collaborative plan, and express the overall collaborative plan as an interpretable scheduling decision instruction; The task execution module is used to control the drone cluster to perform tasks according to the scheduling instructions.

[0008] In a third aspect of the present application, a computer storage medium is provided, wherein the computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the above method steps.

[0009] In the fourth aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the above method.

[0010] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. This application first plans the initial inspection route for the drone cluster based on the initial inspection task, so that the drone cluster can perform inspection operations according to the optimized route, thereby improving the inspection efficiency; during the inspection process, the drone cluster obtains multi-source inspection data and builds a community portrait, and combines the classification reasoning model to realize the intelligent identification of community management risks, thereby enhancing the system's perception ability; by converting community management risks into structured warning events and disposal tasks, and using the ontology reasoning framework to decompose tasks, a hierarchical subtask description is obtained, thereby realizing the intelligent transformation from risk discovery to task execution; 2. This application ensures the rationality of task allocation by selecting the appropriate task execution subject according to the subtask description and generating the corresponding task set; finally, a multi-agent reinforcement learning framework is used to perform spatiotemporal collaborative arrangement of the task set, forming an overall collaborative plan and converting it into an interpretable scheduling decision instruction, thereby realizing the intelligent collaborative scheduling of drone clusters.

[0011] 3. This application is based on the technical route of multi-source data perception, ontology reasoning and reinforcement learning, which not only improves the automation and intelligence level of community management, but also enhances the reliability and credibility of the system through an explainable scheduling decision-making mechanism. At the same time, the introduction of a multi-agent collaborative framework also significantly improves the system's adaptability to complex scenarios and task execution efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of a community management method based on drones provided in an embodiment of the present application; Figure 2 An architecture diagram of a community management system based on drones provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0013] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0014] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0015] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0016] In order to facilitate understanding of the method and system provided by the embodiments of the present application, before introducing the embodiments of the present application, the background of the embodiments of the present application is first introduced.

[0017] With the continuous advancement of urbanization, community safety management faces increasingly severe challenges. The existing community security system mainly relies on human patrols. This traditional management model has many inherent defects: the limited number of patrol personnel leads to insufficient patrol coverage, and the physical limitations of personnel lead to low patrol efficiency. Especially at night or in bad weather conditions, the work efficiency of patrol personnel is significantly reduced, and they are unable to detect and deal with emergencies in a timely manner, which seriously affects the effectiveness of community safety management.

[0018] Although the widely used intelligent security systems in the market have introduced technical means such as fixed cameras and sensor networks, such fixed security facilities still have many limitations. Due to the fixed installation location of the equipment, it is difficult to achieve full coverage of complex terrain, and monitoring blind spots are formed in locations such as building blind spots and vegetation obstruction areas. At the same time, this static deployment method lacks flexibility and it is difficult to adjust the monitoring strategy in time in the face of dynamically changing scenes in the community. More importantly, the intelligence level of existing intelligent security systems is limited, and it is impossible to actively identify and warn of abnormal situations. The system's emergency response capabilities are insufficient and it is difficult to meet the actual needs of modern community security management. The existence of these problems has seriously restricted the improvement of community security management level, and it is urgent to develop new intelligent security solutions.

[0019] After the background introduction of the above content, those skilled in the art can understand the problems existing in the prior art. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0020] Based on the above background technology, please refer to Figure 1 , Figure 1 A flowchart of a community management method based on drones provided in an embodiment of the present application. The system can be implemented by a computer program or can be run as an independent tool application. Specifically, in an embodiment of the present application, the method can be applied on a server, but can also be applied to electronic devices such as servers. A community management method based on drones includes the following steps: S101, obtaining an initial inspection task of the drone cluster, and planning an initial inspection route of the drone cluster based on the initial inspection task; Specifically, first, the initial inspection task of the drone cluster needs to be obtained through the task receiving module. The initial inspection task contains basic task information such as the boundary coordinates of the inspection area, key monitoring locations, and inspection time periods. The system matches the obtained initial inspection task with the pre-built standard task template library, which stores inspection specifications and execution requirements in different scenarios, and obtains a standardized task description through matching. Subsequently, the system reads the community three-dimensional map data, which contains spatial information such as building height, road distribution, and green areas, and obtains the community management schedule, which records the time and area usage arrangements for various activities. The system fuses the standardized task description with the three-dimensional map data and the management schedule to generate an initial task data packet with specific execution parameters.

[0021] After receiving the initial mission data packet, the system further obtains terrain data, building distribution data and control area information of the inspection area, which together constitute the environmental constraints of the UAV flight. Based on these constraints, the system constructs a multi-objective path optimization model that takes into account the flight safety boundary. The model takes the shortest flight distance, optimal energy consumption and maximum coverage as optimization goals. By running the improved ant colony optimization algorithm to solve the model, a set of optimal flight paths that meet the preset constraints is obtained. These paths can achieve a balance of multiple objectives while ensuring the completion of the mission.

[0022] The system inputs the optimal flight path set into the preset safety assessment rule base for path safety assessment. The safety assessment rule base contains multi-dimensional safety assessment standards such as meteorological conditions, building spacing, and flight altitude. Based on the safety assessment results, combined with task priority and timeliness requirements, the system uses the hierarchical analysis method to calculate the comprehensive score of each flight path, and selects the path with the highest score as the final initial inspection route. This multi-level path planning method ensures the reliability and efficiency of drone cluster inspection operations by comprehensively considering safety, efficiency, and task characteristics.

[0023] Based on the above embodiment, as an optional embodiment, planning the initial inspection route of the drone cluster based on the initial inspection task includes: S201, associating and matching the initial inspection task with the preset task template to obtain a standardized inspection task description; Specifically, a task template library is built, which stores the standard description format of different types of inspection tasks. The task template adopts a semantic network structure and defines the core elements such as inspection objects, inspection requirements, execution constraints and their associations. Each task template contains standardized fields such as task type, inspection items, quality standards and resource requirements.

[0024] The system performs semantic analysis on the initial inspection task and extracts the key feature information of the task. Semantic analysis uses natural language processing technology to extract entities and relationships in the task description through word segmentation, part-of-speech tagging and dependency analysis. The system uses the domain ontology model to semantically annotate the extracted features and establish a feature vector representation.

[0025] The system starts the template matching engine, which adopts a multi-level matching strategy. First, coarse-grained matching is performed by task type to screen out a set of candidate templates. Then, fine-grained matching is performed based on semantic similarity calculation. The system uses a word vector model to calculate the semantic distance between the task description and the template, and selects the template with the highest similarity as the matching result.

[0026] To improve matching accuracy, the system implements feature weight learning based on the attention mechanism. By analyzing historical matching data, the system learns the importance weights of different features in the matching process. The system uses an iterative optimization algorithm to dynamically adjust weight parameters to improve the accuracy of template matching.

[0027] The system fills the specific requirements of the initial inspection task into the matching template through parameter mapping. Parameter mapping uses rule reasoning method to establish a corresponding relationship between task characteristics and template parameters. The system ensures the rationality of parameter filling through constraint checking and automatically adjusts parameters that do not meet the constraints.

[0028] S202, obtaining community geographic information data and a community management schedule, and generating an executable initial task set based on a standardized inspection task description combined with the community geographic information data and the community management schedule; Specifically, we first establish a data management framework to integrate community geographic information data and community management schedules. The geographic information data adopts a multi-layer GIS structure, including spatial information layers such as building distribution, road network, and green space distribution. Each layer of data is organized through a spatial index mechanism to support fast spatial query and analysis.

