Pollution source tracking and positioning method and system

Abstract

The invention provides a pollution source tracking and positioning method and system. The method comprises the following steps: acquiring sky-ground cooperation data of a set area; based on the sky-ground cooperation data, carrying out pollution concentration identification on a set area by utilizing a pollution concentration identification model to obtain a pollution concentration space-time distribution matrix; based on the obtained historical pollution event data of the set area, pollution source positioning is carried out by using a pollution source positioning model, and candidate pollution source position information is obtained; and based on the pollution concentration space-time distribution matrix and the candidate pollution source position information, pollution source tracking and positioning are carried out by using a particle swarm optimization algorithm to obtain optimal pollution source position information in a set area. According to the invention, multi-source information is integrated through sky-ground collaborative data, and an intelligent model and an optimization algorithm are combined, so that rapid and accurate positioning of a pollution source is realized, the tracking period is greatly shortened, the processing efficiency and accuracy are improved, and pollution diffusion is effectively suppressed.

Description

Technical Field

[0001] The present invention relates to the technical field of ecological environment monitoring, and particularly to a method and system for tracking and positioning pollution sources. Background Art

[0002] Currently, ecological environment data is mainly collected by various sensors, and then inspectors complete the inspection of ecological environment protection and the tracking and positioning of pollution sources. Among them, the process of manually completing the inspection of ecological environment protection includes processes such as "manual summary → on-site verification → report writing → pollution source positioning → problem feedback", and the average time consumption for each link is 7 - 15 days.

[0003] The following problems exist in the current traditional pollution source tracking and positioning:

[0004] 1. Relying on various sensors to collect ecological environment data, the data is single and it is difficult to accurately reflect the changes in the ecological environment;

[0005] 2. All links in the process are manual intervention links, and the time consumption for each link is long, which will lead to a decrease in the tracking accuracy. Taking the water environment monitoring of a certain basin as an example, it often takes more than 2 weeks from the discovery of pollutant over - standard to the completion of the source tracing, and at this time, the pollution may have spread.

[0006] In summary, the traditional pollution source tracking and positioning has the disadvantages of slow processing speed, low efficiency and poor accuracy. Summary of the Invention

[0007] In order to overcome the defects of slow processing speed, low efficiency and poor accuracy existing in the above - mentioned traditional pollution source tracking and positioning, the present invention provides a method for tracking and positioning pollution sources, including:

[0008] Obtaining sky - ground - air collaborative data of a set area;

[0009] Based on the sky - ground - air collaborative data, using a pollution concentration recognition model to recognize the pollution concentration of the set area, and obtaining a pollution concentration spatio - temporal distribution matrix;

[0010] Based on the historical pollution event data of the obtained set area, using a pollution source positioning model to position the pollution source, and obtaining candidate pollution source location information;

[0011] Based on the pollution concentration spatio - temporal distribution matrix and the candidate pollution source location information, using a particle swarm optimization algorithm to track and position the pollution source, and obtaining the optimal pollution source location information in the set area.

[0012] Optionally, the pollution concentration recognition model is a spatio - temporal graph convolutional network;

[0013] Based on the sky-ground collaborative data, using a pollution concentration recognition model to recognize the pollution concentration in the set area, the obtained spatio-temporal distribution matrix of pollution concentration includes:

[0014] Based on the sky-ground collaborative data, using the spatio-temporal graph convolutional network to divide the set area into multiple graph nodes;

[0015] Based on the multiple graph nodes, using graph convolution and temporal convolution in the spatio-temporal graph convolutional network to perform spatial association and time dynamic capture, and obtain the spatio-temporal distribution matrix of pollution concentration.

[0016] Optionally, based on the spatio-temporal distribution matrix of pollution concentration and the candidate pollution source location information, using the particle swarm optimization algorithm to perform pollution source tracking and positioning, and obtaining the optimal pollution source location information in the set area includes:

[0017] Taking the candidate pollution source location information as the initial position of each particle in the particle swarm optimization algorithm, and randomly generating the velocity of each particle;

[0018] Based on the initial position and velocity of each particle, perform pollution diffusion simulation to obtain the simulated diffusion distribution matrix of each particle; according to the simulated diffusion distribution matrix of each particle and the spatio-temporal distribution matrix of pollution concentration, determine the fitness function value of each particle; based on the fitness function value of each particle and the preset constraint conditions, perform multiple iterative updates on the position and velocity of each particle until the maximum number of iterations is reached, and take the position of the particle with the highest fitness function value at this time as the optimal pollution source location information in the set area.

[0019] Optionally, the determining the fitness function value of each particle according to the simulated diffusion distribution matrix of each particle and the spatio-temporal distribution matrix of pollution concentration includes:

[0020] Based on each particle, determining the mean square error between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration as the fitness function value; or

[0021] Based on each particle, determining the spatial correlation index between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration as the fitness function value.

[0022] Optionally, after using the particle swarm optimization algorithm to perform pollution source tracking and positioning based on the spatio-temporal distribution matrix of pollution concentration and the candidate pollution source location information, and obtaining the optimal pollution source location information in the set area, it further includes:

[0023] Obtaining the environmental observation data of the optimal pollution source location information, where the environmental observation data includes real-time wind speed, terrain, and pollution concentration;

[0024] Input the environmental observation data into the pollution source diffusion prediction model for diffusion path simulation to obtain the pollution diffusion path.

