Atmospheric pollution monitoring method based on unmanned aerial vehicle remote sensing and machine learning

Abstract

The invention relates to an air pollution monitoring method based on unmanned aerial vehicle remote sensing and machine learning. According to the method, atmospheric pollutant concentration is monitored in real time by acquiring fixed monitoring stations and satellite remote sensing data, a pollution abnormal signal is generated when the concentration exceeds a preset threshold value, an unmanned aerial vehicle group is scheduled according to the signal to perform three-dimensional pollution data acquisition according to a preset path, and initial pollution distribution data is obtained; and dynamically adjusting a monitoring path and sensor configuration parameters of the unmanned aerial vehicle group by using a reinforcement learning algorithm, collecting real-time monitoring data, performing pollution traceability calculation in combination with a space-time diagram neural network model, generating a pollution source positioning result, then performing a pollution regulation and control instruction, and executing feedback control. By adopting the method, the accuracy of pollution source positioning and the timeliness of regulation and control can be effectively improved, and a scientific basis and technical support are provided for air pollution control.

Description

Technical Field

[0001] The present invention belongs to the field of atmospheric monitoring, and in particular relates to an atmospheric pollution monitoring method based on unmanned aerial vehicle (UAV) remote sensing and machine learning. Background Art

[0002] Traditional monitoring methods, such as ground-based monitoring stations and satellite remote sensing, are no longer able to meet the demands of refined monitoring and source tracing. In recent years, the rapid development of drone technology, with its advantages such as high flexibility, high spatial and temporal resolution, and low cost, has gradually become an important supplementary means of air pollution monitoring. Drones can be equipped with a variety of miniaturized, high-precision sensors, enabling real-time monitoring of air pollutants such as PM2.5 and ozone. Combined with machine learning algorithms, drone monitoring data can more efficiently locate and trace pollution sources.

[0003] Traditionally, air pollution monitoring relies primarily on a network of fixed ground-based monitoring stations, combined with remote sensing data from meteorological and environmental satellites. This data is used to collect pollutant concentration data and create regional pollution distribution maps. To trace pollution sources, numerical simulation methods based on Gaussian diffusion models or Lagrangian particle models are often used, combined with meteorological data for reverse inference calculations.

[0004] However, current monitoring methods and traditional approaches suffer from the following issues: the limited density of fixed monitoring stations makes it difficult to monitor pollution with high temporal and spatial resolution, particularly in capturing sudden pollution events; satellite remote sensing data, limited by revisit cycles and spatial resolution, struggles to reflect dynamic changes in pollution; and traditional diffusion models are poorly adaptable to complex terrain and meteorological conditions, resulting in limited accuracy in locating pollution sources. These issues severely hinder the effectiveness of precise air pollution prevention and control. Summary of the Invention

[0005] Based on this, it is necessary to provide an atmospheric pollution monitoring method based on UAV remote sensing and machine learning that can achieve high-precision positioning and control in response to the above technical problems.

[0006] In the first aspect, this application provides an air pollution monitoring method based on drone remote sensing and machine learning, comprising:

[0007] Obtain data from fixed monitoring stations and satellite remote sensing to monitor the concentration of atmospheric pollutants in real time. When the concentration exceeds the preset threshold, a pollution anomaly signal result is generated.

[0008] Based on the abnormal pollution signal results, the drone group is dispatched to collect three-dimensional pollution data along the preset path to obtain the initial pollution distribution data;

[0009] Based on the initial pollution distribution data, the monitoring path and sensor configuration parameters of the drone swarm are dynamically adjusted through the reinforcement learning algorithm to obtain the adaptive monitoring network configuration parameters;

[0010] Based on the adaptive monitoring network configuration parameters, the drone swarm collects real-time monitoring data, performs pollution source tracing calculations through a spatiotemporal graph neural network model, and generates pollution source location results.

[0011] Based on the pollution source positioning results, pollution control instructions are generated and feedback control is performed.

[0012] In one embodiment, based on the pollution anomaly signal results, a drone swarm is dispatched to collect three-dimensional pollution data along a preset path to obtain initial pollution distribution data, including:

[0013] Obtain the pollution area range data corresponding to the pollution abnormal signal results and construct a three-dimensional monitoring grid. The three-dimensional monitoring grid is distributed in a rectangular shape on the horizontal plane and is a fixed height layer from the ground in the vertical direction;

[0014] Based on the pollutant type characteristics in the three-dimensional monitoring grid, the corresponding sensor combination is selected to configure the drone monitoring equipment for data collection to obtain the initial pollution distribution data.

[0015] In one embodiment, based on the pollutant type characteristics in the three-dimensional monitoring grid, a corresponding sensor combination is selected to configure the drone monitoring device. After obtaining the initial pollution distribution data, the following steps are further included:

[0016] Based on the initial pollution distribution data, the spatial concentration change rate of pollutants is calculated using the following gradient analysis algorithm:

[0017]

[0018] in, is the spatial concentration change rate of pollutants, is the east-west concentration gradient of the pollutant on the horizontal plane, is the north-south concentration gradient of the pollutant on the horizontal plane, is the vertical concentration gradient, is the time change rate, α, β, γ are the spatial weight coefficients, and λ is the time attenuation factor.

