Urban atmosphere PM2.5 concentration inversion and dynamic early warning system fused with multi-source remote sensing data

Abstract

The invention relates to the technical field of atmosphere PM2.5 concentration early warning, and discloses an urban atmosphere PM2.5 concentration inversion and dynamic early warning system fused with multi-source remote sensing data. Comprising a main control module, a sky-land-sea three-dimensional monitoring network module, a multi-source data fusion module, a digital twin deduction module, a dynamic emission reduction strategy generation module, a safety early warning module and an auxiliary module, and the signal control module is used for performing signal control on the sky-land-sea three-dimensional monitoring network module, the multi-source data fusion module, the digital twinborn deduction module, the dynamic emission reduction strategy generation module, the safety early warning module and the auxiliary module and is used for coordinating operation and data scheduling of the modules. According to the invention, monitoring blind areas are eliminated through satellite-unmanned aerial vehicle-ground-ocean cooperation, full-space coverage is realized, and PM2.5 inversion error lt is realized through multi-source data fusion; and high-precision inversion is realized, environment and economic targets are balanced through an MORL algorithm, and an abnormal root is positioned through causal reasoning, so that an intelligent decision-making effect is achieved.

Description

Technical Field

[0001] This invention relates to the field of atmospheric PM2.5 concentration early warning technology, specifically to an urban atmospheric PM2.5 concentration inversion and dynamic early warning system that integrates multi-source remote sensing data. Background Technology

[0002] Atmospheric PM2.5 (fine particulate matter) refers to solid or liquid particulate matter in the air with an aerodynamic diameter of 2.5 micrometers or less.12 Its diameter is less than 1 / 2045 the thickness of a human hair. It can remain suspended in the air for extended periods and enter the lungs and even the bloodstream through respiration. To ensure the safety of residents, it is usually necessary to deploy atmospheric PM2.5 concentration inversion and dynamic early warning systems to monitor PM2.5 concentrations in real time.

[0003] Traditional atmospheric PM2.5 concentration inversion and dynamic early warning systems rely on manual intervention for multi-source data fusion, have low standardization, such as the difference between MODIS and Sentinel data formats leading to fusion errors, and traditional physical models (such as WRF-Chem) take more than 12 hours to calculate, which cannot meet the needs of dynamic early warning. In addition, blind spots in urban street and valley terrain monitoring are common, and existing sensor networks cannot capture microscale pollution distribution. Therefore, a new urban atmospheric PM2.5 concentration inversion and dynamic early warning system that integrates multi-source remote sensing data is proposed to solve the above-mentioned problems. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an urban atmospheric PM2.5 concentration inversion and dynamic early warning system that integrates multi-source remote sensing data. This system boasts advantages such as high intelligence and better monitoring and early warning effects. It solves the problems of traditional atmospheric PM2.5 concentration inversion and dynamic early warning systems, which rely on manual intervention for multi-source data fusion, have low standardization (e.g., differences in MODIS and Sentinel data formats leading to fusion errors), require traditional physical models (such as WRF-Chem) to take over 12 hours to calculate (failing to meet dynamic early warning requirements), and suffer from widespread blind spots in urban street and valley terrain monitoring, making it impossible for existing sensor networks to capture microscale pollution distribution.

[0005] (II) Technical Solution To achieve the aforementioned goals of high intelligence and better monitoring and early warning effects, this invention provides the following technical solution: a system for inverting and dynamically warning urban atmospheric PM2.5 concentration by integrating multi-source remote sensing data, comprising a main control module, a three-dimensional monitoring network module (sky, land, and sea), a multi-source data fusion module, a digital twin simulation module, a dynamic emission reduction strategy generation module, a safety early warning module, and an auxiliary module. The main control module provides signal control to the three-dimensional monitoring network module (sky, land, and sea), the multi-source data fusion module, the digital twin simulation module, the dynamic emission reduction strategy generation module, the safety early warning module, and the auxiliary module, coordinating the operation and data scheduling of each module. The aforementioned three-dimensional monitoring network module, encompassing the sky, land, and sea, enables full-space PM2.5 data collection through satellites, drones, ground sensors, and ocean buoys. The multi-source data fusion module uses a spatiotemporal dual-domain feature decoupling algorithm (STFD) and an asymmetric graph convolutional network (AGCN) to integrate heterogeneous data; The digital twin simulation module combines the WRF-Chem physical model with the GCN-GRU spatiotemporal recursive network to achieve minute-level diffusion simulation. The dynamic emission reduction strategy generation module outputs optimized control schemes based on multi-objective reinforcement learning (MORL) and causal inference engine; The safety early warning module achieves closed-loop early warning through dynamic threshold algorithm, pollution source tracing model and emergency response simulation; The auxiliary module provides data visualization, user interaction, and system management functions.