[0029] The system digitizes the community management schedule and converts the schedule information into a time series data structure. The schedule includes various community activities, facility maintenance, security inspections and other management matters. The system organizes activities in time sequence through timestamps and establishes time indexes to support efficient time queries.

[0030] The system builds a spatiotemporal constraint analyzer that associates and analyzes standardized inspection task descriptions with geographic information data and schedules. The analyzer first locates the inspection target location in the geographic information system and determines the boundaries and obstacle distribution of the inspection area through spatial analysis. At the same time, the system extracts activity information for relevant time periods from the schedule to identify potential spatiotemporal conflicts.

[0031] The system uses a task decomposition algorithm to divide inspection tasks according to spatial areas and time windows. The spatial division uses a grid method to divide the inspection area into sub-areas suitable for drone operations. The time division is based on the idle time in the schedule, and the system determines the optimal inspection time for each sub-area through time window analysis.

[0032] To ensure the feasibility of the mission, the system has established a constraint verification mechanism. This mechanism checks whether the mission execution meets the drone performance limitations, regulatory requirements and safety specifications. The system calculates the drone's endurance through an energy consumption model and evaluates the execution conditions in combination with weather forecast data to ensure that the generated mission is technically feasible.

[0033] The system implements a task priority assessment module, which determines the execution order of subtasks through multi-dimensional scoring. The scoring dimensions include factors such as task urgency, regional importance, and resource efficiency. The system uses hierarchical analysis to calculate the weight of each dimension and obtains a comprehensive priority score through weighted summation.

[0034] S203, obtaining community topography data, building layout data, and sensitive area distribution data corresponding to the initial task set; Specifically, a multi-source data acquisition framework was established, which integrated multiple data sources such as remote sensing images, 3D laser scanning and field measurements. The system unified data from different sources into the same coordinate system through data fusion algorithms to build a high-precision digital model of the community.

[0035] The system processes the community topography data through the terrain analysis module. This module uses digital elevation model technology to generate the community's elevation distribution map through rasterization. The system calculates the terrain slope and slope direction through the slope analysis algorithm, and identifies the terrain feature points and change areas. At the same time, the system establishes a terrain classification system to divide the community terrain into flat land, slopes, steps and other types of areas.

[0036] To obtain building layout data, the system uses building information modeling technology to construct a three-dimensional model of community buildings. The system obtains the plane outline of the building through the contour extraction algorithm and generates a three-dimensional structural model of the building in combination with the height information. The system also records the attribute information of the building, including parameters such as building type, floor height, and usage function.

[0037] The system has established a sensitive area identification module, which classifies and marks important facilities, densely populated areas and special functional areas in the community based on multi-level sensitivity assessment standards. The system determines the boundaries of sensitive areas through spatial analysis and sets corresponding protection levels and access restrictions.

[0038] To improve the timeliness of data, the system implements a real-time update mechanism. Through the deployed sensor network, the environmental change data is collected in real time. The system adopts an incremental update strategy to timely update the dynamic information of topography, building status and sensitive areas.

[0039] The system builds a spatial data index structure and uses a quadtree algorithm to organize and manage geographic spatial data. This index structure supports fast spatial queries and regional statistics, improving data access efficiency. The system also establishes a data consistency check mechanism to ensure that the spatial relationships between different types of data are accurate.

[0040] The system implements a data quality control module, which checks the accuracy and completeness of the collected data by setting multi-level data validation rules. The system uses anomaly detection algorithms to identify noise and errors in the data, and uses data repair algorithms to handle missing and outliers.

[0041] To support subsequent path planning, the system has established a scene understanding module. This module uses a semantic segmentation algorithm to perform scene analysis on spatial data and identify passable areas, obstacles, and dangerous areas. Based on the results of scene analysis, the system generates a flight constraint map to mark the no-fly zones and altitude restrictions for drones.

[0042] S204, constructing a multi-constraint and multi-objective path optimization model based on the flight route according to the community topography data, building layout data and sensitive area distribution data; Specifically, a UAV flight path model is constructed, which describes the performance boundary of the UAV under different flight states. The flight path includes core parameters such as speed, altitude, and turning radius, and the feasible flight state space is determined through dynamic equations and performance constraints.

[0043] The system establishes an environmental constraint processing module, which converts community terrain data into a height constraint function. The system calculates the minimum safe flight altitude for each location point through a terrain analysis algorithm, and uses a three-dimensional grid division method to establish a spatial constraint map, marking the vertical and horizontal restrictions of different areas.

[0044] For building obstacle avoidance constraints, the system uses an expansion algorithm to process building layout data. By setting a safety buffer zone, the system expands the building outline outward to form a no-fly zone. The system uses a distance field algorithm to calculate the shortest distance from any point to an obstacle and construct a continuous obstacle avoidance constraint function.

[0045] The system has formulated a multi-level protection strategy for sensitive areas, setting different avoidance distances and flight altitude requirements according to the sensitivity level. The system describes the impact of sensitive areas through weight functions, establishes a distance-based attenuation model, and quantifies sensitivity constraints into mathematical expressions.

[0046] In terms of dynamic constraints, the system establishes a kinematic model for the UAV. This model takes into account thrust limitations, energy consumption, and aerodynamic characteristics, and describes the motion laws of the UAV through state equations. The system uses predictive control methods to evaluate path feasibility and ensure that the planned trajectory meets dynamic constraints.

[0047] S205, using heuristic search and intelligent optimization algorithms to solve the multi-constraint multi-objective path optimization model to obtain a set of Pareto optimal flight paths that meet preset constraint conditions; Specifically, an improved non-dominated sorting genetic algorithm is used to solve the multi-constraint multi-objective path optimization model. The algorithm represents the flight path as a series of three-dimensional waypoint sequences through path encoding, and each waypoint contains position coordinates and flight altitude information.

[0048] During the optimization process, the selection strategy based on crowding maintains the diversity of the population. The sparseness of each solution is evaluated by calculating the crowding distance of the solution in the target space. For solutions with smaller crowding, the probability of their selection is increased to maintain the uniform distribution of the solution set in the target space.

[0049] In genetic operations, adaptive crossover and mutation operators are designed. The crossover operation uses arithmetic crossover to control the globality and locality of the search by dynamically adjusting the crossover probability. The mutation operation uses Gaussian mutation, and the step length is adaptively adjusted according to the evolutionary algebra to achieve a dynamic balance of search accuracy.

[0050] Constraint violation calculation converts flight path constraints, obstacle constraints and sensitive area constraints into violation functions. The feasibility of the solution is evaluated by calculating the deviation between the path and the constraint conditions. The improved constraint dominance rule is used to give priority to solutions with smaller violation degrees to enter the next generation.

[0051] In order to improve the search efficiency, a local search mechanism is introduced. After each evolution cycle, the dominant solution in the non-dominated solution set is locally optimized. The local search adopts the pattern search method to improve the path quality by adjusting the waypoint position.

[0052] The solution set update strategy maintains external archives to store historical optimal solutions. The Pareto dominance relationship is used to determine whether new solutions enter the archive, and a gridding method is used to control the size of the archive. The archive is pruned regularly to maintain the representativeness and diversity of the solution set.