[0025] Optionally, the pollution source diffusion prediction model is a diffusion model driven by deep reinforcement learning (DRL).

[0026] Optionally, the training process of the pollution source diffusion prediction model includes:

[0027] Obtain sample environmental observation data, which includes meteorological data and pollution concentration data;

[0028] Based on the meteorological data and the pollution concentration data, use the twin-delayed deep deterministic policy gradient algorithm to obtain the action amount of the agent; the action amount includes the diffusion coefficient of the Gaussian plume model and the wind speed adjustment amount;

[0029] Based on the action amount, use the agent to run the Gaussian plume model to generate a simulated concentration field as the state amount of the agent;

[0030] Use the reward function to calculate the reward value of the simulated concentration field and the actual observed concentration field corresponding to the pollution concentration data; based on the reward value, perform iterative training on the DRL-driven diffusion model to obtain the pollution source diffusion prediction model.

[0031] Optionally, the reward function satisfies the following formula:

[0032]

[0033] where r is the reward value, and MSE is the mean square error between the actual pollution path and the simulated pollution path.

[0034] Optionally, the policy gradient algorithm is the proximal policy optimization (PPO) algorithm or the deep deterministic policy gradient (DDPG) algorithm.

[0035] On the other hand, the present invention also provides a pollution source tracking and positioning system, including:

[0036] An acquisition module for acquiring sky-ground collaborative data of a set area;

[0037] A pollution concentration identification module for identifying the pollution concentration of the set area based on the sky-ground collaborative data by using a pollution concentration identification model to obtain a pollution concentration spatio-temporal distribution matrix;

[0038] A pollution source location module, which is used to perform pollution source location based on the obtained historical pollution event data of the set area by using a pollution source location model to obtain candidate pollution source location information; and perform pollution source tracking and location by using a particle swarm optimization algorithm based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information to obtain the optimal pollution source location information in the set area.

[0039] Optionally, the pollution concentration recognition model is a spatio-temporal graph convolutional network;

[0040] The pollution concentration recognition module is specifically used to divide the set area into multiple graph nodes by using the spatio-temporal graph convolutional network based on the sky-earth-air collaborative data; and perform spatial association and time dynamic capture by using graph convolution and temporal convolution in the spatio-temporal graph convolutional network based on the multiple graph nodes to obtain a pollution concentration spatio-temporal distribution matrix.

[0041] Optionally, the pollution source location module is specifically used to use the candidate pollution source location information as the initial position of each particle in the particle swarm optimization algorithm and randomly generate the velocity of each particle; perform pollution diffusion simulation based on the initial position and velocity of each particle to obtain a simulated diffusion distribution matrix of each particle; determine the fitness function value of each particle according to the simulated diffusion distribution matrix of each particle and the pollution concentration spatio-temporal distribution matrix; and perform multiple iterative updates on the position and velocity of each particle based on the fitness function value of each particle and preset constraint conditions until the maximum number of iterations is reached, and use the position of the particle with the highest fitness function value at this time as the optimal pollution source location information in the set area.

[0042] Optionally, the pollution source location module is specifically used to determine the mean square error between the simulated diffusion distribution matrix of the particle and the pollution concentration spatio-temporal distribution matrix as the fitness function value based on each particle; or determine the spatial correlation index between the simulated diffusion distribution matrix of the particle and the pollution concentration spatio-temporal distribution matrix as the fitness function value based on each particle.

[0043] Optionally, the pollution source tracking and location system further includes:

[0044] A pollution diffusion path prediction module, which is used to obtain environmental observation data of the optimal pollution source location information, where the environmental observation data includes real-time wind speed, terrain and pollution concentration; input the environmental observation data into the pollution source diffusion prediction model to perform diffusion path simulation to obtain a pollution diffusion path.

[0045] Optionally, the pollution source diffusion prediction model is a diffusion model driven by deep reinforcement learning DRL.

[0046] Optionally, the training process of the pollution source diffusion prediction model includes: obtaining sample environmental observation data, where the sample environmental observation data includes meteorological data and pollution concentration data; based on the meteorological data and the pollution concentration data, using the dual-delay deep deterministic policy gradient algorithm to obtain the action quantity of the intelligent agent; the action quantity includes the diffusion coefficient of the Gaussian plume model and the wind speed adjustment quantity; based on the action quantity, using the intelligent agent to run the Gaussian plume model to generate a simulated concentration field as the state quantity of the intelligent agent; using a reward function to calculate the reward value of the simulated concentration field and the actual observation concentration field corresponding to the pollution concentration data; based on the reward value, iteratively training the DRL-driven diffusion model to obtain the pollution source diffusion prediction model.

[0047] Optionally, the reward function satisfies the following formula:

[0048]

[0049] Wherein, r is the reward value, and MSE is the mean square error between the actual pollution path and the simulated pollution path.

[0050] Optionally, the policy gradient algorithm is the Proximal Policy Optimization (PPO) algorithm or the Deep Deterministic Policy Gradient (DDPG) algorithm.