[0019] In one embodiment, after calculating the spatial concentration change rate of pollutants using a gradient analysis algorithm based on the initial pollution distribution data, the method further includes:

[0020] Based on the spatial concentration change rate of pollutants, a path optimization evaluation function is constructed. The evaluation function includes three optimization objectives: pollution gradient, information entropy, and flight energy consumption.

[0021] The path optimization evaluation function is iteratively optimized through the reinforcement learning algorithm to generate a set of optimized monitoring paths for the drone swarm;

[0022] The changes in pollution gradient indicators are monitored in real time based on the optimized monitoring path set of the drone swarm. When the pollution gradient indicator is detected to exceed the preset dynamic threshold, the result of the number of drones that need to be increased is generated based on the gradient exceeding the standard.

[0023] In one embodiment, the real-time monitoring data collected by the drone swarm is controlled according to the configuration parameters of the adaptive monitoring network, and pollution source tracing calculations are performed through a spatiotemporal graph neural network model to generate pollution source location results, including:

[0024] The real-time monitoring data collected by the drone swarm is controlled by the adaptive monitoring network configuration parameters to construct a spatiotemporal correlation graph structure. The spatiotemporal coordinates of each monitoring point are used as graph nodes, and the edge weights between nodes are obtained using a distance exponential function.

[0025] Through the spatiotemporal graph neural network, multi-layer feature aggregation is performed on the spatiotemporal correlation graph to generate a pollution diffusion feature matrix;

[0026] The pollution diffusion characteristic matrix is integrated with the real-time meteorological data, and the pollution source probability distribution map is obtained through the reverse diffusion model. The calculation formula of the reverse diffusion model is:

[0027] P=f(G,W,M)

[0028] Among them, P is the pollution probability distribution, G is the spatiotemporal correlation graph, W is the edge weight matrix, and M is the meteorological parameter matrix.

[0029] In one embodiment, the pollution diffusion characteristic matrix is integrated with real-time meteorological data, and a pollution source probability distribution map is obtained by calculating the reverse diffusion model, including:

[0030] Generate a spatial diffusion coefficient matrix based on the node distribution characteristics in the spatiotemporal correlation graph structure;

[0031] Obtain meteorological correction factors based on wind speed, wind direction and atmospheric stability parameters in real-time meteorological data;

[0032] Based on the spatial diffusion coefficient matrix and meteorological correction factors, the contribution probability of each potential pollution source is generated through the reverse diffusion model;

[0033] The contribution probability of each potential pollution source is normalized to generate a standardized pollution source probability distribution map.

[0034] In one embodiment, generating pollution control instructions and performing feedback control based on pollution source location results includes:

[0035] Conduct credibility assessment based on pollution source location results, and generate corresponding pollution control instructions when the assessment results meet the preset credibility conditions;

[0036] Send pollution control instructions to the environmental monitoring system to perform pollution control operations;

[0037] Collect environmental monitoring data after performing pollution control operations;

[0038] The parameters of the spatiotemporal graph neural network model are updated based on environmental monitoring data to obtain subsequent pollution source tracing calculation optimization results.

[0039] Secondly, this application also provides an air pollution monitoring system based on drone remote sensing and machine learning, including:

[0040] The data acquisition module is used to obtain fixed monitoring station and satellite remote sensing data to monitor the concentration of atmospheric pollutants in real time. When the concentration exceeds the preset threshold, it generates a pollution anomaly signal result;

[0041] The drone dispatching module is used to dispatch drone groups to collect three-dimensional pollution data along a preset path according to the pollution anomaly signal results to obtain initial pollution distribution data;

[0042] The adaptive adjustment module is used to dynamically adjust the monitoring path and sensor configuration parameters of the drone swarm based on the initial pollution distribution data through the reinforcement learning algorithm to obtain the adaptive monitoring network results;

[0043] The pollution source tracing module is used to calculate the pollution source based on the real-time monitoring data collected by the adaptive monitoring network through the spatiotemporal graph neural network model to generate pollution source location results;

[0044] The control instruction generation and execution module is used to generate pollution control instructions and perform feedback control based on the pollution source positioning results.

[0045] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements any step of the above embodiment when executing the computer program.

[0046] In a fourth aspect, the present application further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, any step of the above embodiment is implemented.

[0047] The above-mentioned atmospheric pollution monitoring method based on UAV remote sensing and machine learning can conduct high-precision real-time monitoring of atmospheric pollutant concentrations by integrating remote sensing data and realize adaptive monitoring network parameter configuration and source tracing positioning through reinforcement learning algorithm. It can effectively improve the accuracy of pollution source positioning and the timeliness of regulation, and provide scientific basis and technical support for atmospheric pollution control. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 Schematic diagram of the implementation environment of the air pollution monitoring method based on UAV remote sensing and machine learning of the present invention;

[0050] Figure 2 This is a flow chart of the air pollution monitoring method based on UAV remote sensing and machine learning of the present invention;

[0051] Figure 3 This is a structural block diagram of the air pollution monitoring system based on drone remote sensing and machine learning of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0053] The atmospheric pollution monitoring method based on UAV remote sensing and machine learning provided in the embodiment of the present application can be applied to Figure 1 In the implementation environment shown, the monitoring terminal 101 communicates with the server 102 via a network. The data storage system can store data that the server 102 needs to process. The data storage system can be integrated with the server 102 or placed on a cloud or other network server. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, and tablet computers. The server 102 can be implemented as a standalone server or a server cluster consisting of multiple servers.