[0053] The convergence judgment adopts a multi-criteria termination mechanism. By monitoring the changes in the solution set of multiple generations, the convergence of the optimization process is evaluated. When the improvement of the solution set is lower than the threshold and meets the preset number of iterations, the calculation process is terminated.

[0054] Post-process the final Pareto optimal solution set. Eliminate sharp turns in the path through smoothing to ensure the continuity of the flight path. Verify that the smoothed path still meets all constraints and make local adjustments if necessary.

[0055] S206, inputting the Pareto optimal flight path set into a preset knowledge reasoning rule library to obtain a safety assessment result for each flight path; Specifically, a preset knowledge reasoning rule base is built based on the knowledge graph, which contains expert experience and historical data of flight safety assessment. The knowledge reasoning rules are expressed using fuzzy logic to transform qualitative safety assessment criteria into quantitative assessment rules.

[0056] Each Pareto optimal flight path is input into the knowledge reasoning rule base for safety assessment. First, the characteristic parameters of the flight path are extracted, including key indicators such as waypoint spacing, turning angle, and flight altitude change rate. The path information is converted into the input form required for rule reasoning through the feature extraction module.

[0057] Comprehensive evaluation of path safety is performed based on fuzzy reasoning mechanism. Path characteristic parameters are mapped to safety evaluation space through fuzzy membership function. Fuzzy rules use if-then form to describe safety judgment criteria, and rule weights are set in combination with expert knowledge.

[0058] During the reasoning process, a multi-level safety assessment framework is used. The first level evaluates the safe distance between the path and obstacles, and evaluates the collision risk through the minimum distance calculation. The second level evaluates the dynamic feasibility of the path, and determines the maneuverability requirements based on the turning radius and climbing rate. The third level evaluates the avoidance effect between the path and sensitive areas, and calculates the impact degree of sensitive areas.

[0059] The inference results are quantitatively expressed in risk levels. The fuzzy inference results are transformed into clear safety indicators through the inverse fuzzification process. The risk levels are divided into three levels: low risk, medium risk and high risk, and specific risk scores are given.

[0060] In order to improve the accuracy of the assessment, a situational awareness mechanism is introduced. The safety assessment rules are dynamically adjusted according to different time periods, weather conditions and task requirements. The dynamic optimization of the assessment criteria is achieved through the adaptive adjustment of the rule weights.

[0061] Construct a safety assessment report generation module. Record the assessment results of each flight path in detail, including the scores of various safety indicators, risk point distribution and improvement suggestions. The assessment report is presented in a structured form to facilitate subsequent analysis and decision-making.

[0062] The reliability of the evaluation results is verified by cross-validation. The stability and consistency of the evaluation results are verified by cross-checking different evaluation rule combinations. For paths with significant differences in evaluation results, key analysis and review are performed.

[0063] S207, calculating a comprehensive score for each flight path through a strategy evaluation method based on the safety assessment results and the priority and timeliness requirements included in the initial task set; Specifically, a hierarchical analysis evaluation model is constructed, and the safety assessment results, task priorities and timeliness requirements are used as evaluation indicators. By establishing a judgment matrix, the weight coefficients of each evaluation indicator are calculated to reflect the importance of different indicators in the comprehensive score. The safety assessment results are standardized and the risk level is converted into a quantitative score. The linear normalization method is used to unify the evaluation indicators of different scales into the same scoring range. The standardization process ensures the comparability of various indicators. The task priority information is extracted from the initial task set, and a priority quantification model is established. The priority coefficient is set according to the importance of the task, and the qualitative priority description is converted into a quantitative score through the priority function. The priority score reflects the urgency of task execution. The timeliness requirement is converted into a time constraint score. The degree of timeliness satisfaction is calculated by comparing the expected execution time of the flight path with the completion time limit required by the task. The time constraint score is described by a nonlinear function to highlight the penalty effect of task delay.

[0064] A comprehensive scoring calculation model was constructed, and multiple evaluation indicators were integrated using a weighted summation method. The scores of each indicator were multiplied by the corresponding weights and then summed to obtain a comprehensive score for each flight path. The scoring calculation took into account the mutual influence and compensation relationship between the indicators.

[0065] Introduce a sensitivity analysis mechanism to evaluate the impact of weight changes on the comprehensive score. Through weight perturbation analysis, identify key influencing factors and verify the stability of the scoring results. Sensitivity analysis provides a basis for weight adjustment.

[0066] Establish a ranking mechanism for the scoring results. Arrange the flight paths in descending order according to the comprehensive scores to form a path optimization sequence. The ranking results reflect the comprehensive performance of different paths in meeting multiple mission requirements.

[0067] Conduct comparative analysis of the scoring results. Use scatter plots and other visualization methods to display the score distribution of different paths in each evaluation dimension. Comparative analysis reveals the advantages and disadvantages of path performance.

[0068] Record the intermediate results of the scoring process, including the original score, standardized value and weight coefficient of each indicator. Detailed scoring records facilitate result traceability and verification, and improve the transparency of the evaluation process.

[0069] S209, selecting the flight path with the highest comprehensive score as the initial inspection route.

[0070] Specifically, based on the obtained comprehensive scoring results, the Pareto optimal flight path set is sorted and selected. According to the size of the comprehensive score, the flight path with the highest score is directly determined as the initial inspection route.

[0071] The selection process first analyzes the characteristics of the flight path with the highest score, which shows good comprehensive performance in multiple dimensions such as safety assessment results, mission priority satisfaction and timeliness requirements. By extracting the waypoint sequence of this path, a complete initial inspection route definition is formed.

[0072] The flight path with the highest score is selected as the initial inspection route, ensuring that the route is optimal in the multi-objective trade-off. This path not only meets the flight path constraints and environmental constraints, but also achieves a good balance in terms of path length, safety margin and energy consumption.

[0073] By selecting the flight path with the highest score as the initial inspection route, the transformation from multiple Pareto optimal paths to a single execution plan is achieved. This selection method based on comprehensive scoring ensures that the initial inspection route meets the predetermined mission requirements and execution conditions.

[0074] S102, controlling the drone cluster to perform inspection operations based on the initial inspection route, and obtaining multi-source inspection data of the inspection area through the drone cluster during the inspection operation; Specifically, the drone cluster is first grouped and controlled based on the initial inspection route, using a master-slave cluster control architecture, with one drone acting as the host responsible for task coordination and the remaining drones acting as slaves to perform specific inspection tasks. The system uses an adaptive formation control algorithm to calculate the flight state parameters of each drone in real time, including flight speed, heading angle, and spacing, to ensure that the drone cluster maintains a stable formation structure when performing inspection tasks.

[0075] During the inspection process, the drone cluster obtains comprehensive data of the inspection area through the multi-source sensor system it carries. The multi-source sensor system includes high-definition visible light cameras, infrared thermal imagers, lidars, and environmental sensors, which collect different types of inspection data. The high-definition visible light camera collects real-time image data of the inspection area, the infrared thermal imager obtains the temperature distribution information of the target object, the lidar scans to obtain three-dimensional point cloud data for spatial reconstruction, and the environmental sensor collects environmental parameters including temperature, humidity, noise, etc.