[0051] On the other hand, the present invention also provides an electronic device, including: at least one processor and a memory; the memory and the processor are connected by a bus;

[0052] The memory is used to store one or more programs;

[0053] When the one or more programs are executed by the at least one processor, the pollution source tracking and positioning method described in any one of the above is implemented.

[0054] On the other hand, the present invention also provides a readable storage medium, on which an execution program is stored, and when the execution program is executed, the pollution source tracking and positioning method described in any one of the above is implemented.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] The present invention provides a method and system for tracking and locating pollution sources. The method includes: acquiring sky-earth-ground collaborative data of a set area; based on the sky-earth-ground collaborative data, using a pollution concentration identification model to identify the pollution concentration of the set area, and obtaining a pollution concentration spatio-temporal distribution matrix; based on the historical pollution event data of the acquired set area, using a pollution source location model to locate the pollution source, and obtaining candidate pollution source location information; based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information, using a particle swarm optimization algorithm to track and locate the pollution source, and obtaining the optimal pollution source location information in the set area. The present invention integrates multi-source information through sky-earth-ground collaborative data, combines intelligent models and optimization algorithms, realizes rapid and accurate location of pollution sources, greatly shortens the tracking period, improves the processing efficiency and accuracy, and effectively curbs the spread of pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a schematic flowchart of the pollution source tracking and location method of the present invention;

[0058] Figure 2 is a schematic structural diagram of the pollution source tracking and location system of the present invention;

[0059] Figure 3 is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The following further elaborates on the specific embodiments of the present invention with reference to the accompanying drawings.

[0061] Embodiment 1:

[0062] A pollution source tracking and location method provided by the present invention, the schematic flowchart is as Figure 1 shown, including:

[0063] Step 101: Acquire sky-earth-ground collaborative data of a set area;

[0064] Step 102: Based on the sky-earth-ground collaborative data, use a pollution concentration identification model to identify the pollution concentration of the set area, and obtain a pollution concentration spatio-temporal distribution matrix;

[0065] Step 103: Based on the historical pollution event data of the acquired set area, use a pollution source location model to locate the pollution source, and obtain candidate pollution source location information;

[0066] Step 104: Based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information, use a particle swarm optimization algorithm to track and locate the pollution source, and obtain the optimal pollution source location information in the set area.

[0067] A pollution source tracking and positioning method provided by an embodiment of the present invention is applied to an electronic device, which can be a personal computer (PC), a server, etc.

[0068] In the present invention, the electronic device collects sky-ground collaborative data of a set area and performs pollution source tracking and positioning based on the sky-ground collaborative data. Among them, the sky-ground collaborative data includes space-based data, air-based data, and ground-based data. Among them, the space-based data includes but is not limited to satellite data, and the satellite data can be hyperspectral images; the air-based data includes but is not limited to drone data; the ground-based data includes but is not limited to sensor data.

[0069] Exemplarily, sky-ground collaborative data is collected through space-based monitoring, air-based monitoring, and ground-based monitoring. First, a low-earth orbit satellite with an orbital altitude not exceeding 500 km is deployed to achieve space-based monitoring. The low-earth orbit satellite is equipped with a hyperspectral imager with a wavelength band not exceeding 200 and a synthetic aperture radar (SAR). Among them, the hyperspectral imager can obtain spectral data of a set area, and combined with the ability of the SAR to penetrate clouds, the spatio-temporal adaptive compressive sensing algorithm is used to reconstruct the image of the occluded area to obtain radar data. Among them, the set area can be detected at regular time intervals, and the set time interval does not exceed one hour.

[0070] Among them, the spatio-temporal adaptive compressive sensing algorithm can be implemented by the following formula:

[0071]

[0072] Among them, Y is the observation data collected by the satellite, A is a preset sensing matrix, X is the reconstructed image, λ is a preset regularization parameter, and ‖X‖ TV represents the total variation (TV) norm of the reconstructed image X.

[0073] Air-based monitoring is achieved by deploying a drone powered by a hydrogen fuel cell. Among them, the endurance of the drone can reach 10 hours. The drone is equipped with a lidar and a gas chromatograph, and can autonomously fly along the pollutant concentration gradient based on the dynamic trajectory planning algorithm to collect three-dimensional pollution distribution data in real time.

[0074] Among them, the dynamic trajectory planning algorithm can be implemented by the following formula:

[0075]

[0076] Among them, C(x, y, t) represents the pollutant concentration field collected by the drone, v(t) represents the drone speed, and P routerepresents the flight trajectory, T represents the total length of the time window considered during the trajectory planning of the UAV, and t represents the time point.

[0077] Ground-based monitoring is achieved by deploying multiple micro sensors. The sensors transmit data such as PM2.5, VOCs concentration, temperature, and humidity through the Long Range Wide Area Network (LoRaWAN) protocol, and form a high-precision ground pollution map after processing noise through adaptive Kalman filtering.

[0078] In the present invention, a pollution concentration recognition model is deployed in the electronic device. After the electronic device obtains the sky-earth-air collaborative data of a set area, it can input the sky-earth-air collaborative data into the pollution concentration recognition model, so that the pollution concentration recognition model recognizes the pollution concentration of the set area according to the sky-earth-air collaborative data and obtains a pollution concentration spatio-temporal distribution matrix.

