Unmanned aerial vehicle precise source control method and system based on embodied intelligence

Abstract

The application discloses a UAV atmospheric pollution accurate tracing control method and system based on embodied intelligence, which comprises the following steps: predicting atmospheric pollutant concentration, triggering real-time early warning when the pollution exceeds the standard, dispatching UAV to carry out patrol, tracing the pollution source based on the atmospheric pollutant concentration data of the monitoring station, combining the three-dimensional wind field and the Lagrangian diffusion model, planning the patrol path of the UAV based on the pollution source tracing result, using the AI model to carry out quantitative analysis of the pollution event based on the pollution source tracing result, the UAV return data and the patrol path, and generating the pollution event report. The application focuses on solving the problem that the UAV path planning is disconnected with the tracing result, and realizes the sequential control process of early warning, tracing, path planning, patrol collection and analysis report.

Description

Technical Field

[0001] This invention relates to the field of ecological environment monitoring and pollution prevention technology, specifically to a method and system for precise source tracing and control of air pollution based on embodied intelligence unmanned aerial vehicles (UAVs). Background Technology

[0002] As ecological and environmental governance develops towards digitalization, intelligence, and three-dimensionality, drone technology is increasingly being used in air pollution monitoring, emergency source tracing, and law enforcement evidence collection. Existing drone monitoring systems typically employ gas sensors or multispectral monitoring equipment, performing sampling flights along pre-set routes and transmitting the collected data back to ground stations or cloud platforms for analysis to assist in determining pollution distribution and locating sources. While this technology has achieved some success in pollution emergency response and regional patrols, problems remain, including rigid route planning, fragmented data processing, and limited accuracy of source tracing models.

[0003] For example, patent application CN110763804A, entitled "A System and Method for Tracing Air Pollution Sources Based on UAVs," discloses a technical solution for rapidly locating pollution sources using UAVs equipped with air monitoring devices, calculating the direction of pollution sources through concentration changes and wind direction information. Patent application CN112580893A, entitled "A Method for Multi-UAV Atmospheric Monitoring Path Planning Based on an Improved Genetic Algorithm," proposes an approach to optimize multi-UAV trajectory planning using an improved genetic algorithm. While optimizing path length and turning angles, it is disconnected from the pollution source tracing process. With fixed monitoring station locations and static meteorological parameters, the algorithm does not incorporate pollution weights, wind field changes, and other factors into its fitness assessment, failing to achieve intelligent path adaptive optimization for source tracing tasks. Patent application CN120542713A, entitled "A System for Urban Air Pollution Monitoring and Prevention Based on Radar and AI Collaboration Using UAVs," further combines radar scanning, AI recognition, and UAV inspection to achieve multi-source information fusion and intelligent decision-making.

[0004] The aforementioned technologies still have limitations in intelligent adaptive path planning, 3D wind field modeling, and pollution source tracing, and lack comprehensive performance capabilities such as real-time learning, autonomous scheduling, and automatic report generation. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a method and system for precise source tracing and control of air pollution using unmanned aerial vehicles (UAVs) based on embodied intelligence, which improves the timeliness, accuracy, and intelligence of environmental monitoring.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] The present invention provides a method for precise source tracing and control of air pollution from unmanned aerial vehicles (UAVs) based on embodied intelligence, comprising:

[0008] Predict air pollutant concentrations and trigger real-time warnings when pollution levels exceed standards, and dispatch drones to conduct patrols;

[0009] Based on atmospheric pollutant concentration data from monitoring stations, combined with three-dimensional wind fields and Lagrange diffusion models, pollution sources can be traced.

[0010] Based on the pollution source tracing results, a genetic algorithm is used to plan the drone's patrol path, outputting an optimal execution path and several alternative patrol paths;

[0011] Based on the pollution source tracing results, drone data transmission and patrol routes, an AI model is used to conduct quantitative analysis of pollution events and generate pollution event reports.

[0012] A further improvement of the present invention is that a spatiotemporal graph neural network is used to predict the concentration of air pollutants. The input of the spatiotemporal graph neural network includes the historical air pollutant concentration sequence, historical meteorological element sequence, and spatial relationship diagram of each monitoring station. The input in the actual prediction stage includes the air pollutant concentration sequence, the meteorological element sequence, and the spatial relationship diagram of the station collected in real time during the current and historical preset time periods of each monitoring station. The output is the predicted value of air pollutant concentration for each monitoring station in the future preset time window and the warning sign of exceeding the standard. The warning sign of exceeding the standard includes the anomaly sign of the corresponding monitoring station and the regional risk level.