[0054] In combination with the above implementation environment, the application scenarios of the embodiments of the present application are explained.

[0055] The embodiments of this application are applicable to complex urban air pollution monitoring and control scenarios, especially emergency monitoring and disposal scenarios for sudden air pollution incidents in industrial parks. Environmental monitoring personnel use a swarm of deployed drones to monitor the concentration of atmospheric pollutants in the park in real time. When abnormal emissions are detected, the monitoring terminal automatically dispatches the drone swarm to conduct precise three-dimensional pollution scanning, and quickly locates the pollution source based on intelligent algorithms, guiding law enforcement personnel to quickly go to the target enterprise for verification and disposal, thus achieving rapid closed-loop management from pollution discovery to precise tracing, shortening the traditional pollution tracing process that takes several hours to a dozen minutes, significantly improving the environmental supervision efficiency of industrial parks.

[0056] This is only an example and does not limit the specific application scenario.

[0057] In an exemplary embodiment, Figure 2 As shown in the figure, a method for monitoring atmospheric pollution based on UAV remote sensing and machine learning is provided. Figure 1 Taking the monitoring terminal in FIG. 1 as an example, the method includes the following steps S111 to S115:

[0058] S111, obtain fixed monitoring station and satellite remote sensing data, monitor the concentration of atmospheric pollutants in real time, and generate a pollution abnormality signal result when the concentration is detected to exceed a preset threshold.

[0059] Optionally, the monitoring terminal receives real-time PM2.5 data (sampling frequency 1 time / minute) from multiple fixed monitoring stations in the urban area and NO2 remote sensing data (spatial resolution 5.5km×3.5km) from the Sentinel-5P satellite. When the monitoring terminal detects that the PM2.5 concentration at a monitoring station downwind of an industrial park exceeds the preset threshold of 150μg / m3 for three consecutive times, 3 When the pollution is detected, a pollution anomaly signal including time, location and exceeding standard factor is automatically generated.

[0060] S112: Based on the abnormal pollution signal results, the drone group is dispatched to collect three-dimensional pollution data along a preset path to obtain initial pollution distribution data.

[0061] Specifically, when a monitoring terminal generates an abnormal pollution signal, a swarm of drones on standby is automatically activated to conduct three-dimensional monitoring of pollutant concentrations in the target area according to a pre-programmed three-dimensional spiral scanning path. The drones' multi-parameter sensors simultaneously collect pollution data at different altitudes, and edge computing nodes fuse this data in real time to generate an initial pollution distribution map containing spatial distribution characteristics. A drone swarm is a collaborative monitoring system consisting of 3-6 multi-rotor drones with autonomous flight capabilities. The three-dimensional pollution data is a structured dataset containing longitude, latitude, and altitude coordinates and corresponding pollutant concentrations. The initial pollution distribution data is the spatial distribution matrix of pollutants obtained from the first round of scanning.

[0062] S113, based on the initial pollution distribution data, the monitoring path and sensor configuration parameters of the drone swarm are dynamically adjusted through a reinforcement learning algorithm to obtain the adaptive monitoring network configuration parameters.

[0063] For example, the monitoring terminal uses a deep reinforcement learning model to analyze pollution diffusion trends based on initially collected pollution distribution data. This dynamically optimizes the flight paths and sensor operating modes of the drone swarm, and through real-time evaluation of monitoring efficiency and energy consumption, it develops an adaptive network configuration solution that includes optimal path planning and sensor tuning parameters. These adaptive network configuration parameters include a set of dynamically adjustable operating parameters, such as the drone's three-dimensional coordinate sequence, sensor sampling frequency, and detection sensitivity.

[0064] S114, according to the adaptive monitoring network configuration parameters, the real-time monitoring data collected by the drone group is controlled, and the pollution source tracing calculation is performed through the spatiotemporal graph neural network model to generate the pollution source positioning result.

[0065] Optionally, based on dynamically optimized drone network configuration parameters, the fleet is commanded to perform high-precision environmental sampling. The acquired spatiotemporal sequence monitoring data is fed into a pre-trained graph neural network model. By integrating meteorological fields with pollution diffusion characteristics, a reverse deduction is performed to generate a positioning result containing the coordinates of the pollution source and a probability weight. The pollution source positioning result is a structured output including the coordinates of the suspected source (longitude, latitude, and altitude), a probability confidence level, and a contribution assessment (a quantitative indicator of the impact on the monitoring point).

[0066] S115: Generate pollution control instructions and perform feedback control based on the pollution source positioning result.