[0079] Exemplarily, the pollution concentration recognition model extracts and fuses features from the sky-earth-air collaborative data, takes satellite pixels, UAV waypoints, and sensor positions as graph nodes, dynamically calculates the cross-platform correlation degree through the attention mechanism, and generates a pollution concentration spatio-temporal distribution matrix.

[0080] In addition, a pollution source location model is also deployed in the electronic device. The electronic device can input the historical pollution event data of a set area into the pollution source location model, so that the pollution source location model locates the pollution source according to the historical pollution event data and obtains the candidate pollution source location information that may be the pollution source in the set area. Among them, the historical pollution event data includes but is not limited to pollution type, pollution concentration, meteorological conditions, pollution range, and geographical information, etc.

[0081] Among them, the pollution source location model can output the candidate pollution source location information by means of outputting a probability heat map or discrete position points, etc.

[0082] The electronic device inputs the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information into the particle swarm optimization algorithm together. The particle swarm optimization algorithm is initialized based on the candidate pollution source location information. Specifically, the candidate pollution source location information is used as the initial position of the particle. By iteratively simulating the pollution diffusion process, calculating the mean square error between the simulation result and the pollution concentration spatio-temporal distribution matrix, and optimizing the particle trajectory in combination with the geographical obstacle constraint conditions, the optimal pollution source location information is obtained.

[0083] The electronic device determines the optimal pollution source location information determined by the particle swarm optimization algorithm as the optimal pollution source location information in the set area.

[0084] Through multi-platform collaborative observation, the present invention obtains sky-ground collaborative data, eliminates the blind area of a single data source, and effectively captures the spatio-temporal evolution law of pollution through dynamic correlation analysis of the pollution source concentration identification model. Based on the pollution source location model and the particle swarm optimization algorithm under complex constraints, combined with the spatio-temporal distribution matrix of pollution concentration, it maintains high-efficiency optimization ability, so as to accurately locate the optimal pollution source location information.

[0085] In order to accurately locate the pollution source and improve the efficiency of pollution source location, on the basis of the above embodiments, in the present invention, the above pollution source concentration identification model is a spatio-temporal graph convolutional network;

[0086] Based on the sky-ground collaborative data, using the pollution source concentration identification model to identify the pollution concentration in the set area, the obtained spatio-temporal distribution matrix of pollution concentration includes:

[0087] Based on the sky-ground collaborative data, using the spatio-temporal graph convolutional network to divide the set area into multiple graph nodes;

[0088] Based on multiple graph nodes, using graph convolution and temporal convolution in the spatio-temporal graph convolutional network to perform spatial correlation and time dynamic capture, and obtain the spatio-temporal distribution matrix of pollution concentration.

[0089] In the present invention, the pollution source concentration identification model is a spatio-temporal graph convolutional network (Spatial Temporal Graph Convolutional Networks, ST-GCN).

[0090] The electronic device inputs the sky-ground collaborative data into the ST-GCN. After receiving the sky-ground collaborative data, the ST-GCN divides the set area into multiple graph nodes. Exemplarily, the ST-GCN uses the satellite pixels, UAV flight points, and sensor positions of the set area carried by the sky-ground collaborative data as graph nodes. Based on multiple graph nodes, the ST-GCN uses graph convolution and temporal convolution to perform spatial correlation and time dynamic capture, and obtains the spatio-temporal distribution matrix of pollution concentration.

[0091] Exemplarily, the ST-GCN adopts a dual-branch structure, and the two branches are respectively a spatial branch and a temporal branch. Among them, the spatial branch is used for spatial feature extraction to obtain spatial features, specifically by aggregating the features of adjacent graph nodes through graph convolution operations; the temporal branch is used for temporal feature extraction to obtain temporal features, specifically by capturing the periodic changes of pollution diffusion through temporal convolution.

[0092] In the spatial feature extraction stage, ST-GCN calculates the correlation degree between each graph node through an attention mechanism and dynamically generates an adjacency matrix. Among them, this attention mechanism can adaptively strengthen the weights of the connection edges of the nodes around the pollution source to achieve cross-platform association. In the temporal feature extraction stage, the temporal convolutional layer adopts a dilated convolutional structure to effectively capture long-term pollution evolution data while maintaining computational efficiency.

[0093] Among them, the attention mechanism can be implemented through the following formula:

[0094]

[0095] Among them, e ij represents the weight of the connection edge between graph node i and graph node j, h i represents the feature corresponding to graph node i, h j represents the feature corresponding to graph node j, W q and W k are preset learnable parameters, and d is a preset parameter.

[0096] ST-GCN fuses the extracted spatial features and temporal features to obtain a spatio-temporal distribution matrix of pollution concentration.