[0013] A further improvement of this invention lies in: combining a three-dimensional wind field with a Lagrange diffusion model to trace the pollution source, including:

[0014] A three-dimensional wind field of the target area is constructed, coupled with a Lagrange diffusion model, and the pollution contribution rate of each pollution source to the pollution of the target area is calculated.

[0015] Based on the pollution contribution rate of each pollution source to the target area, a contribution list is obtained.

[0016] A further improvement of this invention lies in: based on the pollution source tracing results, using a genetic algorithm to plan the drone's patrol path, including:

[0017] Based on the pollution source tracing results, a set of patrol points and patrol routes are generated;

[0018] Based on genetic algorithms, the loitering path is encoded and an equal-length sequence is designed.

[0019] Generate the initial population , This represents the m-th patrol path containing N patrol points. Number of patrol routes;

[0020] Based on the drone loitering path planning objective, a fitness function is determined. The planning objective includes path length, flight energy consumption, and contribution coverage.

[0021] ;

[0022] in: For the fitness function, For the loitering path, For the loitering points obtained based on the contribution list mapping The contribution of pollution For drones at patrol points arrive Euclidean distance, Total energy consumption, These are the weighting coefficients. Chromosome length The patrol point number;

[0023] Update the loitering path based on the fitness function;

[0024] For the loitering path, a multi-objective optimization index is introduced based on the fitness function;

[0025] The loitering path is evaluated using a weighted multi-objective optimization index:

[0026] ;

[0027] in: For comprehensive evaluation function, This is a path length metric. Energy consumption indicators To contribute to coverage metrics, , , The weights are respectively for path length, energy consumption, and contribution coverage indicators;

[0028] The loitering paths are sorted based on the comprehensive evaluation function. The loitering path with the highest evaluation result is selected as the optimal execution path, and several loitering paths with the highest evaluation results are output as alternative paths.

[0029] A further improvement of this invention is that the expressions for the introduced multi-objective optimization indices are as follows:

[0030] Contribution Coverage Metrics:

[0031] ;

[0032] Path length metric:

[0033] ;

[0034] Energy consumption indicators:

[0035] .

[0036] A further improvement of this invention lies in: based on the pollution source tracing results and patrol routes, using an AI model to perform quantitative analysis of pollution events, including:

[0037] The system processes real-time monitoring data from monitoring stations, predicted air pollutant concentration data, pollution source tracing results, and air pollutant concentration data and video data transmitted back after drone patrols. This includes establishing correspondences based on time, spatial location, pollutant type, and data source, and ensuring that similar air pollutant concentration data use consistent units of measurement.

[0038] The emission intensity of pollution sources is adjusted based on nearby monitoring stations;

[0039] Substituting the corrected emission intensity of the pollution source into the Lagrange diffusion model, the atmospheric pollutant concentration and pollution contribution of the corrected monitoring station were calculated.

[0040] Based on the corrected atmospheric pollutant concentrations and pollution contribution rates of the monitoring stations, an AI model is used to comprehensively score pollution events and generate reports.

[0041] A further improvement of the present invention is that: the emission intensity of the pollution source is corrected based on the nearby monitoring stations, including:

[0042] Sort the pollution sources by spatial distance from nearest to farthest, and select the nearest one. Each patrol point is used as a monitoring station, resulting in a set:

[0043] ;

[0044] in: For patrol points, For monitoring site collection, The patrol point number;

[0045] Given a three-dimensional wind field and diffusion parameters, the atmospheric pollutant concentration at the monitoring station from the pollution source is denoted as:

[0046] ;

[0047] in: As a source of pollution Emission intensity;

[0048] Pollution sources were inverted at monitoring stations using the least squares method. Emission intensity:

[0049] ;

[0050] in: For the corrected pollution source Emission intensity, For patrol points The corresponding measured concentrations of air pollutants.

[0051] A further improvement of this invention lies in: utilizing an AI model to comprehensively score pollution events, wherein the expression for the comprehensive score is:

[0052] ;

[0053] in: For comprehensive scoring, Contribute a score to the pollution source. To score the pollution intensity, Score the spatial impact. For the evaluation of the cruise flight, , , , These are adjustable weighting coefficients.