[0067] Specifically, a credibility assessment is generated based on the pollution source location results, and hierarchical control instructions are automatically generated and pushed to relevant entities. Simultaneously, drone verification flights are launched to conduct closed-loop monitoring of control effectiveness, forming a complete "locate-dispose-verify" control loop. Pollution control instructions are structured control commands that include the control target (enterprise / facility ID), control measures (production suspension / limitation / maintenance, etc.), and execution time (immediate / within 1 hour, etc.). Feedback control is a dynamic adjustment system that includes tracking the execution status of instructions and analyzing pollutant concentration trends.

[0068] In the above-mentioned atmospheric pollution monitoring method based on drone remote sensing and machine learning, through the intelligent collaborative drone swarm dynamic networking and spatiotemporal graph neural network tracing algorithm, the full process automation from pollution anomaly detection, precise tracing to closed-loop control is realized, which significantly improves the spatiotemporal resolution of pollution monitoring and the efficiency of emergency response.

[0069] In one embodiment, based on the pollution anomaly signal results, a drone swarm is dispatched to collect three-dimensional pollution data along a preset path to obtain initial pollution distribution data, including:

[0070] S211, obtaining pollution area range data corresponding to the pollution abnormality signal result, and constructing a three-dimensional monitoring grid. The three-dimensional monitoring grid is distributed in a rectangular shape on the horizontal plane and is a height layer with a fixed distance from the ground in the vertical direction.

[0071] By analyzing the spatial impact range of pollution anomalies, a three-dimensional monitoring network framework covering the target area is automatically generated. This framework is divided into equally spaced matrices horizontally and vertically into three-dimensional layers at preset heights (e.g., 50 meters), constructing a standardized spatial monitoring unit system. The pollution area range data includes three-dimensional spatial envelope data of longitude span, latitude span, and impact height; the three-dimensional monitoring grid is a spatial index system composed of horizontal grid coordinates (X, Y) and vertical grid (Z).

[0072] S212, based on the pollutant type characteristics in the three-dimensional monitoring grid, select the corresponding sensor combination and configure the drone monitoring equipment to collect data to obtain initial pollution distribution data.

[0073] For example, based on the characteristics of pollutant types identified within the grid, the optimal sensor combination solution is automatically matched and loaded onto the drone monitoring equipment. Through a standardized sampling process, the three-dimensional spatial distribution data of pollutants is obtained to form a structured initial pollution distribution data set. Pollutant type characteristics include chemical component identification (such as SO2 / Nox / PM2.5, etc.) and physical property (gaseous / particulate matter) detection results; the sensor combination configuration is an equipment installation solution optimized according to monitoring needs, including but not limited to main sensors (for major pollutants), auxiliary sensors (cross-validation) and meteorological sensors (temperature, humidity / wind speed).

[0074] In the above-mentioned atmospheric pollution data collection method based on three-dimensional grid monitoring, a high-precision three-dimensional spatial representation of pollution distribution is achieved through adaptive sensor configuration that intelligently matches pollutant characteristics and a three-dimensional grid scanning strategy, providing a reliable data foundation for subsequent pollution tracing.

[0075] In one embodiment, based on the pollutant type characteristics in the three-dimensional monitoring grid, a corresponding sensor combination is selected to configure the drone monitoring device. After obtaining the initial pollution distribution data, the following steps are further included:

[0076] S311, based on the initial pollution distribution data, calculate the spatial concentration change rate of the pollutant using the following gradient analysis algorithm:

[0077]

[0078] in, is the spatial concentration change rate of pollutants, is the east-west concentration gradient of the pollutant on the horizontal plane, is the north-south concentration gradient of the pollutant on the horizontal plane, is the vertical concentration gradient, is the time change rate, α, β, γ are the spatial weight coefficients, and λ is the time attenuation factor.

[0079] Specifically, a multidimensional gradient field analysis algorithm is used to calculate the spatial and temporal characteristics of pollutants, generating quantitative indicators of pollutant diffusion dynamics that provide a basis for decision-making in optimizing subsequent monitoring strategies. The spatial weight coefficient is a parameter used to balance the contributions of gradients in different directions (α+β+γ=1); the temporal decay factor is a coefficient reflecting the sensitivity of pollutant concentration to temporal changes (0<λ≤1).

[0080] Assume that the initial pollution distribution data obtained in Industrial Park B is as follows:

[0081] Spatial grid: 100m×100m×50m(X×Y×Z); time interval: Δt=10 minutes; PM2.5 concentration (μg / m 3 ) as shown in the following table:

[0082]

[0083] The central difference method was used to calculate the isotropic gradient (unit: μg / m 3 km):

[0084] Spatial gradient calculation:

[0085]

[0086] Time gradient calculation: (Converted to hourly rate of change);

[0087] Weighted composite calculation:

[0088] Take typical weight coefficients: α = 0.4, β = 0.4, γ = 0.2, λ = 0.2;

[0089]

[0090] Generate the diffusion feature vector as:

[0091] Comprehensive change rate: -222μg / m 3 km; Main diffusion direction: westward, vertically downward; Change intensity level: Level III (moderate intensity);

[0092] Recommended strategy: Densify the monitoring grid westward and downward.