[0097] In order to accurately locate the pollution source and improve the efficiency of pollution source location, based on the above embodiments, in the present invention, based on the spatio-temporal distribution matrix of pollution concentration and the candidate pollution source location information, the particle swarm optimization algorithm is used to track and locate the pollution source, and the optimal pollution source location information in the set area includes:

[0098] Use the candidate pollution source location information as the initial position of each particle in the particle swarm optimization algorithm, and randomly generate the velocity of each particle;

[0099] Based on the initial position and velocity of each particle, conduct pollution diffusion simulation to obtain the simulated diffusion distribution matrix of each particle; according to the simulated diffusion distribution matrix of each particle and the spatio-temporal distribution matrix of pollution concentration, determine the fitness function value of each particle; based on the fitness function value of each particle and the preset constraint conditions, update the position and velocity of each particle iteratively for multiple times until the maximum number of iterations is reached, and use the position of the particle with the highest fitness function value at this time as the optimal pollution source location information in the set area.

[0100] In the present invention, through the particle swarm optimization algorithm combined with the spatio-temporal distribution matrix of pollution concentration, the optimal pollution source location information can be screened out from the candidate pollution source location information.

[0101] Specifically, the candidate pollution source location information is used as the initial position of each particle in the particle swarm optimization algorithm, and the velocity of each particle is randomly generated. Based on the initial position and velocity of each particle, a pollution diffusion simulation is carried out to obtain the simulated diffusion distribution matrix of each particle.

[0102] The electronic device determines the fitness function value of each particle according to the simulated diffusion distribution matrix of each particle and the spatio-temporal distribution matrix of pollution concentration. And based on the fitness function value of each particle and the preset constraint conditions, the position and velocity of each particle are iteratively updated multiple times until the maximum number of iterations is reached, and the position of the particle with the highest fitness function value at this time is used as the optimal pollution source location information in the set area.

[0103] Exemplarily, the constraint conditions include, but are not limited to, geographical obstacle constraints and pollution source release time constraints.

[0104] In the present invention, in the process of iterative optimization based on the particle swarm optimization algorithm, each particle can simulate Gaussian plume diffusion according to the current position to generate a simulated diffusion distribution matrix.

[0105] The termination condition setting of the particle swarm optimization algorithm can be the maximum number of iterations or the fitness function value increasing less than the threshold for 10 consecutive generations, etc.

[0106] In order to accurately locate the pollution source and improve the efficiency of pollution source location, on the basis of the above embodiments, in the present invention, the determination of the fitness function value of each particle according to the simulated diffusion distribution matrix of each particle and the spatio-temporal distribution matrix of pollution concentration includes:

[0107] Based on each particle, the mean square error between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration is determined as the fitness function value; or

[0108] Based on each particle, the spatial correlation coefficient between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration is determined as the fitness function value.

[0109] In the present invention, the electronic device can determine the fitness function value in various ways, including but not limited to based on the mean square error or based on the spatial correlation coefficient.

[0110] In one example, for each particle, the electronic device determines the mean square error between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration, and determines this mean square error as the fitness function value.

[0111] In another example, for each particle, the electronic device determines the spatial correlation coefficient between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration, and determines this spatial correlation coefficient as the fitness function value.

[0112] In order to accurately locate the pollution source and improve the efficiency of pollution source location, based on the above embodiments, in the present invention, after using the particle swarm optimization algorithm to track and locate the pollution source based on the spatio-temporal distribution matrix of pollution concentration and the candidate pollution source location information, and obtaining the optimal pollution source location information in the set area, it further includes:

[0113] Obtain the environmental observation data of the optimal pollution source location information, where the environmental observation data includes real-time wind speed, terrain, and pollution concentration;

[0114] Input the environmental observation data into the pollution source diffusion prediction model to simulate the diffusion path and obtain the pollution diffusion path.

[0115] In the present invention, after the electronic device determines the optimal pollution source location information, the electronic device can also monitor and simulate the pollution at the optimal pollution source location information to obtain the pollution diffusion path.

[0116] Specifically, the electronic device obtains the environmental observation data of the optimal pollution source location information, and the environmental observation data includes but is not limited to real-time wind speed, terrain, and pollution concentration. A pollution source diffusion prediction model is deployed in the electronic device. After obtaining the environmental observation data, the electronic device inputs the environmental observation data into the pollution source diffusion prediction model to simulate the diffusion path and obtain the pollution diffusion path.

[0117] Among them, the electronic device can normalize the real-time wind speed, terrain, and pollution concentration of the optimal pollution source location information respectively, and then splice the normalization results into a spatio-temporal tensor. The electronic device inputs the spatio-temporal tensor into the pollution source diffusion prediction model.

[0118] In order to accurately locate the pollution source and improve the efficiency of pollution source location, based on the above embodiments, in the present invention, the above pollution source diffusion prediction model is a diffusion model driven by deep reinforcement learning DRL.

[0119] In order to accurately locate the pollution source and improve the efficiency of pollution source location, based on the above embodiments, in the present invention, the training process of the above pollution source diffusion prediction model includes:

[0120] Obtain sample environmental observation data, where the sample environmental observation data includes meteorological data and pollution concentration data;

[0121] Based on the meteorological data and the pollution concentration data, use the twin-delayed deep deterministic policy gradient algorithm to obtain the action amount of the intelligent agent; the action amount includes the diffusion coefficient of the Gaussian plume model and the wind speed adjustment amount;

[0122] Based on the action amount, use the intelligent agent to run the Gaussian plume model to generate a simulated concentration field as the state amount of the intelligent agent;

[0123] Using the reward function, calculate the reward value of the simulated concentration field corresponding to the actual observed concentration field of the pollution concentration data; based on the reward value, iteratively train the DRL-driven diffusion model to obtain the pollution source diffusion prediction model.