[0076] The system adopts a distributed data collection architecture. Each drone is equipped with an edge computing unit for preprocessing and preliminary analysis of the collected raw data. The edge computing unit performs operations such as data compression, noise filtering, and feature extraction to reduce the data transmission load. At the same time, the system establishes a data fusion mechanism based on spatiotemporal synchronization, and uniformly processes data from different sensors through timestamp alignment and spatial registration technology.

[0077] S103, constructing a community portrait based on multi-source inspection data, and inputting the community portrait into a preset classification reasoning model to output community management risks; Specifically, the multi-source inspection data is first cleaned and standardized, including removing outliers, supplementing missing data, and unifying data formats. Then, the system uses a multi-level feature extraction method to construct a community portrait, which includes three levels: physical space features, dynamic behavior features, and social relationship features.

[0078] At the level of physical space features, the system uses point cloud data collected by LiDAR to reconstruct the community's three-dimensional model and extract spatial indicators such as building density, green coverage, and road accessibility. Through high-definition image analysis, it extracts scene elements such as facility distribution, vehicle parking, and environmental sanitation to form a digital expression of the community's physical environment.

[0079] At the level of dynamic behavior characteristics, the system extracts dynamic indicators such as changes in crowd density, vehicle flow patterns, and frequency of facility use based on time series data analysis. The system uses spatiotemporal clustering algorithms to identify hot spots and high-frequency events in the community and build a community activity pattern map. At the same time, environmental sensor data analysis is used to extract dynamic environmental characteristics such as noise distribution and air quality changes.

[0080] At the level of social relationship characteristics, the system constructs the community social network structure by analyzing residents' activity trajectories and interactive behaviors, and extracts social characteristic indicators such as community cohesion and neighborhood relationship density. The system uses relationship graph technology to organize these characteristics into a structured knowledge network.

[0081] Based on the above three levels of features, the system uses tensor decomposition technology to integrate various features and build a unified community portrait model. The model uses a multi-dimensional tensor structure to store community features and realizes correlation analysis and pattern mining between features through tensor decomposition.

[0082] The system inputs the constructed community portrait into the preset classification reasoning model, which adopts a hybrid architecture combining deep learning and rule reasoning. The deep learning part uses graph neural network to process the structured features in the community portrait and extract potential risk patterns. The rule reasoning part is based on the domain expert knowledge base, matches the discovered abnormal patterns with the preset risk rules, and outputs specific community management risks.

[0083] The classification reasoning model adopts a hierarchical reasoning mechanism, first identifying hardware risks such as facility failures and safety hazards at the physical level, then discovering dynamic risks such as abnormal activities and violations at the behavioral level, and finally predicting social risks such as neighborhood conflicts and mass incidents at the social level. The system assesses the risk level of each type of hidden danger and generates a detailed risk description report.

[0084] Based on the above embodiment, as an optional embodiment, building a community portrait based on multi-source inspection data includes: S301, performing multi-modal data fusion processing on multi-source inspection data to obtain fused inspection data; Specifically, for different modal data such as image data, video data, infrared thermal imaging data and laser point cloud data contained in multi-source inspection data, a multimodal data fusion processing method is used for unified processing. Multimodal data fusion achieves data integration through two levels: feature-level fusion and decision-level fusion.

[0085] In feature-level fusion, each modality data is first preprocessed, including noise removal, data enhancement, and feature extraction. Color, texture, and edge features are extracted from image data, time series features and motion features are extracted from video data, temperature distribution features are extracted from infrared thermal imaging data, and geometric features and spatial distribution features are extracted from laser point cloud data.

[0086] In decision-level fusion, a deep learning model is used to fuse the extracted features. The attention mechanism is used to assign weights to different modal features to achieve adaptive fusion of features. The fusion process takes into account the characteristics and relevance of various types of data to ensure that the fusion result retains the key information of the original data.

[0087] Through multimodal data fusion processing, the fused inspection data is obtained.

[0088] S302, constructing a community three-dimensional model based on the fused inspection data, and performing spatial registration on the community three-dimensional model to obtain a registered community three-dimensional model; Specifically, a 3D model of the community is constructed based on the fused inspection data. The 3D geometric structure of community elements such as buildings, roads, and vegetation is reconstructed using laser point cloud data and image data. Through feature point matching and depth estimation algorithms, the 2D image information is mapped to 3D space to form a 3D model with real texture.

[0089] The generated community 3D model is spatially registered. First, the feature points and registration control points in the model are extracted to establish the spatial coordinate transformation relationship. A combination of rigid transformation and non-rigid transformation is used to achieve accurate alignment of the model in spatial position, direction and scale.

[0090] In the process of spatial registration, GPS positioning information and geographic information system data are used as references to ensure that the 3D model is consistent with the actual geographic coordinate system. The registration parameters are optimized through the iterative closest point algorithm to improve the spatial accuracy of the model.

[0091] Through spatial registration, a registered 3D model of the community is obtained. The registered model has accurate spatial position information and proportional relationship, and realizes the precise correspondence between the virtual model and the actual environment. This spatially registered 3D model provides an accurate spatial reference framework for the construction of community portraits.

[0092] S303, associating and annotating the registered community three-dimensional model with the multi-source inspection data to form structured data with spatial location information; Specifically, the registered community 3D model is used as a spatial reference to associate and annotate the multi-source inspection data. Based on the spatiotemporal information when the inspection data is collected, a mapping relationship is established between the data and the corresponding spatial position in the 3D model.

[0093] In the process of associative annotation, the inspection data such as image data, video data, infrared thermal imaging data, etc. are annotated to the corresponding positions of the 3D model using the spatial coordinate correspondence. The associative relationship between the inspection data and the geometric elements of the model is established through the spatial index mechanism.

[0094] The annotation process uses semantic segmentation to match the target objects identified in the inspection data with the entities in the 3D model. By establishing a semantic labeling system, a structured association between the inspection data and the spatial location is achieved.

[0095] Through associative annotation, structured data with spatial location information is formed. This structured data contains the spatial location, attribute characteristics and time information of the inspection object, so that the inspection data has a clear spatial reference. Structural processing improves the efficiency of data retrieval and analysis, and provides data support for the accurate construction of community portraits.

[0096] S304: Based on preset semantic rules, semantic annotation is performed on the structured data to obtain a community portrait.

[0097] Specifically, the structured data is semantically annotated based on preset semantic rules. The preset semantic rules define the semantic attributes and relationships of various elements in the community, including standardized description rules such as building types, facility status, and environmental characteristics.

[0098] In the process of semantic annotation, spatial entities in structured data are matched with semantic rules. Through the semantic reasoning engine, data features are mapped to corresponding semantic labels to form a semantic association network between entities.

[0099] Semantic annotation uses natural language processing technology to convert abnormal conditions and facility status found during inspections into standardized semantic descriptions. The annotation process follows preset semantic rules to ensure the consistency and standardization of the description.

[0100] After semantic annotation, a community portrait is obtained. The community portrait contains the location information, attribute characteristics and semantic relationships of spatial entities, forming a complete knowledge expression of the community. This semantic community portrait supports intelligent data retrieval and analysis, and provides a knowledge basis for community management decisions.

[0101] S104, generating structured warning events based on the categories and levels of community management risks, and determining corresponding disposal tasks based on the warning events; Specifically, the hidden dangers in community management are intelligently classified and risk assessed, and a multi-dimensional classification system is used to describe the hidden dangers in a standardized manner. The classification system includes dimensions such as the source of hidden dangers, scope of impact, and degree of urgency. The system converts unstructured hidden danger descriptions into structured hidden danger feature vectors through semantic analysis technology.