[0093] In the above-mentioned dynamic monitoring method of atmospheric pollution based on multidimensional gradient analysis, by integrating the weighted calculation model of spatial three-dimensional gradient and time change rate, accurate quantitative characterization of pollutant diffusion trends is achieved, providing a scientific decision-making basis for the intelligent path planning of drone swarms.

[0094] In one embodiment, after calculating the spatial concentration change rate of pollutants using a gradient analysis algorithm based on the initial pollution distribution data, the method further includes:

[0095] S411, based on the spatial concentration change rate of pollutants, construct a path optimization evaluation function, which includes three optimization objectives: pollution gradient, information entropy, and flight energy consumption.

[0096] Alternatively, a multi-objective optimization function can be constructed based on the calculated spatial concentration variation characteristics of pollutants. This dynamic weight allocation mechanism balances the three core indicators of pollution gradient, information entropy, and flight energy consumption to generate the optimal flight strategy decision for the drone swarm. The pollution gradient indicator is a vector parameter that reflects the direction and intensity of pollutant diffusion; the information entropy indicator is a parameter that quantifies the value of the monitoring data.

[0097] S412, iteratively optimize the path optimization evaluation function through the reinforcement learning algorithm to generate a set of optimized monitoring paths for the drone swarm.

[0098] Specifically, the deep reinforcement learning model based on the Actor-Critic framework performs multiple rounds of iterative optimization on the path optimization evaluation function through a continuous training process of interaction with the environment, and finally outputs a drone swarm collaborative monitoring strategy that includes the optimal flight path sequence and sensor configuration scheme.

[0099] S413, monitor the changes of pollution gradient indicators in real time according to the optimized monitoring path set of the drone group. When the pollution gradient indicator is monitored to exceed the preset dynamic threshold, the result of the number of drones to be increased is generated based on the gradient exceeding the standard.

[0100] For example, by real-time analysis of the pollution gradient data stream collected by the drone group, when it is detected that the gradient value of a certain direction exceeds the dynamic threshold (such as the west gradient > 200 μg / m 3 km), automatically calculates the magnitude and direction of the gradient exceeding the standard, and generates recommendations on the number of additional drones and their deployment positions based on the preset expansion algorithm.

[0101] In the above-mentioned UAV collaborative monitoring method based on dynamic optimization, the efficient allocation of UAV swarm monitoring resources and real-time and accurate tracking of pollution diffusion trends are achieved through multi-objective reinforcement learning algorithms and adaptive expansion mechanisms.

[0102] In an exemplary embodiment, the real-time monitoring data collected by the drone swarm is controlled according to the configuration parameters of the adaptive monitoring network, and the pollution source tracing calculation is performed through the spatiotemporal graph neural network model to generate the pollution source location results, including:

[0103] S511, constructing a spatiotemporal correlation graph structure based on the real-time monitoring data collected by the drone swarm according to the configuration parameters of the adaptive monitoring network, wherein the spatiotemporal correlation graph structure includes the spatiotemporal coordinates of each monitoring point as a graph node and using a distance exponential function to obtain the edge weights between the nodes.

[0104] Optionally, based on the configuration parameters of the current drone swarm, the spatiotemporal monitoring data collected in real time are mapped into a graph structure data model, where the spatiotemporal coordinates (longitude, latitude, altitude, timestamp) of each monitoring point are converted into graph nodes, and the spatial correlation strength between nodes is calculated as the edge weight through the distance decay function to construct a complete spatiotemporal correlation graph of pollution diffusion.

[0105] S512, performing multi-layer feature aggregation on the spatiotemporal correlation graph through a spatiotemporal graph neural network to generate a pollution diffusion feature matrix.

[0106] Specifically, the constructed spatiotemporal correlation graph is input into the pre-trained graph neural network model, the spatial correlation features of pollution diffusion are extracted through the spatial graph convolution layer, and the dynamic evolution law is captured by the temporal convolution layer, and finally the feature matrix representing the spatiotemporal diffusion pattern of pollutants is output.

[0107] S513, the pollution diffusion characteristic matrix is integrated with the real-time meteorological data, and a pollution source probability distribution map is obtained by calculating the reverse diffusion model. The calculation formula of the reverse diffusion model is:

[0108] P=f(G,W,M)

[0109] Among them, P is the pollution probability distribution, G is the spatiotemporal correlation graph, W is the edge weight matrix, and M is the meteorological parameter matrix.

[0110] For example, a pollution diffusion feature matrix (dimension 50×16, 50 monitoring points) was combined with real-time meteorological data (wind speed 3.2 m / s, northwest wind direction, stability class B) for feature-level fusion. The reverse diffusion model was iterated 100 times, and a probability distribution map was output to identify three potential sources:

[0111] Grid coordinates Contamination probability (100，200) 0.87 (150，180) 0.92 (120，220) 0.45

[0112] In the above-mentioned pollution source tracing method based on spatiotemporal graph neural network, high-precision positioning of pollution sources and visualization of three-dimensional spatial probability distribution are achieved through multi-source data fusion and the reverse diffusion model driven by physical mechanisms.

[0113] In one embodiment, the pollution diffusion characteristic matrix is integrated with real-time meteorological data, and a pollution source probability distribution map is obtained by calculating the reverse diffusion model, including:

[0114] S611: Generate a spatial diffusion coefficient matrix according to the node distribution characteristics in the spatiotemporal association graph structure.