[0124] In the present invention, the pollution source diffusion prediction model is a diffusion model driven by Deep Reinforcement Learning (DRL).

[0125] The training process of the DRL-driven diffusion model includes: obtaining sample environmental observation data, which includes meteorological data and pollution concentration data. Based on the meteorological data and pollution concentration data, using the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, obtain the action quantity of the intelligent agent; the action quantity includes the diffusion coefficient of the Gaussian plume model and the wind speed adjustment quantity; based on the action quantity, use the intelligent agent to run the Gaussian plume model to generate a simulated concentration field as the state quantity of the intelligent agent; using the reward function, calculate the reward value of the simulated concentration field corresponding to the actual observed concentration field of the pollution concentration data; based on the reward value, iteratively train the DRL-driven diffusion model to obtain the pollution source diffusion prediction model.

[0126] Specifically, the TD3 algorithm is implemented by the following formula:

[0127]

[0128] Among them, Q(s,a) represents the expected return value of the current state quantity s and the action quantity a, s is the state quantity, a is the action quantity, E represents the operation of taking the expected value, r and γ are preset parameters, represents the output of the target Q function, which is used to stabilize the training process and reduce the overestimation problem, π φ’ (s’) represents the action output by the target policy network according to the next state s’, θ’ i and φ’ are both parameters of the target policy network, and i may represent one of the two Q networks, i = 1, 2.

[0129] In the present invention, the action quantity of the intelligent agent includes the diffusion coefficient of the Gaussian plume model and the wind speed adjustment quantity. Among them, the diffusion coefficient is used to control the horizontal diffusion range of pollutants, and the wind speed adjustment quantity is used to adjust the advection term in the Gaussian plume model to match the actual wind field.

[0130] In order to accurately locate the pollution source and improve the efficiency of pollution source location, on the basis of the above embodiments, in the present invention, the above reward function satisfies the following formula:

[0131]

[0132] Among them, r is the reward value, and MSE is the mean square error between the actual pollution path and the simulated pollution path.

[0133] In the present invention, the reward function can be the reciprocal of the mean square error between the simulated concentration field and the actual observed concentration field to achieve maximum alignment.

[0134] Exemplarily, the reward function can be implemented by the following formula:

[0135]

[0136] Among them, r is the reward value, and MSE is the mean square error between the actual pollution path and the simulated pollution path.

[0137] In the present invention, it is possible to judge whether the pollution source diffusion prediction model converges based on the reward function. If it is determined that the pollution source diffusion prediction model converges, the action amount at this time is used as the network parameter of the pollution source diffusion prediction model. If it is determined that the pollution source diffusion prediction model does not converge, the action amount is optimized based on the policy gradient algorithm until convergence.

[0138] In order to accurately locate the pollution source and improve the efficiency of pollution source location, on the basis of the above embodiments, in the present invention, the above policy gradient algorithm is the Proximal Policy Optimization (PPO) algorithm or the Deep Deterministic Policy Gradient (DDPG) algorithm.

[0139] In the present invention, the policy gradient algorithm is the Proximal Policy Optimization (PPO) algorithm or the Deep Deterministic Policy Gradient (DDPG) algorithm.

[0140] On the basis of the disclosure of the present invention, a lightweight AI chip can be deployed at the sensor node. Through the model distillation and compression algorithm, the algorithms and models provided by the present invention are run for distillation and compression to obtain the ST-GCN micro model, realizing local pollution concentration anomaly detection and early warning.

[0141] The following uses a specific embodiment to illustrate the embodiments of the present invention. The following steps are included in this embodiment:

[0142] (1) Obtain the sky-ground collaborative data of the set area;

[0143] (2) Based on the sky-ground collaborative data, use the pollution concentration identification model to identify the pollution concentration of the set area to obtain the pollution concentration spatio-temporal distribution matrix;

[0144] (3) Based on the historical pollution event data of the set area obtained, use the pollution source location model to locate the pollution source and obtain the candidate pollution source location information;

[0145] (4) Based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information, use the particle swarm optimization algorithm to track and locate the pollution source, and obtain the optimal pollution source location information in the set area.

[0146] (5) Obtain the environmental observation data of the optimal pollution source location information, and the environmental observation data includes real-time wind speed, terrain and pollution concentration;

[0147] (6) Input the environmental observation data into the pollution source diffusion prediction model to simulate the diffusion path and obtain the pollution diffusion path.

[0148] The present invention integrates multi-source information through sky-ground-sky cooperation data, combines intelligent models and optimization algorithms, realizes rapid and accurate positioning of pollution sources, greatly shortens the tracking period, improves the processing efficiency and accuracy, and effectively curbs pollution diffusion.