[0102] The system classifies hidden dangers based on a preset risk assessment matrix, which comprehensively considers the two dimensions of probability of occurrence and degree of harm. For each hidden danger, the system calculates its risk score and classifies it into four levels: major hidden danger, serious hidden danger, general hidden danger and minor hidden danger. The system adopts a fuzzy comprehensive evaluation method to achieve accurate classification of hidden danger levels by establishing an evaluation index set and a weight set.

[0103] After completing the hazard classification and level assessment, the system starts the warning event generation mechanism. This mechanism builds a standard description of the warning event based on the event ontology model. The event ontology model defines the core elements of the warning event, including event type, location, impact range, duration and other attributes. The system uses the ontology reasoning engine to map the hazard characteristics to the warning event attributes and generate a structured warning event description.

[0104] For each warning event, the system uses a task decomposition algorithm to generate the corresponding disposal task. The task decomposition algorithm is based on a preset disposal procedure library, which decomposes the complex disposal process into a series of specific execution tasks. The disposal procedure library stores standard disposal processes for different types of hidden dangers, including emergency response, resource allocation, and collaborative disposal. The system selects the most suitable disposal procedure through rule matching and refines the task according to the actual situation.

[0105] The system uses a task dependency graph model to describe the relationship between processing tasks. The task dependency graph represents the execution order and parallel relationship of tasks through a directed acyclic graph structure. Based on the critical path analysis method, the system determines the priority and timing of tasks and assigns execution parameters to each task, including execution time limit, required resources and execution subject.

[0106] To ensure the feasibility of the disposal task, the system obtains available resource information in real time, including drone status, emergency supplies, human resources, etc., and makes the best match through the resource scheduling algorithm. The resource scheduling algorithm considers the two optimization goals of resource utilization efficiency and response time, and uses a heuristic search method to solve the optimal resource allocation plan.

[0107] Based on the above embodiment, as an optional embodiment, a structured warning event is generated based on the category and level of community management hidden dangers, and a corresponding handling task is determined based on the warning event, including: Obtaining treatment rules that match the category of community management hazards from a preset hazard treatment rule library; Determine the priority of handling community management risks based on the level of community management risks; According to the disposal rules and disposal priorities, structured warning events are generated, and based on the warning events and combined with the preset task templates, corresponding disposal tasks are generated.

[0108] Specifically, for the detected community management hidden dangers, the matching disposal rules are first obtained from the preset hidden danger disposal rule library. The hidden danger disposal rule library stores the standard disposal processes, required resources and disposal requirements for different types of hidden dangers. The corresponding disposal rules are extracted by matching the hidden danger categories.

[0109] The disposal priority is determined based on the level of community management hidden dangers. The hidden danger level includes evaluation indicators such as the degree of danger, scope of impact and urgency. According to the comprehensive score of these indicators, the disposal priority is divided into three levels: emergency disposal, priority disposal and general disposal.

[0110] According to the disposal rules and disposal priorities, structured warning events are generated. Warning events contain standardized information such as hidden danger location, type description, hazard level, disposal time limit, etc. Based on the specific circumstances of the warning event, the corresponding disposal task is generated in combination with the preset task template.

[0111] S105, decomposing the disposal task based on a preset ontology reasoning framework to obtain multiple hierarchical subtask descriptions, and selecting a corresponding execution task subject based on the subtask descriptions; Specifically, the preset ontology reasoning framework is loaded, which includes the task ontology model and the reasoning rule library. The task ontology model uses a hierarchical conceptual structure to describe the task domain knowledge, and defines core concepts such as task type, execution conditions, resource requirements, and their relationships. The reasoning rule library stores the expert experience of task decomposition and describes the logical relationship of task decomposition through a formal rule language.

[0112] The system inputs the disposal task into the ontology reasoning framework and starts the top-down task decomposition process. First, the system extracts the key elements of the disposal task through semantic analysis, including task objectives, execution constraints, completion standards, etc. Based on these elements, the system performs concept matching in the task ontology and determines the category attributes of the task.

[0113] Task decomposition is performed recursively, and the system calls the corresponding decomposition rules according to the task category. The decomposition rules describe the composition relationship of tasks based on the AND or tree structure, and the system obtains the subtask set of each layer through rule reasoning. For each subtask, the system continues to apply the decomposition rules until an atomic task that cannot be decomposed is obtained. These atomic tasks constitute the lowest-level execution unit.

[0114] During the task decomposition process, the system also builds a task dependency network, which describes the pre- and post-requisite and parallel relationships between subtasks through a directed graph structure. The system uses critical path analysis to determine the timing of task execution and annotates the time window and resource requirements for each subtask.

[0115] For the decomposed subtask description, the system starts the execution subject selection mechanism. The mechanism first extracts the task execution requirements based on the task feature vector, including technical requirements, equipment specifications, response time limits and other dimensions. The system retrieves candidate subjects that meet the requirements from the execution subject resource pool, including different types of drones, ground robots and manual intervention.

[0116] The system uses a multi-objective decision model to match execution subjects. The decision objectives include execution efficiency, resource consumption and task suitability. The system calculates the weight of each indicator through the hierarchical analysis method and uses the fuzzy comprehensive evaluation method to comprehensively score the candidate subjects. The system selects the subject with the highest score as the final execution subject and generates an execution task allocation plan.

[0117] Based on the above embodiment, as an optional embodiment, the disposal task is decomposed based on a preset ontology reasoning framework to obtain multiple hierarchical subtask descriptions, including: S401, acquiring task decomposition rules and task semantic description related to the handling task from a preset ontology knowledge base; Specifically, the task decomposition rules and task semantic descriptions related to the disposal task are obtained from the preset ontology knowledge base. The ontology knowledge base is a standardized knowledge system that contains concepts, relationships and rules in the field of community management.

[0118] During the acquisition process, the ontology reasoning engine performs semantic analysis on the disposal task to identify the core elements and key attributes of the task. The system matches the corresponding task decomposition rules in the ontology knowledge base based on the semantic features of the task.

[0119] The task decomposition rules define the decomposition methods and hierarchical relationships of different types of tasks, including the dependencies, execution order, and resource requirements between tasks. The task semantic description standardizes the expression and attribute definition of each task node.

[0120] By extracting task decomposition rules and semantic descriptions from the ontology knowledge base, standardized knowledge support is provided for subsequent task decomposition. This ontology-based task parsing method ensures the standardization and semantic consistency of task decomposition and lays the foundation for building a hierarchical subtask system.

[0121] S402, splitting the handling task into multiple first-level subtasks based on the task decomposition rule; Specifically, the disposal task is split based on the task decomposition rule to form multiple first-level subtasks. The task decomposition rule defines the basic components and splitting criteria of the disposal task, including task objectives, execution conditions and completion criteria.

[0122] In the process of task decomposition, the core objectives and key links of the disposal task are first identified. According to the task classification system defined in the task decomposition rules, the disposal task is divided into first-level subtasks such as on-site investigation, emergency response, resource scheduling, and follow-up.

[0123] Each first-level subtask has clear task boundaries and execution requirements. Through the constraints of task decomposition rules, the connectivity and integrity between first-level subtasks are ensured to avoid omissions or duplications in the task splitting process.

[0124] The first-level subtask system formed clearly shows the main components of the disposal task and provides a framework structure for subsequent detailed decomposition.