[0115] Optionally, based on the spatial density distribution characteristics of nodes in the spatiotemporal correlation graph structure, a spatial variation coefficient matrix reflecting the regional diffusion characteristics is generated through a kernel density estimation algorithm, and the differences in pollutant diffusion capacity in different geographical locations are quantified. The basic value of the spatial diffusion coefficient is 1.0 corresponding to standard atmospheric conditions, 1 indicates enhanced diffusion (such as open areas), and <1 indicates suppressed diffusion (such as densely built-up areas).

[0116] S612: Obtain a meteorological correction factor based on wind speed, wind direction, and atmospheric stability parameters in the real-time meteorological data.

[0117] Specifically, based on real-time wind speed, wind direction and atmospheric stability parameters, the meteorological correction factor is dynamically calculated through a multi-parameter coupled weight function to calibrate the physical transmission parameters in the diffusion model.

[0118] S613, based on the spatial diffusion coefficient matrix and meteorological correction factors, generates the contribution probability of each potential pollution source through the reverse diffusion model.

[0119] For example, the spatial diffusion coefficient matrix and the meteorological correction factor are tensor-fused and input into a reverse diffusion model based on physical constraints for probability deduction, and the quantitative value of the possibility of each grid point as a pollution source is output to form a standardized probability distribution field.

[0120] S614: Normalize the contribution probability of each potential pollution source to generate a standardized pollution source probability distribution map.

[0121] Among them, the Softmax function is used to normalize the contribution probability of pollution sources in the entire region to generate a standardized heat map that conforms to the probability distribution characteristics, in which the sum of the probability values of each grid point is 1, intuitively showing the spatial distribution characteristics of pollution sources.

[0122] In the above pollution source inversion method based on multi-factor coupling, the accurate quantitative evaluation of pollution source contribution and the visual probability distribution expression are achieved by integrating geographic spatial characteristics and real-time meteorological correction of reverse diffusion calculation.

[0123] In an exemplary embodiment, generating pollution control instructions and performing feedback control based on pollution source location results includes:

[0124] S711, performing a credibility assessment based on the pollution source location result, and generating a corresponding pollution control instruction when the assessment result meets a preset credibility condition.

[0125] Optionally, a credibility score is performed by analyzing the spatial clustering and temporal consistency characteristics of the probability distribution of pollution sources. When the comprehensive score exceeds a threshold of 85 points, a structured regulatory instruction containing specific control measures and time requirements is automatically generated.

[0126] S712: Send the pollution control instruction to the environmental monitoring system to perform pollution control operations.

[0127] For example, structured control instructions are pushed to the environmental monitoring system through a standardized API interface, triggering the automatic execution of the preset pollution control protocol, and establishing a real-time tracking mechanism for the instruction execution status.

[0128] S713, collecting environmental monitoring data after performing the pollution control operation.

[0129] Specifically, through a hybrid monitoring network consisting of drone swarms and fixed monitoring stations, multi-parameter environmental data of the target area are collected at a preset time frequency to form a verification data set containing the trend of pollutant concentration changes.

[0130] S714, based on the environmental monitoring data, the parameters of the spatiotemporal graph neural network model are updated to obtain the subsequent pollution source tracing calculation optimization results.

[0131] Optionally, the newly collected environmental monitoring data are used as training samples through an online learning mechanism, and the gradient descent algorithm is used to adjust the weight parameters of the spatiotemporal graph neural network model to optimize the model's fitting accuracy to the pollution diffusion law.

[0132] In the above-mentioned intelligent environmental control method based on closed-loop feedback, the precise implementation of pollution control measures and the continuous performance improvement of the monitoring and traceability system are achieved through the credibility-driven precise regulation and model dynamic optimization mechanism.

[0133] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0134] Based on the same inventive concept, the present application also provides a system for implementing the aforementioned method for air pollution monitoring based on drone remote sensing and machine learning. The implementation solution provided by this system is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the air pollution monitoring system based on drone remote sensing and machine learning provided below can be found in the above-mentioned limitations of the air pollution monitoring method based on drone remote sensing and machine learning, and will not be repeated here.

[0135] In an exemplary embodiment, Figure 3 As shown, an air pollution monitoring system 10 based on drone remote sensing and machine learning is provided, comprising:

[0136] The data acquisition module 11 is used to obtain fixed monitoring station and satellite remote sensing data to monitor the concentration of atmospheric pollutants in real time. When the concentration exceeds a preset threshold, a pollution abnormality signal result is generated;

[0137] The drone dispatching module 12 is used to dispatch the drone group to collect three-dimensional pollution data along a preset path according to the pollution abnormality signal results to obtain initial pollution distribution data;

[0138] The adaptive adjustment module 13 is used to dynamically adjust the monitoring path and sensor configuration parameters of the drone group based on the initial pollution distribution data through the reinforcement learning algorithm to obtain the adaptive monitoring network results;

[0139] The pollution source tracing module 14 is used to perform pollution source tracing calculations based on the real-time monitoring data collected by the adaptive monitoring network through a spatiotemporal graph neural network model to generate pollution source location results;

[0140] The control instruction generation and execution module 15 is used to generate pollution control instructions and perform feedback control according to the pollution source positioning result.