[0149] The present invention relates to a sky-ground integrated ecological environment monitoring system and a pollution source tracking method. Based on a three-dimensional monitoring network constructed by satellite remote sensing, low-altitude drone patrols and ground sensors, it realizes comprehensive perception and accurate tracing of air, water and soil pollution. The present invention includes a data acquisition module, a pollution source automatic identification module, a dynamic data visualization platform and a pollution diffusion simulation module. Through multi-source data fusion and machine learning algorithms, it can quickly locate pollution sources and simulate their diffusion paths, providing a scientific basis for pollution control. The present invention has the advantages of wide monitoring coverage and high positioning accuracy, and is widely applicable to the field of comprehensive ecological environment monitoring.

[0150] Embodiment 2:

[0151] Based on the same inventive concept, the present invention also provides a pollution source tracking and positioning system, the structural schematic diagram is as Figure 2 shown, including:

[0152] An acquisition module 201 for acquiring sky-ground cooperation data of a set area;

[0153] A pollution concentration identification module 202 for identifying the pollution concentration of a set area based on the sky-ground cooperation data by using a pollution concentration identification model to obtain a pollution concentration spatio-temporal distribution matrix;

[0154] The pollution source location module 203 is used to locate the pollution source based on the historical pollution event data of the set area obtained, using the pollution source location model to obtain the candidate pollution source location information; based on the spatio-temporal distribution matrix of pollution concentration and the candidate pollution source location information, using the particle swarm optimization algorithm to track and locate the pollution source, and obtain the optimal pollution source location information in the set area.

[0155] In a specific implementation manner, the pollution concentration recognition model is a spatio-temporal graph convolutional network;

[0156] The pollution concentration recognition module 202 is specifically used to divide the set area into multiple graph nodes based on the sky-earth-air collaborative data, using the spatio-temporal graph convolutional network; based on the multiple graph nodes, using graph convolution and temporal convolution in the spatio-temporal graph convolutional network to perform spatial association and time dynamic capture, and obtain the spatio-temporal distribution matrix of pollution concentration.

[0157] In a specific implementation manner, the pollution source location module 203 is specifically used to take the candidate pollution source location information as the initial position of each particle in the particle swarm optimization algorithm, and randomly generate the velocity of each particle; perform pollution diffusion simulation based on the initial position and velocity of each particle to obtain the simulated diffusion distribution matrix of each particle; determine the fitness function value of each particle according to the simulated diffusion distribution matrix of each particle and the spatio-temporal distribution matrix of pollution concentration; based on the fitness function value of each particle and the preset constraint conditions, perform multiple iterative updates on the position and velocity of each particle until the maximum number of iterations is reached, and take the position of the particle with the highest fitness function value at this time as the optimal pollution source location information in the set area.

[0158] In a specific implementation manner, the pollution source location module 203 is specifically used to determine the mean square error between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration as the fitness function value based on each particle; or, determine the spatial correlation index between the simulated diffusion distribution matrix of the particle and the spatio-temporal distribution matrix of pollution concentration as the fitness function value based on each particle.

[0159] In a specific implementation manner, the pollution source tracking and location system further includes:

[0160] The pollution diffusion path prediction module 204 is used to obtain the environmental observation data of the optimal pollution source location information, and the environmental observation data includes real-time wind speed, terrain and pollution concentration; input the environmental observation data into the pollution source diffusion prediction model to perform diffusion path simulation, and obtain the pollution diffusion path.

[0161] In a specific implementation manner, the pollution source diffusion prediction model is a diffusion model driven by deep reinforcement learning DRL.

[0162] In a specific implementation manner, the training process of the pollution source diffusion prediction model includes: obtaining sample environmental observation data, which includes meteorological data and pollution concentration data; based on the meteorological data and the pollution concentration data, using the dual-delay deep deterministic policy gradient algorithm to obtain the action quantity of the intelligent agent; the action quantity includes the diffusion coefficient of the Gaussian plume model and the wind speed adjustment quantity; based on the action quantity, using the intelligent agent to run the Gaussian plume model to generate a simulated concentration field as the state quantity of the intelligent agent; using the reward function to calculate the reward value of the simulated concentration field and the actual observed concentration field corresponding to the pollution concentration data; based on the reward value, performing iterative training on the DRL-driven diffusion model to obtain the pollution source diffusion prediction model.

[0163] In a specific implementation manner, the reward function satisfies the following formula:

[0164]

[0165] Where r is the reward value, and MSE is the mean square error between the actual pollution path and the simulated pollution path.

[0166] In a specific implementation manner, the policy gradient algorithm is the proximal policy optimization (PPO) algorithm or the deep deterministic policy gradient (DDPG) algorithm.

[0167] Embodiment 3:

[0168] As Figure 3 shown, the present invention also provides an electronic device, which may be a computer device, a single-chip microcomputer device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected through a bus; the memory can be used to store an execution program, and the exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, and the data can be called and / or modified when the instructions are executed.

[0169] The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of a pollution source tracking and positioning method in the above embodiments.

[0170] Embodiment 4:

[0171] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device-readable storage medium (Memory). The electronic device-readable storage medium is a memory device in the electronic device and is used to store programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device and, of course, the extended storage medium supported by the electronic device. The storage medium provides a storage space, and this storage space stores the operating system of the terminal. And, in this storage space, one or more instructions suitable for being loaded and executed by the processor are also stored. These instructions can be one or more executable programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. By the processor loading and executing one or more instructions stored in the storage medium, the steps of a pollution source tracking and positioning method in the above embodiments can be implemented.