[0054] The present invention provides a drone-based air pollution precision source tracing and control system based on embodied intelligence, comprising:

[0055] The real-time alarm and data acquisition module is used to predict the concentration of air pollutants and trigger real-time warnings when pollution exceeds the standard, and dispatch drones to carry out patrols;

[0056] The pollution source tracing module is used to trace pollution sources and generate a pollution source contribution list based on atmospheric pollutant concentration data from monitoring stations, combined with three-dimensional wind fields and Lagrange diffusion models.

[0057] The UAV route planning module is used to plan the UAV patrol flight path based on the pollution source contribution list and using a genetic algorithm, output an optimal execution path and several alternative paths, and control the UAV to complete the patrol flight and data collection according to the optimal execution path.

[0058] The AI ​​report generation and intelligent decision-making module is used to perform quantitative analysis of pollution events based on pollution source tracing results, patrol routes, and drone data, and generate pollution event reports using AI models.

[0059] The beneficial effects of this invention are as follows: This invention integrates real-time monitoring, predictive alerts, pollution source tracing, and dynamic drone patrols, achieving rapid closed-loop processing from data acquisition to decision execution, thus improving response speed and control efficiency. By combining high-resolution three-dimensional wind fields and Lagrange diffusion models, it achieves pollution source location with a spatial resolution of hundreds of meters and provides a quantified list of pollution source contributions, significantly improving the accuracy and timeliness of source tracing. This invention employs a fitness function and genetic algorithm to perform multi-objective optimization of the patrol path, achieving a comprehensive balance between path length, contribution coverage, and energy consumption. After the patrol, the data transmitted back by the drone is used to correct the emission intensity of the pollution source and perform a comprehensive score, improving the credibility of the report analysis and the speed of control decision-making. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the method flow in an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0062] This embodiment is a drone-based air pollution precision source tracing and control system based on embodied intelligence, including:

[0063] The real-time alarm and data acquisition module is used to predict the concentration of air pollutants and trigger real-time warnings when pollution exceeds the standard, and dispatch drones to carry out patrols;

[0064] The pollution source tracing module is used to trace pollution sources and generate a pollution source contribution list based on atmospheric pollutant concentration data from monitoring stations, combined with three-dimensional wind fields and Lagrange diffusion models.

[0065] The UAV route planning module is used to plan the UAV patrol flight path based on the pollution source contribution list and using a genetic algorithm, output an optimal execution path and several alternative paths, and control the UAV to complete the patrol flight and data collection according to the optimal execution path.

[0066] The AI ​​report generation and intelligent decision-making module is used to perform quantitative analysis of pollution events based on pollution source tracing results, patrol routes, and drone data, and generate pollution event reports using AI models.

[0067] The real-time alarm and data acquisition module constructs a lightweight prediction architecture based on a spatiotemporal graph neural network to perform short-term forecasts of air pollutant concentrations at distributed monitoring stations and trigger tiered warnings. The inputs to the spatiotemporal graph neural network training phase include historical air pollutant concentration sequences, historical meteorological element sequences, and a spatial relationship diagram for each monitoring station. The inputs to the actual prediction phase include real-time air pollutant concentration sequences, concurrent meteorological element sequences, and a spatial relationship diagram for each monitoring station within the current and historical preset time periods. The air pollutant concentration sequences characterize the pollutant changes at each monitoring station within the preset time window. The meteorological element sequences include at least one of wind speed, wind direction, temperature, and humidity. The spatial relationship diagram characterizes the spatial correlation between different monitoring stations. The tiered warning decision is based on an adaptive threshold strategy. For the target pollutant at the i-th monitoring station, let the historical baseline be Bᵢ. The historical baseline Bᵢ is determined based on the mean and standard deviation of the corresponding pollutant concentration at the i-th monitoring station within the historical preset time period, expressed as:

[0068] ;

[0069] in: Let i be the historical baseline of the i-th monitoring station. Let be the average pollutant concentration of the i-th monitoring station within a preset historical time period. denoted as the standard deviation of the corresponding pollutant concentration, and k is the adjustment coefficient.

[0070] The historical baseline is corrected based on real-time meteorological data to obtain the adaptive warning threshold for the i-th monitoring station. The expression is:

[0071] ;

[0072] in: Let i be the adaptive early warning threshold for the i-th monitoring station. is the real-time meteorological correction factor corresponding to the i-th monitoring station, which is determined based on the difference between the current meteorological elements and the historical meteorological elements.