[0125] S403, performing semantic analysis on each first-level subtask to determine the task type, task goal, and execution requirements of the first-level subtask; Specifically, semantic analysis is performed on each first-level subtask, and the core semantic information of the task is extracted through semantic parsing technology. Semantic analysis is based on the task semantic description rules to identify keywords and semantic structures in the task description.

[0126] First, determine the task type of the first-level subtask, including inspection, disposal, coordination, etc. Through semantic feature matching, match the task description with the preset task type standard to determine the basic attributes of the task.

[0127] The analysis of task objectives extracts the semantic elements related to the objectives in the task description and clarifies the specific indicators and acceptance criteria for task completion. The semantic analysis of task objectives ensures the clarity of task orientation.

[0128] S404, further split the first-level subtask into multiple second-level subtasks according to the task type, task goal, execution requirements and task decomposition rules, and repeat the above task decomposition process until all subtasks reach a granularity that can be directly executed, and obtain multiple hierarchical subtask descriptions.

[0129] Specifically, based on the determined task type, task goal and execution requirements, combined with the task decomposition rules, the first-level subtasks are further divided. The task decomposition rules define the detailed standards and splitting methods for different task types.

[0130] The system splits the first-level subtask into multiple second-level subtasks according to the task decomposition rules. For example, the field survey task is split into second-level subtasks such as data collection, field recording, and problem location. Each second-level subtask inherits the constraints of the parent task and has more specific execution content.

[0131] For the second-level subtasks obtained by splitting, the system continues to perform the same decomposition process. Through task type judgment, goal decomposition and requirement refinement, the second-level subtasks are further split into third-level subtasks. This recursive task decomposition continues until all subtasks reach a granularity that can be directly executed.

[0132] The multiple hierarchical subtask descriptions finally formed constitute a complete task execution system.

[0133] S106, generating a task set corresponding to the execution task subject according to the execution task subject and the corresponding subtask description; Specifically, we first establish a task set generation framework, which converts the abstract task description into a specific execution instruction set based on the characteristics of the execution task subject and the requirements of the subtask description. The system uses a task template library to store standard execution templates for different types of execution task subjects, which define the basic process and control parameters of task execution.

[0134] The system performs semantic analysis on the subtask description and extracts the key elements of task execution, including execution location, operation object, execution time limit and completion standard, etc. These elements are matched with the parameters in the task template through semantic mapping. The system adopts parametric modeling method to convert abstract task requirements into specific execution parameters.

[0135] For the drone-like mission subject, the system generates a flight control instruction set, including take-off and landing point coordinates, waypoint sequence, flight altitude, speed curve and other flight parameters. The system calculates the optimal flight path through the trajectory planning algorithm, taking into account factors such as obstacle avoidance, energy consumption optimization and mission constraints. At the same time, the system generates sensor control instructions and defines data collection strategies and processing procedures.

[0136] For ground robots that perform tasks, the system generates motion control instruction sets and operation instruction sets. Motion control instructions include parameters such as path planning, speed control, and posture adjustment, and operation instructions define specific operation action sequences. The system generates smooth motion trajectories through motion planning algorithms to ensure that the robot can complete tasks safely and efficiently.

[0137] The system builds an independent task queue for each task execution subject and uses a task scheduling algorithm to sort and organize tasks. The task scheduling algorithm takes into account task priority, time constraints, and resource utilization efficiency, and solves the optimal execution order through dynamic programming methods. The system also establishes a task synchronization mechanism to ensure coordination and cooperation between multiple task execution subjects.

[0138] To improve the reliability of task execution, the system embeds an exception handling mechanism in the task set. This mechanism defines the handling procedures for different types of abnormal situations, including strategies for coping with communication interruptions, equipment failures, environmental interference, etc. The system detects the execution status in real time through status monitoring, and immediately initiates the corresponding processing flow once an abnormality is found.

[0139] Based on the above embodiment, as an optional embodiment, a task set corresponding to the execution task subject is generated according to the execution task subject and the corresponding subtask description, including: S501, acquiring capability attribute information of each task execution subject, wherein the capability attribute information includes executable task types, task execution efficiency and task execution constraint conditions; Specifically, the system obtains the capability attribute information of each task execution subject, which is the basis for task allocation. The task execution subject includes specific staff, equipment or system.

[0140] First, the system reads the executable task type information of the task execution subject and clarifies the scope of tasks that each subject can undertake. For example, a task execution subject has the execution capabilities of inspection, maintenance, emergency response and other task types. Secondly, the task execution efficiency data is obtained, including quantitative indicators such as historical task completion status and average execution time. Task execution efficiency reflects the actual work ability and work quality of the task execution subject.

[0141] Finally, the system extracts task execution constraints, including working time restrictions, geographical area restrictions, qualification requirements, etc. These constraints determine the scope of tasks that can be undertaken by the task execution entity.

[0142] S502, based on the capability attribute information, the subtask descriptions are screened to obtain candidate subtasks that match the capabilities of each task execution subject; Specifically, based on the acquired capability attribute information, the subtask descriptions are screened and matched. The screening process determines the candidate subtasks suitable for each task execution subject by comparing the capability characteristics of the task execution subject with the subtask requirements.

[0143] First, the system selects subtasks of matching types based on the executable task types of the task execution subject. Through the correspondence between task types, the task types that the task execution subject is not competent for are excluded. Next, the system evaluates whether the task execution subject meets the timeliness requirements of the subtasks based on the task execution efficiency data. The completion time limit of the subtask is matched with the efficiency index of the task execution subject to select candidate subtasks with time requirements. Then, the system checks the task execution constraints to ensure that the candidate subtasks meet the various restriction requirements of the task execution subject. Subtasks that exceed the capabilities of the task execution subject are eliminated through matching and filtering of constraint conditions.

[0144] S503, determining the execution order of each candidate subtask according to the task dependency relationship of the candidate subtasks; Specifically, a reasonable execution order is determined based on the task dependency relationship between candidate subtasks. The task dependency relationship reflects the connection and association constraint relationship between subtasks.

[0145] First, the system analyzes the input-output relationship of the candidate subtasks and identifies the data flow and resource sharing between tasks. By sorting out the information transmission chain between tasks, it is clear which tasks need to rely on the execution of other tasks. Then, the system builds a task dependency network and connects the candidate subtasks with dependencies to form a directed graph structure. The task dependency network clearly shows the order and parallel possibility between tasks. Then, the system uses a topological sorting algorithm to calculate the execution sequence of the candidate subtasks based on the task dependency network. On the premise of ensuring that the dependencies are met, the system optimizes the sorting results to maximize the efficiency of task execution.

[0146] S504: Generate a task set based on the execution order and in combination with the task execution constraints of each task execution subject.

[0147] Specifically, this application generates a final task set based on the determined execution order and the task execution constraints of the task execution subject. The task set is the specific work content and execution plan assigned to each task execution subject.

[0148] First, the system matches the execution order with the working time limit of the task execution subject to ensure that the task arrangement meets the working time constraint. The system reasonably allocates the task execution period according to the work shift and available time of the task execution subject. Next, the system considers the geographical area restrictions of the task execution subject and optimizes the combination allocation of tasks in similar areas. By clustering geographical locations, the travel time of the task execution subject between different regions is reduced. Then, the system combines the qualification requirements and professional restrictions of the task execution subject to ensure that the task assignment meets the relevant specifications and standards. The system checks the execution qualification requirements of each task to ensure that the task execution subject has the corresponding execution qualifications.