[0172] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0173] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0174] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that after reading the present invention, various changes, modifications, or equivalent replacements can still be made to the specific implementation manners of the application. However, these changes, modifications, or equivalent replacements are all within the scope of the protection of the claims pending for the application.

Claims

1. A method for tracking and positioning pollution sources, characterized in that, Including: Obtaining sky-earth-ground collaborative data of a set area; Based on the sky-earth-ground collaborative data, using a pollution concentration recognition model to recognize the pollution concentration of the set area, and obtaining a pollution concentration spatio-temporal distribution matrix; Based on the historical pollution event data of the obtained set area, using a pollution source location model to locate the pollution source, and obtaining candidate pollution source location information; Based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information, using a particle swarm optimization algorithm to track and locate the pollution source, and obtaining the optimal pollution source location information in the set area.

2. The method according to claim 1, characterized in that The pollution concentration recognition model is a spatio-temporal graph convolutional network; The step of, based on the sky-earth-ground collaborative data, using a pollution concentration recognition model to recognize the pollution concentration of the set area, and obtaining a pollution concentration spatio-temporal distribution matrix includes: Based on the sky-earth-ground collaborative data, using the spatio-temporal graph convolutional network to divide the set area into multiple graph nodes; Based on the multiple graph nodes, using graph convolution and temporal convolution in the spatio-temporal graph convolutional network to perform spatial association and time dynamic capture, and obtaining a pollution concentration spatio-temporal distribution matrix.

3. The method according to claim 1 or 2, characterized in that, The step of, based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information, using a particle swarm optimization algorithm to track and locate the pollution source, and obtaining the optimal pollution source location information in the set area includes: Taking the candidate pollution source location information as the initial position of each particle in the particle swarm optimization algorithm, and randomly generating the velocity of each particle; Based on the initial position and velocity of each particle, performing pollution diffusion simulation to obtain a simulated diffusion distribution matrix of each particle; according to the simulated diffusion distribution matrix of each particle and the pollution concentration spatio-temporal distribution matrix, determining the fitness function value of each particle; based on the fitness function value of each particle and preset constraint conditions, iteratively updating the position and velocity of each particle multiple times until the maximum number of iterations is reached, and taking the position of the particle with the highest fitness function value at this time as the optimal pollution source location information in the set area.

4. The method according to claim 3, characterized in that, The step of, according to the simulated diffusion distribution matrix of each particle and the pollution concentration spatio-temporal distribution matrix, determining the fitness function value of each particle includes: Based on each particle, determining the mean square error between the simulated diffusion distribution matrix of the particle and the pollution concentration spatio-temporal distribution matrix as the fitness function value; or Based on each particle, determining the spatial correlation coefficient between the simulated diffusion distribution matrix of the particle and the pollution concentration spatio-temporal distribution matrix as the fitness function value.

5. The method according to any one of claims 1-4, characterized in that, After the step of, based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information, using a particle swarm optimization algorithm to track and locate the pollution source, and obtaining the optimal pollution source location information in the set area, further including: Obtaining environmental observation data of the optimal pollution source location information, where the environmental observation data includes real-time wind speed, terrain, and pollution concentration; Inputting the environmental observation data into the pollution source diffusion prediction model to perform diffusion path simulation, and obtaining a pollution diffusion path.

6. The method according to claim 5, wherein The pollution source diffusion prediction model is a diffusion model driven by deep reinforcement learning DRL.

7. The method according to claim 6, wherein The training process of the pollution source diffusion prediction model includes: Obtain sample environmental observation data, where the sample environmental observation data includes meteorological data and pollution concentration data; Based on the meteorological data and the pollution concentration data, use the dual-delay deep deterministic policy gradient algorithm to obtain the action amount of the agent; the action amount includes the diffusion coefficient of the Gaussian plume model and the wind speed adjustment amount; Based on the action amount, use the agent to run the Gaussian plume model to generate a simulated concentration field as the state amount of the agent; Use the reward function to calculate the reward value of the simulated concentration field and the actual observed concentration field corresponding to the pollution concentration data; perform iterative training on the DRL-driven diffusion model based on the reward value to obtain the pollution source diffusion prediction model.

8. The method according to claim 7, wherein The reward function satisfies the following formula: where r is the reward value, and MSE is the mean square error between the actual pollution path and the simulated pollution path.

9. The method according to claim 7, wherein The policy gradient algorithm is the Proximal Policy Optimization (PPO) algorithm or the Deep Deterministic Policy Gradient (DDPG) algorithm.

10. A pollution source tracking and positioning system, characterized in that, It includes: An acquisition module for acquiring sky-ground collaborative data of a set area; A pollution concentration identification module for identifying the pollution concentration of the set area using a pollution concentration identification model based on the sky-ground collaborative data to obtain a pollution concentration spatio-temporal distribution matrix; A pollution source location module for locating the pollution source using a pollution source location model based on the historical pollution event data of the set area obtained to obtain candidate pollution source location information; Based on the pollution concentration spatio-temporal distribution matrix and the candidate pollution source location information, use the particle swarm optimization algorithm to perform pollution source tracking and location to obtain the optimal pollution source location information in the set area.

Images