[0073] When the predicted concentration of air pollutants at the i-th monitoring station in the future preset time window is greater than the corresponding adaptive warning threshold When the concentration of pollutant exceeds the preset concentration limit, an anomaly flag is generated for the i-th monitoring station. .in:

[0074] =1 indicates that the predicted concentration of air pollutants is greater than the corresponding adaptive warning threshold. If the predicted concentration of air pollutants is greater than the preset concentration limit of the corresponding pollutant, it indicates that the i-th monitoring station is abnormal, that is, the pollution exceeds the standard.

[0075] =0 indicates that the predicted concentration of air pollutants is less than or equal to the corresponding adaptive warning threshold. If the predicted concentration of air pollutants is less than or equal to the preset concentration limit of the corresponding pollutant, it indicates that no abnormality has occurred at the i-th monitoring station.

[0076] The regional risk level is determined based on the proportion of monitoring stations generating anomaly markers to the total number of monitoring stations in the target area. Let N be the total number of monitoring stations in the target area, and the regional anomaly ratio be... The expression is:

[0077] ;

[0078] when When the risk level is less than the first proportional threshold, the area is classified as low risk; when... When the risk level of an area is greater than or equal to the first proportional threshold and less than the second proportional threshold, the area risk level is determined to be medium risk; when When the risk level is greater than or equal to the second proportional threshold, the area is classified as high-risk. The warning indicator includes the anomaly marker at the corresponding monitoring station and the area's risk level. Through this process, timely identification and early warning of abnormal air quality in the region are achieved, providing data support for warning triggering, drone patrol scheduling, pollution source tracing, and emergency response.

[0079] For the precise identification and quantitative management of pollution sources in the target area, the pollution source tracing module performs pollution source tracing based on high-resolution three-dimensional wind fields and Lagrange diffusion models, including:

[0080] The three-dimensional wind field of the target area is constructed. In this embodiment, a three-dimensional small-scale wind field is constructed with a resolution of 100 meters to 1 kilometer. It can accurately simulate the impact of terrain undulations, building obstruction and local airflow disturbance on pollutant transport, and significantly improve the accuracy of regional pollution paths and diffusion patterns.

[0081] The coupled Lagrange diffusion model overcomes the limitations of the traditional steady-state Gaussian model, enabling simulation of the dynamic migration and transport of pollutants under unsteady conditions. Using the Lagrange diffusion model, key high-value locations of pollutants can be identified within seconds, and their sources can be traced, including industrial discharge outlets, major traffic arteries, and other potential pollution sources, providing real-time source tracing capabilities for regional pollution events.

[0082] The spatial distribution of air pollutant concentrations is represented by the Lagrange diffusion model as follows:

[0083] (1);

[0084] in: To be located in the spatial position of the target area, i.e., grid point The location is due to the pollution source. The concentration of atmospheric pollutants contributed, i.e., the single-source concentration, is used for calculating the pollution contribution rate, generating a pollution source contribution inventory, and optimizing the drone patrol path. As a source of pollution Emission intensity, As a source of pollution coordinates , , These are diffusion parameters that are dynamically adjusted based on wind speed, turbulence intensity, and terrain.

[0085] Combined with expression (1), the contribution rate of each pollution source to the pollution of the target area is calculated, and the contribution list of pollution sources is generated, providing a quantitative basis for pollution source control.

[0086] The system in this embodiment generates a contribution list using a quantitative calculation method for the pollution contribution of each pollution source to the target area. The expression is:

[0087] (2);

[0088] in: The cumulative concentration of all pollution sources at the same grid point. For the set of grid points in the target area, As a source of pollution Contribution rate of pollution to the target area. Expression (2) quantifies the contribution rate of each pollution source in the entire target area by integrating the target area, providing a clear numerical basis for pollution source control.

[0089] This embodiment achieves refined location and quantitative assessment of pollution sources through the aforementioned technical means. This not only improves the response speed and accuracy of pollution incidents but also provides reliable scientific data support for pollution emission management, emergency dispatch, and regulation formulation. It enables source identification with a spatial resolution of hundreds of meters for pollution sources such as industrial discharge outlets and transportation roads, and provides a contribution list within seconds. Compared to Gaussian models or traditional large-scale models, this significantly improves the accuracy and timeliness of source tracing.