[0149] The final generated task set contains complete information such as task content, execution time, execution location, etc. This task set generation method based on execution order and constraints realizes the scientificity and rationality of task allocation and improves overall work efficiency.

[0150] S107, for the task execution subject, a preset multi-agent reinforcement learning framework is used to perform spatiotemporal collaborative arrangement of the task set to form an overall collaborative plan, and the overall collaborative plan is expressed as an interpretable scheduling decision instruction; Specifically, a preset multi-agent reinforcement learning framework is constructed, which models each task execution subject as an independent intelligent agent, and realizes the spatiotemporal coordination of tasks through deep reinforcement learning methods. The system adopts a layered architecture design, including an environmental perception layer, a decision learning layer, and a collaborative control layer.

[0151] At the environmental perception layer, the system constructs the task environment state space, which contains dimensional information such as task temporal and spatial distribution, resource constraints, and execution conditions. The system converts high-dimensional environmental information into low-dimensional feature vectors through the state encoder, and uses the attention mechanism to highlight key state information to provide input for decision learning.

[0152] In the decision learning layer, the system configures a deep neural network as a policy network and a value network for each agent. The policy network is responsible for generating action decisions, including task selection, execution timing, and path planning; the value network evaluates the current state and the value of the decision, and guides the direction of policy optimization. The system adopts a distributed learning architecture and stores and reuses historical decision data through an experience playback mechanism.

[0153] The system has designed a multi-level reward function system, including evaluation indicators such as task completion, resource utilization, and spatiotemporal coordination. The reward function guides the agent to learn the optimal decision-making strategy. The system adopts a strategy optimization algorithm based on the trust region to improve the convergence speed while ensuring learning stability.

[0154] At the collaborative control layer, the system implements an inter-agent communication protocol based on the attention mechanism, enabling each agent to perceive the status and intention of other agents. The system models the interaction relationship between agents through graph neural networks to achieve group-level decision optimization. At the same time, the system adopts a hierarchical collaborative mechanism to coordinate tasks at both the local and global levels.

[0155] The system transforms the task timing relationship into mathematical constraints through spatiotemporal constraint modeling, and uses mixed integer programming to solve the initial scheduling plan. On this basis, the plan is continuously optimized and adjusted through reinforcement learning to achieve dynamic coordination of tasks. The system has established a task conflict detection mechanism to avoid conflicts during execution through predictive planning.

[0156] S108, controlling the drone cluster to execute the task according to the scheduling decision.

[0157] Specifically, this application establishes a UAV mission execution control framework, which parses the scheduling decision instructions into specific flight control instructions and mission operation instructions. The system adopts a distributed control architecture, including an instruction parsing layer, a flight control layer, and a mission execution layer.

[0158] At the instruction parsing layer, the system converts the scheduling decision instructions into standardized control parameters through the instruction parser. The instruction parser uses semantic analysis technology to identify the key control elements in the instructions, including flight paths, mission requirements, time constraints, etc. The system converts these elements into control quantities that can be directly executed by the drone through parameter mapping.

[0159] At the flight control layer, the system establishes a flight controller based on model prediction. The controller predicts the motion state of the drone through the dynamic model and generates the optimal control input in real time. The system uses an adaptive control algorithm to compensate for external disturbances and model errors through online parameter estimation to ensure the accuracy and robustness of flight control.

[0160] The system implements a multi-level navigation control strategy, including global path planning and local trajectory generation. Global path planning generates the optimal route by considering terrain constraints and no-fly zones. Local trajectory generation uses a dynamic window method to ensure flight safety through real-time obstacle avoidance. The system also establishes a path tracking controller to ensure that the UAV accurately tracks the planned trajectory through feedback control.

[0161] At the mission execution layer, the system builds a mission state machine to manage the transition and control of different mission stages. The state machine defines the standard process of mission execution, including takeoff, enroute flight, mission operation, and return landing. The system controls state transitions through an event trigger mechanism to ensure the continuity and integrity of mission execution.

[0162] Based on the above embodiment, as an optional embodiment, the overall coordination scheme is represented as an interpretable scheduling decision instruction, including: Based on the preset instruction template library, the overall coordination plan is converted into a standardized scheduling instruction sequence. Add an explainable description to each scheduling instruction to obtain an explainable scheduling instruction.

[0163] Specifically, the overall coordination plan is converted into scheduling decision instructions that can be understood and executed by the executors. The instruction template library stores various standardized instruction formats and expression specifications.

[0164] First, the system selects an instruction template that matches the overall coordination plan from the instruction template library. The instruction template defines the basic structure and key elements of the scheduling instruction, including standard fields such as instruction object, execution time, and execution content. The system fills the specific content of the overall coordination plan into the instruction template to form a standardized scheduling instruction sequence. Next, the system adds an explainable description to each scheduling instruction. The explainable description contains explanatory content such as instruction background, execution purpose, and expected effect. The system generates clear and easy-to-understand instruction descriptions by extracting relevant information from the overall coordination plan to help executors understand the specific meaning and execution points of the instructions.

[0165] This conversion method based on instruction templates ensures the standardization and uniformity of scheduling instructions. By adding interpretable descriptions, the comprehensibility and executability of instructions are improved, and misunderstandings and deviations in the execution of instructions are reduced. The final interpretable scheduling decision instructions provide a strong guarantee for the smooth implementation of the overall collaborative solution.

[0166] On the other hand, the present application also provides a community management system based on drones, such as Figure 2 The system comprises: Route planning module 1, used to obtain the initial inspection task of the drone cluster and plan the initial inspection route of the drone cluster based on the initial inspection task; Data acquisition module 2, used to control the drone cluster to perform inspection operations based on the initial inspection route, and obtain multi-source inspection data of the inspection area through the drone cluster during the inspection operation; Hidden danger identification module 3, used to build community portraits based on multi-source inspection data, and input the community portraits into a preset classification reasoning model to output community management hidden dangers; The disposal task generation module 4 is used to generate structured warning events based on the categories and levels of community management hidden dangers, and determine corresponding disposal tasks based on the warning events; The task decomposition module 5 is used to decompose the disposal task based on the preset ontology reasoning framework, obtain multiple hierarchical subtask descriptions, and select the corresponding execution task subject based on the subtask descriptions; The task set generation module 6 is used to generate a task set corresponding to the execution task subject according to the execution task subject and the corresponding subtask description; The scheduling decision instruction module 7 is used to perform spatiotemporal collaborative arrangement of the task set for the task execution subject using a preset multi-agent reinforcement learning framework to form an overall collaborative plan, and express the overall collaborative plan as an interpretable scheduling decision instruction; The task execution module 8 is used to control the drone cluster to perform tasks according to the scheduling instructions.

[0167] Please refer to Figure 3 The application also discloses an electronic device. Figure 3 The electronic device 300 may include: at least one processor 301 , at least one network interface 304 , a user interface 303 , a memory 305 , and at least one communication bus 302 .

[0168] The communication bus 302 is used to realize the connection and communication between these components.

[0169] The user interface 303 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0170] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0171] Among them, the processor 301 may include one or more processing cores. The processor 301 uses various interfaces and lines to connect various parts in the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and calling data stored in the memory 305. Optionally, the processor 301 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field~Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 301 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 301, and it can be implemented separately through a chip.