[0141] In one embodiment, the drone scheduling module 12 is further configured to:

[0142] Obtain the pollution area range data corresponding to the pollution abnormal signal results and construct a three-dimensional monitoring grid. The three-dimensional monitoring grid is distributed in a rectangular shape on the horizontal plane and is a fixed height layer from the ground in the vertical direction;

[0143] Based on the pollutant type characteristics in the three-dimensional monitoring grid, the corresponding sensor combination is selected to configure the drone monitoring equipment for data collection to obtain the initial pollution distribution data.

[0144] In one embodiment, the drone scheduling module 12 is further configured to:

[0145] Based on the initial pollution distribution data, the spatial concentration change rate of pollutants is calculated using the following gradient analysis algorithm:

[0146]

[0147] in, is the spatial concentration change rate of pollutants, is the east-west concentration gradient of the pollutant on the horizontal plane, is the north-south concentration gradient of the pollutant on the horizontal plane, is the vertical concentration gradient, is the time change rate, α, β, γ are the spatial weight coefficients, and λ is the time attenuation factor.

[0148] In one embodiment, the drone scheduling module 12 is further configured to:

[0149] Based on the spatial concentration change rate of pollutants, a path optimization evaluation function is constructed. The evaluation function includes three optimization objectives: pollution gradient, information entropy, and flight energy consumption.

[0150] The path optimization evaluation function is iteratively optimized through the reinforcement learning algorithm to generate a set of optimized monitoring paths for the drone swarm;

[0151] The changes in pollution gradient indicators are monitored in real time based on the optimized monitoring path set of the drone swarm. When the pollution gradient indicator is detected to exceed the preset dynamic threshold, the result of the number of drones that need to be increased is generated based on the gradient exceeding the standard.

[0152] In one embodiment, the pollution source tracing module 14 is further configured to:

[0153] The real-time monitoring data collected by the drone swarm is controlled by the adaptive monitoring network configuration parameters to construct a spatiotemporal correlation graph structure. The spatiotemporal coordinates of each monitoring point are used as graph nodes, and the edge weights between nodes are obtained using a distance exponential function.

[0154] Through the spatiotemporal graph neural network, multi-layer feature aggregation is performed on the spatiotemporal correlation graph to generate a pollution diffusion feature matrix;

[0155] The pollution diffusion characteristic matrix is integrated with the real-time meteorological data, and the pollution source probability distribution map is obtained through the reverse diffusion model. The calculation formula of the reverse diffusion model is:

[0156] P=f(G,W,M)

[0157] Among them, P is the pollution probability distribution, G is the spatiotemporal correlation graph, W is the edge weight matrix, and M is the meteorological parameter matrix.

[0158] In one embodiment, the pollution source tracing module 14 is further configured to:

[0159] Generate a spatial diffusion coefficient matrix based on the node distribution characteristics in the spatiotemporal correlation graph structure;

[0160] Obtain meteorological correction factors based on wind speed, wind direction and atmospheric stability parameters in real-time meteorological data;

[0161] Based on the spatial diffusion coefficient matrix and meteorological correction factors, the contribution probability of each potential pollution source is generated through the reverse diffusion model;

[0162] The contribution probability of each potential pollution source is normalized to generate a standardized pollution source probability distribution map.

[0163] In one embodiment, the control instruction generation and execution module 15 is further configured to:

[0164] Conduct credibility assessment based on pollution source location results, and generate corresponding pollution control instructions when the assessment results meet the preset credibility conditions;

[0165] Send pollution control instructions to the environmental monitoring system to perform pollution control operations;

[0166] Collect environmental monitoring data after performing pollution control operations;

[0167] The parameters of the spatiotemporal graph neural network model are updated based on environmental monitoring data to obtain subsequent pollution source tracing calculation optimization results.

[0168] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the atmospheric pollution monitoring method based on drone remote sensing and machine learning as described above are implemented.

[0169] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0170] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0171] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A method for monitoring air pollution based on drone remote sensing and machine learning, characterized in that: The method comprises: Acquire fixed monitoring station and satellite remote sensing data to monitor the concentration of atmospheric pollutants in real time, and generate a pollution anomaly signal result when the concentration exceeds a preset threshold; According to the abnormal pollution signal results, the drone group is dispatched to collect three-dimensional pollution data along a preset path to obtain initial pollution distribution data; Based on the initial pollution distribution data, the monitoring path and sensor configuration parameters of the drone swarm are dynamically adjusted through a reinforcement learning algorithm to obtain adaptive monitoring network configuration parameters; According to the adaptive monitoring network configuration parameters, the real-time monitoring data collected by the drone group is controlled, and the pollution source tracing calculation is performed through the spatiotemporal graph neural network model to generate the pollution source positioning result; Based on the pollution source positioning result, a pollution control instruction is generated and feedback control is performed.