[0090] After completing the pollution source tracing analysis, efficient patrol routes are automatically generated based on pollution sources with high contribution rates, enabling unmanned, intelligent, and rapid-response patrol missions. The following key factors are considered when generating the patrol route:

[0091] Pollution contribution rate: Prioritize patrols of pollution sources that contribute significantly to regional pollution;

[0092] Flight resource constraints: drone battery power, flight distance, and total energy consumption;

[0093] Source environmental factors: wind speed and wind direction affect the contribution list through three-dimensional wind field and Lagrange diffusion model, and further affect the contribution coverage evaluation of the patrol point;

[0094] Closed-loop cruise constraint: Ensure that the drone's starting point and ending point are consistent to achieve a complete closed loop.

[0095] When the real-time alarm and data acquisition module triggers a real-time warning, the drone dynamically plans its flight path, thereby improving the timeliness of tracing and evidence collection tasks, reducing the overall response time from days to minutes, and enabling rapid decision-making and precise control.

[0096] Encoding and designing equal-length sequences for the loitering path includes:

[0097] Encoding scheme: Each drone's patrol path is represented as a chromosome. Each gene The location of a corresponding monitoring station, i.e., a patrol point, is denoted as L in this embodiment, and L=N.

[0098] Sequence length consistency handling: To meet the requirement of consistent chromosome length in genetic algorithms, a method of padding duplicate points is used. For example: There are S pollution sources. The number of randomly selectable patrol points is set to 2S-1, and the chromosome length is fixed at N. This ensures that all chromosomes have the same length while allowing the planned route to avoid traversing all pollution sources. All patrol points with numbers greater than N are occupancy points, such as chromosomes. The flight path indicates that it only includes pollution source 1.

[0099] Generate the initial population , This represents the m-th patrol path containing N patrol points. Number of patrol routes;

[0100] The fitness function is designed, including the UAV path planning objectives, which include: shortest path length, maximum coverage contribution, and minimum energy consumption. The designed fitness function is:

[0101] (3);

[0102] in: For the fitness function, For the loitering path, For the patrol points obtained by mapping based on the pollution source contribution list The pollution contribution of the patrol point is used to characterize the patrol point. The degree of coverage of high-contribution pollution areas, For drones at patrol points arrive Euclidean distance, Total energy consumption, These are the weighting coefficients. This represents chromosome length.

[0103] The energy consumption calculation expression is:

[0104] (4);

[0105] in: For flight speed, The total mass of the drone, For flight energy consumption, a dynamic energy consumption model is used for estimation.

[0106] Perform genetic manipulation:

[0107] (5);

[0108] in: Let be the probability that the m-th patrol path is selected. For the first Each loitering path The fitness function value, For the first Each loitering path The fitness function value. In this embodiment, the parent is selected through roulette or tournament.

[0109] cross:

[0110] Using the single-point crossover method, the parent chromosome is split at a specific point, and fragments are exchanged to generate offspring, thus creating a new path sequence:

[0111] (6);

[0112] in: For the new path sequence, , , For genes;

[0113] Mutations:

[0114] Randomly swap the positions of two genes in the new path sequence, or randomly replace them with repeating sites:

[0115] (7);

[0116] in: , For genes. Through mutation, population diversity is increased, avoiding getting trapped in local optima.

[0117] Ultimately, the most fit elite individuals are retained, while offspring chromosomes are introduced to form a new generation.

[0118] In this embodiment, the UAV route planning is based on the pollution source tracing results and contribution list, and the optimal path is selected before the patrol mission is executed. The atmospheric pollutant concentration data, meteorological elements, self-state data and video data collected during the patrol are transmitted back after the patrol and used for subsequent pollution event analysis, source strength correction, comprehensive scoring and report generation.

[0119] Based on the fitness function, a multi-objective optimization index is introduced. In the multi-objective optimization stage, the genetic algorithm uses a comprehensive evaluation function. As a basis for fitness, path iterative optimization is performed, where multi-objective optimization metrics include:

[0120] Path length metric Energy consumption indicators :

[0121] (8);

[0122] (9);

[0123] Contribution Coverage Metrics :

[0124] (10).

[0125] Taking into account multiple objective optimization indicators, this embodiment uses a weighted multi-objective optimization approach to evaluate the loitering path:

[0126] (11);

[0127] in: , , These are the weights for the path length indicator, energy consumption indicator, and contribution coverage indicator, respectively. , , Adjustable, with constraints including upper limits on battery power, flight time, no-fly zones, and closed-loop constraints on takeoff and landing points. Optimal or suboptimal loitering schemes that satisfy multiple objectives are obtained through methods such as genetic algorithms.

[0128] By using genetic algorithms for selection, crossover, mutation, and elite retention, the loitering paths are iteratively optimized to obtain a path ranking result that meets the comprehensive requirements of path length, energy consumption, and contribution coverage.