[0172] Among them, the memory 305 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read~Only Memory). Optionally, the memory 305 includes a non-transitory computer-readable medium. The memory 305 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 305 may optionally be at least one storage system located away from the aforementioned processor 301. Refer to Figure 3 The memory 305 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program for a flocculant addition analysis method.

[0173] exist Figure 3 In the electronic device 300 shown, the user interface 303 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 301 can be used to call the application program storing the road assessment method in the memory 305. When executed by one or more processors 301, the electronic device 300 executes one or more methods in the above-mentioned embodiments. It should be noted that for the aforementioned method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application. In the above embodiments, the description of each embodiment has its own emphasis. For the parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0174] In the several implementations provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of the system or unit can be electrical or other forms.

[0175] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0176] The present application also provides a computer storage medium that can store multiple instructions, which are suitable for being loaded and executed by a processor as described above. Figure 1 The road assessment method of the embodiment shown in the figure can be found in the specific implementation process. Figure 1 The specific description of the illustrated embodiment will not be repeated here.

[0177] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0178] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory, including a number of instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: various media that can store program codes, such as USB flash drives, mobile hard drives, magnetic disks or optical disks.

[0179] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0180] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A community management method based on drones, characterized in that: The method comprises: Obtaining an initial inspection task for the drone cluster, and planning an initial inspection route for the drone cluster based on the initial inspection task; Controlling the drone cluster to perform inspection operations based on the initial inspection route, and acquiring multi-source inspection data of the inspection area through the drone cluster during the inspection operation; Building a community portrait based on the multi-source inspection data, and inputting the community portrait into a preset classification reasoning model to output community management risks; Generate structured warning events based on the categories and levels of the community management hidden dangers, and determine corresponding disposal tasks based on the warning events; Decomposing the processing task based on a preset ontology reasoning framework to obtain multiple hierarchical subtask descriptions, and selecting a corresponding execution task subject based on the subtask descriptions; Generate a task set corresponding to the execution task subject according to the execution task subject and the corresponding subtask description; For the task execution subject, a preset multi-agent reinforcement learning framework is used to perform spatiotemporal collaborative arrangement of the task set to form an overall collaborative plan, and the overall collaborative plan is expressed as an interpretable scheduling decision instruction; The drone cluster is controlled to perform tasks according to the scheduling decision instructions.

2. The method according to claim 1, characterized in that: The planning of the initial inspection route of the drone cluster based on the initial inspection task includes: Associating and matching the initial inspection task with a preset task template to obtain a standardized inspection task description; Acquire community geographic information data and a community management schedule, and generate an executable initial task set based on the standardized inspection task description combined with the community geographic information data and the community management schedule; Obtaining community topography data, building layout data, and sensitive area distribution data corresponding to the initial task set; Constructing a multi-constraint and multi-objective path optimization model based on the flight route according to the community topography data, the building layout data and the sensitive area distribution data; The multi-constraint and multi-objective path optimization model is solved by using a heuristic search and intelligent optimization algorithm to obtain a set of Pareto optimal flight paths that meet preset constraints; Inputting the Pareto optimal flight path set into a preset knowledge reasoning rule library to obtain a safety assessment result for each flight path; Calculating a comprehensive score for each flight path using a strategy evaluation method based on the safety assessment results and the priority and timeliness requirements included in the initial task set; The flight path with the highest comprehensive score is selected as the initial inspection route.

3. The method according to claim 1, characterized in that: The constructing of a community portrait based on the multi-source inspection data includes: Performing multi-modal data fusion processing on the multi-source inspection data to obtain fused inspection data; Based on the fused inspection data, a community three-dimensional model is constructed, and the community three-dimensional model is spatially registered to obtain a registered community three-dimensional model; Associating and annotating the registered community three-dimensional model with the multi-source inspection data to form structured data with spatial location information; Based on preset semantic rules, the structured data is semantically annotated to obtain the community portrait.

4. The method according to claim 1, characterized in that The generating of structured warning events based on the categories and levels of the community management hidden dangers, and determining corresponding handling tasks based on the warning events, include: Obtaining a disposal rule that matches the category of the community management hidden danger from a preset hidden danger disposal rule library; Determining the priority of handling the community management hidden danger based on the level of the community management hidden danger; The structured warning event is generated according to the handling rule and the handling priority, and the corresponding handling task is generated based on the warning event and in combination with a preset task template.

5. The method according to claim 1, characterized in that: The processing task is decomposed based on the preset ontology reasoning framework to obtain multiple hierarchical subtask descriptions, including: Acquire task decomposition rules and task semantic description related to the handling task from a preset ontology knowledge base; Based on the task decomposition rule, split the handling task into multiple first-level subtasks; Performing semantic analysis on each of the first-level subtasks to determine the task type, task objective, and execution requirements of the first-level subtask; According to the task type, the task objective, the execution requirements and the task decomposition rules, the first-level subtask is further split into multiple second-level subtasks, and the above task decomposition process is repeated until all subtasks reach a granularity that can be directly executed, thereby obtaining the multiple hierarchical subtask descriptions.

6. The method according to claim 1, characterized in that The step of generating a task set corresponding to the execution task subject according to the execution task subject and the corresponding subtask description includes: Acquiring capability attribute information of each of the task execution subjects, wherein the capability attribute information includes executable task types, task execution efficiency, and task execution constraints; Based on the capability attribute information, the subtask descriptions are screened to obtain candidate subtasks that match the capabilities of each task execution subject; Determining the execution order of each of the candidate subtasks according to the task dependencies of the candidate subtasks; Based on the execution order and in combination with the task execution constraints of each task execution subject, the task set is generated.

7. The method according to claim 1, characterized in that The step of expressing the overall coordination scheme as an interpretable scheduling instruction comprises: Based on the preset instruction template library, the overall coordination scheme is converted into a standardized scheduling instruction sequence. An explainable description is added to each scheduling instruction to obtain the explainable scheduling instruction.

8. A community management system based on drones, characterized in that: The system comprises: A route planning module, used to obtain an initial inspection task of the drone cluster and plan an initial inspection route of the drone cluster based on the initial inspection task; A data acquisition module, used to control the drone cluster to perform inspection operations based on the initial inspection route, and to acquire multi-source inspection data of the inspection area through the drone cluster during the inspection operation; A hidden danger identification module is used to construct a community portrait based on the multi-source inspection data, and input the community portrait into a preset classification reasoning model to output community management hidden dangers; A disposal task generation module, used to generate structured warning events based on the categories and levels of the community management hidden dangers, and determine corresponding disposal tasks based on the warning events; A task decomposition module is used to decompose the processing task based on a preset ontology reasoning framework to obtain multiple hierarchical subtask descriptions, and select a corresponding execution task subject based on the subtask descriptions; A task set generation module, used to generate a task set corresponding to the execution task subject according to the execution task subject and the corresponding subtask description; The scheduling decision instruction module is used to use a preset multi-agent reinforcement learning framework to perform spatiotemporal collaborative arrangement of the task set for the task execution subject, form an overall collaborative plan, and express the overall collaborative plan as an interpretable scheduling decision instruction; The task execution module is used to control the drone cluster to perform tasks according to the scheduling instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes a processor, a memory and a transceiver, the memory is used to store instructions, the transceiver is used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

Images