2. The method according to claim 1, characterized in that According to the abnormal pollution signal result, the drone group is dispatched to collect three-dimensional pollution data along a preset path to obtain initial pollution distribution data, including: Obtaining pollution area range data corresponding to the pollution abnormality signal result and constructing a three-dimensional monitoring grid, wherein the three-dimensional monitoring grid is distributed in a rectangular shape on the horizontal plane and is a height layer at a fixed distance from the ground in the vertical direction; Based on the pollutant type characteristics in the three-dimensional monitoring grid, a corresponding sensor combination is selected to configure the drone monitoring equipment for data collection to obtain initial pollution distribution data.

3. The method according to claim 2, characterized in that After selecting a corresponding sensor combination to configure a drone monitoring device based on the pollutant type characteristics in the three-dimensional monitoring grid and obtaining initial pollution distribution data, the method further includes: Based on the initial pollution distribution data, the spatial concentration change rate of pollutants is calculated using the following gradient analysis algorithm: in, is the spatial concentration change rate of pollutants, is the east-west concentration gradient of the pollutant on the horizontal plane, is the north-south concentration gradient of the pollutant on the horizontal plane, is the vertical concentration gradient, is the temporal rate of change, β, β, and γ are the spatial weight coefficients, and λ is the temporal attenuation factor.

4. The method according to claim 3, characterized in that After calculating the spatial concentration change rate of pollutants using a gradient analysis algorithm based on the initial pollution distribution data, the method further includes: Based on the spatial concentration change rate of the pollutants, a path optimization evaluation function is constructed, wherein the evaluation function includes three optimization objectives: pollution gradient, information entropy, and flight energy consumption; The path optimization evaluation function is iteratively optimized by a reinforcement learning algorithm to generate a set of optimized monitoring paths for the drone swarm; The changes in the pollution gradient index are monitored in real time according to the optimized monitoring path set of the drone group. When the pollution gradient index is monitored to exceed a preset dynamic threshold, the result of the number of drones to be increased is generated based on the gradient exceeding the standard.

5. The method according to claim 1, wherein The real-time monitoring data collected by the drone swarm is controlled according to the configuration parameters of the adaptive monitoring network, and pollution source tracing calculation is performed through the spatiotemporal graph neural network model to generate pollution source positioning results, including: Constructing a spatiotemporal correlation graph structure using the real-time monitoring data collected by the drone swarm according to the adaptive monitoring network configuration parameters, wherein the spatiotemporal correlation graph structure includes using the spatiotemporal coordinates of each monitoring point as a graph node and using a distance exponential function to obtain edge weights between nodes; Performing multi-layer feature aggregation on the spatiotemporal correlation graph through the spatiotemporal graph neural network to generate a pollution diffusion feature matrix; The pollution diffusion characteristic matrix is integrated with the real-time meteorological data, and the pollution source probability distribution map is obtained by calculating the reverse diffusion model. The calculation formula of the reverse diffusion model is: P=f(G,W,M) Among them, P is the pollution probability distribution, G is the spatiotemporal correlation graph, W is the edge weight matrix, and M is the meteorological parameter matrix.

6. The method according to claim 5, characterized in that The pollution diffusion characteristic matrix is integrated with real-time meteorological data, and a pollution source probability distribution map is obtained by calculating the reverse diffusion model, including: generating a spatial diffusion coefficient matrix according to node distribution characteristics in the spatiotemporal association graph structure; Obtaining a meteorological correction factor based on wind speed, wind direction, and atmospheric stability parameters in the real-time meteorological data; Based on the spatial diffusion coefficient matrix and the meteorological correction factor, generating the contribution probability of each potential pollution source through the reverse diffusion model; The contribution probability of each potential pollution source is normalized to generate a standardized pollution source probability distribution map.

7. The method according to claim 1, characterized in that Generating pollution control instructions and executing feedback control according to the pollution source positioning result includes: Performing a credibility assessment based on the pollution source location result, and generating corresponding pollution control instructions when the assessment result meets a preset credibility condition; sending the pollution control instruction to the environmental monitoring system to perform pollution control operations; collecting environmental monitoring data after performing the pollution control operations; The parameters of the spatiotemporal graph neural network model are updated based on the environmental monitoring data to obtain subsequent pollution source tracing calculation optimization results.

8. An air pollution monitoring system based on drone remote sensing and machine learning, characterized by: The system comprises: A data acquisition module is used to acquire fixed monitoring station and satellite remote sensing data to monitor the concentration of atmospheric pollutants in real time. When it is detected that the concentration exceeds a preset threshold, a pollution abnormality signal result is generated; A drone dispatching module is used to dispatch a group of drones to collect three-dimensional pollution data along a preset path according to the pollution abnormality signal results to obtain initial pollution distribution data; An adaptive adjustment module is used to dynamically adjust the monitoring path and sensor configuration parameters of the drone swarm based on the initial pollution distribution data through a reinforcement learning algorithm to obtain an adaptive monitoring network result; A pollution source tracing module is used to perform pollution source tracing calculations based on the real-time monitoring data collected by the adaptive monitoring network through a spatiotemporal graph neural network model to generate pollution source location results; The control instruction generation and execution module is used to generate pollution control instructions and perform feedback control according to the pollution source positioning result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images