[0129] After iteration, the various loitering paths are ranked based on a comprehensive evaluation function. The loitering path with the highest evaluation result is selected as the optimal execution path, and several other loitering paths with high evaluation results are output as candidate paths. The UAV prioritizes completing the loitering along the optimal execution path. When flight conditions, no-fly zones, or battery constraints prevent the optimal execution path from being executed, a path that meets the constraints is selected from the candidate loitering paths for execution.

[0130] After completing its patrol flight along the selected path and returning to its take-off and landing point, the drone transmits back data on atmospheric pollutant concentrations, meteorological elements, its own status data, and video data. The own status data includes at least one of the following: flight speed, remaining battery power, position, and heading.

[0131] The AI ​​report generation and intelligent decision-making module, based on an AI model agent, addresses the high-precision and timeliness requirements of regional air pollution management by enabling automated analysis and report generation for pollution events. Compared to existing technologies that rely on manual compilation or simple rule-based report generation, this module significantly improves report generation speed, analysis accuracy, and control efficiency through multi-source data fusion and intelligent judgment, achieving fully automated processing within minutes. This embodiment processes multi-source data, including real-time monitoring data from monitoring stations, predicted air pollutant concentration data, pollution source tracing results, and air pollutant concentration data and video data collected by drones. This includes establishing correspondences based on time, spatial location, pollutant type, and data source, ensuring that similar air pollutant concentration data use consistent units of measurement, and extracting structured feature information from video data that can be used for pollution event analysis.

[0132] To ensure consistency between the pollution source tracing results and the atmospheric pollutant concentration data measured by drone patrols, each pollution source... Introducing local nearest neighbor inversion correction operations, including:

[0133] 1. Select nearby monitoring stations, including:

[0134] In the target area grid, based on pollution sources Sort by spatial distance from nearest to farthest, and select the nearest one. This embodiment takes [number] monitoring stations. The set is obtained as follows:

[0135] (12);

[0136] in: For patrol points, For monitoring site collection, This is the serial number of the patrol point.

[0137] 2. Use the Lagrange diffusion model to infer the pollution source. Emission intensity, including:

[0138] Given a three-dimensional wind field and diffusion parameters, the pollution source The concentrations of air pollutants at the monitoring stations are recorded as follows:

[0139] (13);

[0140] For expression (1) at the loitering point The values ​​taken at the corresponding monitoring stations.

[0141] The pollution source was retrieved from the five nearest monitoring stations using the least squares method. Emission intensity:

[0142] (14);

[0143] Inversion That is, the corrected pollution source The emission intensity, with the original Lagrange diffusion model structure unchanged, only the parameters are adjusted. Perform numerical updates, where For patrol points The corresponding measured concentrations of air pollutants.

[0144] 3. Based on the corrected pollution sources emission intensity Update pollution contribution and pollution contribution rate, including:

[0145] The corrected pollution source emission intensity Substituting into the Lagrange diffusion model, calculate the corrected single-source concentration at the monitoring station:

[0146] (15); in: For the revised loitering point Cumulative concentrations from all pollution sources For the corrected pollution source The concentration of air pollutants at the monitoring stations.

[0147] Combining expression (2), the spatial integral is discretized into a summation over the grid set of the target region, yielding the corrected pollution contribution rate:

[0148] (16);

[0149] in: For the corrected pollution source Contribution rate to pollution in the target area.

[0150] In the fitness function of UAV loitering path planning, i.e., in expression (3), the loitering points in the fitness function are... pollution contribution Replace with the revised loitering point pollution contribution This allows path optimization to directly reflect the pollution source. Pollution distribution after emission intensity correction.

[0151] Complete the pollution source After adjusting for emission intensity and pollution contribution, the AI ​​model performs quantitative assessment and comprehensive evaluation of pollution events. The logical representation of the comprehensive score is as follows:

[0152] (17);

[0153] in: For comprehensive scoring, Contribute a score to the pollution source. To score the pollution intensity, Score the spatial impact. For the evaluation of the cruise flight, , , , These are adjustable weighting coefficients.

[0154] The pollution source contribution score is determined based on the modified set of pollution contribution rates, expressed as:

[0155] (18);

[0156] in: For the number of pollution sources, For the first The contribution score weight of each pollution source.

[0157] The pollution intensity score is determined based on the corrected concentration of air pollutants and the preset concentration limits for the corresponding pollutants, expressed as:

[0158] (19);

[0159] in: For the preset concentration limits of the corresponding pollutants, The number of monitoring stations corresponds to the number of patrol points.

[0160] The expression for spatial impact score is:

[0161] (20);

[0162] in: This refers to the number of monitoring stations where the corrected atmospheric pollutant concentration exceeds the preset concentration limit for the corresponding pollutant.

[0163] The patrol flight verification score is determined based on the consistency between the atmospheric pollutant concentration data collected by the UAV patrol flight and the concentration calculated by the Lagrange diffusion model, as well as the pollution characteristics in the video data. The expression is:

[0164] (twenty one);

[0165] in: A consistency score is given between the concentration measured by the UAV and the concentration calculated by the Lagrange diffusion model. The pollution feature score is derived from drone video data. These are adjustable weighting coefficients.

[0166] This embodiment uses the method of "selecting nearby monitoring stations and using the Lagrange diffusion model to infer the pollution source" to... Emission intensity - with modified , This closed-loop mechanism of "driving drone patrol path planning and comprehensive scoring" is used to analyze pollution sources. The emission intensity and pollution contribution rate are finely revised to make the pollution source analysis and spatial impact assessment in the report closer to the actual measured facts, providing a consistent and quantifiable data basis for the automatically generated source tracing analysis report and control recommendations.

[0167] Based on the actual monitoring situation in a certain area, the generated source tracing analysis report includes the following:

[0168] Source tracing report

[0169] Air quality emergency source tracing report

[0170] Air quality early warning analysis

[0171] Time: 14:00 on June 30, 2025;

[0172] Location: A monitoring station in Dongcheng District;

[0173] Emergency warning type: PM2.5 exceedance alarm;

[0174] Warning level: High;

[0175] Warning: PM2.5 concentration exceeds 150 μg / m³.

[0176] This warning indicates that the monitoring stations experienced severe pollution during a specific period, and the increased PM2.5 concentration may pose a risk to public health, requiring immediate emergency measures for source tracing and control.

[0177] Source tracing analysis

[0178] Source tracing period: July 2, 2025, 02:07:10.435Z to 02:07:16.140Z;

[0179] Weather conditions:

[0180] Temperature: 25°C;

[0181] Humidity: 65%;

[0182] Wind speed: 3 m / s;

[0183] Wind direction: Southeast;

[0184] Weather conditions: Sunny.

[0185] Source tracing results: Through source tracing analysis, the increase in PM2.5 mainly comes from industrial emissions and traffic exhaust. Industrial emissions are concentrated in the industrial area of ​​Dongcheng District, while traffic exhaust is mainly distributed along the city's main roads.

[0186] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0187] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for precise source tracing and control of air pollution from drones based on embodied intelligence, characterized in that, include: Predict air pollutant concentrations and trigger real-time warnings when pollution levels exceed standards, and dispatch drones to conduct patrols; Based on atmospheric pollutant concentration data from monitoring stations, combined with three-dimensional wind fields and Lagrange diffusion models, the pollution sources are traced. Based on the pollution source tracing results, a genetic algorithm is used to plan the drone's patrol path, outputting an optimal execution path and several alternative patrol paths; Based on pollution source tracing results, drone data transmission and patrol routes, AI models are used to conduct quantitative analysis of pollution events and generate pollution event reports. Based on the pollution source tracing results, a genetic algorithm is used to plan the drone's patrol path, including: Based on the pollution source tracing results, a set of patrol points and patrol routes are generated; Based on genetic algorithms, the loitering path is encoded and an equal-length sequence is designed. Generate the initial population , This represents the m-th patrol path containing N patrol points. This refers to the number of patrol routes; Based on the drone loitering path planning objective, a fitness function is determined. The planning objective includes path length, flight energy consumption, and contribution coverage. ； in: For the fitness function, For the loitering path, For the loitering points obtained based on the contribution list mapping The contribution of pollution For drones at patrol points arrive Euclidean distance, Total energy consumption, These are the weighting coefficients. Chromosome length The patrol point number; Update the loitering path based on the fitness function; For the loitering path, a multi-objective optimization index is introduced based on the fitness function; The loitering path is evaluated using a weighted multi-objective optimization index: ； in: For comprehensive evaluation function, This is a path length metric. As an energy consumption indicator, To contribute to coverage metrics, , , The weights are respectively for path length, energy consumption, and contribution coverage indicators; The loitering paths are sorted based on the comprehensive evaluation function. The loitering path with the highest evaluation result is selected as the optimal execution path, and several loitering paths with the highest evaluation results are output as alternative paths. Based on pollution source tracing results and patrol routes, AI models are used to conduct quantitative analysis of pollution events, including: The system processes real-time monitoring data from monitoring stations, predicted air pollutant concentration data, pollution source tracing results, and air pollutant concentration data and video data transmitted back after drone patrols. This includes establishing correspondences based on time, spatial location, pollutant type, and data source, and ensuring that similar air pollutant concentration data use consistent units of measurement. The emission intensity of pollution sources is adjusted based on nearby monitoring stations; Substituting the corrected emission intensity of the pollution source into the Lagrange diffusion model, the atmospheric pollutant concentration and pollution contribution of the corrected monitoring station were calculated. Based on the corrected atmospheric pollutant concentrations and pollution contribution rates of the monitoring stations, an AI model is used to comprehensively score pollution events and generate reports. Correcting the emission intensity of pollution sources based on nearby monitoring stations includes: Sort the pollution sources by spatial distance from nearest to farthest, and select the nearest one. Each patrol point is used as a monitoring station, resulting in a set: ； in: For patrol points, For monitoring site collection, The patrol point number; Given a three-dimensional wind field and diffusion parameters, the atmospheric pollutant concentration at the monitoring station from the pollution source is denoted as: ； in: As a source of pollution Emission intensity; Pollution sources were inverted at monitoring stations using the least squares method. Emission intensity: ； in: For the corrected pollution source emission intensity, For patrol points The corresponding measured concentrations of air pollutants.

2. The method for precise source tracing and control of air pollution from unmanned aerial vehicles based on embodied intelligence as described in claim 1, characterized in that, The spatiotemporal graph neural network is used to predict the concentration of air pollutants. The input of the spatiotemporal graph neural network during the training phase includes the historical air pollutant concentration sequence, historical meteorological element sequence, and spatial relationship diagram of each monitoring station. The input during the actual prediction phase includes the air pollutant concentration sequence collected in real time during the current and historical preset time periods of each monitoring station, the meteorological element sequence of the same period, and the spatial relationship diagram of the station. The output is the predicted value of air pollutant concentration for each monitoring station in the future preset time window and the warning sign of exceeding the standard. The warning sign of exceeding the standard includes the anomaly sign of the corresponding monitoring station and the regional risk level.

3. The method for precise source tracing and control of air pollution from unmanned aerial vehicles based on embodied intelligence as described in claim 1, characterized in that, By combining three-dimensional wind fields with Lagrange diffusion models, the pollution sources can be traced, including: A three-dimensional wind field of the target area is constructed, coupled with a Lagrange diffusion model, and the pollution contribution rate of each pollution source to the pollution of the target area is calculated. Based on the pollution contribution rate of each pollution source to the target area, a contribution list is obtained.

4. The method for precise source tracing and control of air pollution from unmanned aerial vehicles based on embodied intelligence as described in claim 1, characterized in that, The expressions for the introduced multi-objective optimization indices are as follows: Contribution Coverage Metrics: ； Path length metric: ； Energy consumption indicators: 。 5. The method for precise source tracing and control of air pollution from unmanned aerial vehicles based on embodied intelligence according to claim 1, characterized in that, An AI model is used to comprehensively score pollution events, where the expression for the comprehensive score is: ； in: For comprehensive scoring, Contribute a score to the pollution source. To score the pollution intensity, Score the spatial impact. For the evaluation of the cruise flight, , , , These are adjustable weighting coefficients.

6. A UAV-based precise source tracing and control system for air pollution based on embodied intelligence, using the method described in any one of claims 1 to 5, characterized in that: include: The real-time alarm and data acquisition module is used to predict the concentration of air pollutants and trigger real-time warnings when pollution exceeds the standard, and dispatch drones to carry out patrols; The pollution source tracing module is used to trace pollution sources and generate a pollution source contribution list based on atmospheric pollutant concentration data from monitoring stations, combined with three-dimensional wind fields and Lagrange diffusion models. The UAV route planning module is used to plan the UAV patrol flight path based on the pollution source contribution list and using a genetic algorithm, output an optimal execution path and several alternative paths, and control the UAV to complete the patrol flight and data collection according to the optimal execution path. The AI ​​report generation and intelligent decision-making module is used to perform quantitative analysis of pollution events based on pollution source tracing results, patrol routes, and drone data, and generate pollution event reports using AI models.

Images