[0006] Preferably, in the three-dimensional monitoring network module for sky, land, and sea: Satellite data inversion was performed using the ENVI 2018 aerosol inversion plugin, which achieved hourly PM2.5 inversion at a resolution of 1km×1km based on MODIS / Sentinel data. Low-altitude UAV monitoring uses the LSTM trajectory optimization algorithm to dynamically plan scanning paths, covering blind spots in urban streets and valleys; The marine monitoring node integrates SO2 / NO2 sensors and lidar to establish a land-sea pollution transmission coefficient model.

[0007] Preferably, in the multi-source data fusion module: The STFD algorithm dynamically adjusts the fusion weights of satellite and ground data through an attention mechanism, resulting in an inversion error of <5μg / m³. AGCN network processes the spatial correlation between UAV flight paths and fixed monitoring station data, improving modeling accuracy by 40%.

[0008] Preferably, in the digital twin inference module: The WRF-Chem physical model simulates atmospheric chemical processes and provides basic diffusion predictions; The AI-enhanced model combines the spatiotemporal recursive networks of GCN and GRU to achieve minute-level inference and updates.

[0009] Preferably, in the dynamic emission reduction strategy generation module: the MORL algorithm generates a Pareto optimal solution between PM2.5 control and economic losses; the causal inference engine analyzes the correlation between emission anomalies and process parameters through a Bayesian network.

[0010] Preferably, the dynamic threshold early warning algorithm adjusts the early warning threshold based on sliding window statistics (such as the 3σ principle); The pollution source tracing model integrates Bayesian networks and graph neural networks (GNNs) to locate pollution sources; The emergency response simulation is integrated with the digital twin module to simulate the impact of control measures on PM2.5 concentration.

[0011] Preferably, the auxiliary module includes a user interaction module and a system management module; The data visualization module supports 3D displays such as PM2.5 concentration heat maps and pollution diffusion animations; The user interaction module provides public service interfaces such as pollution event subscription and early warning push; The system management module monitors device status, data quality, and manages access permissions.

[0012] Preferably, the main control module adopts a distributed architecture, which supports real-time scheduling of data streams and exception handling between modules.

[0013] Preferably, satellite-to-ground data assimilation is achieved within 90 seconds through edge computing nodes.

[0014] (III) Beneficial Effects Compared with existing technologies, this invention provides an urban atmospheric PM2.5 concentration inversion and dynamic early warning system that integrates multi-source remote sensing data, which has the following beneficial effects: 1. This urban atmospheric PM2.5 concentration inversion and dynamic early warning system, which integrates multi-source remote sensing data, eliminates monitoring blind spots through satellite-UAV-ground-ocean collaboration, achieving full spatial coverage. The fusion of multi-source data reduces the PM2.5 inversion error to <5μg / m³, achieving high-precision inversion. By balancing environmental and economic goals through the MORL algorithm, and causal reasoning to locate the root causes of anomalies, it achieves the effect of intelligent decision-making.

[0015] 2. This urban atmospheric PM2.5 concentration inversion and dynamic early warning system, which integrates multi-source remote sensing data, forms a complete closed loop through dynamic threshold adjustment, pollution source tracing, and emergency simulation. Furthermore, the 3D visualization and public interaction interface enhance governance participation. The system achieves end-to-end intelligent management from data collection to decision support through deep coupling of physical models and AI. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the urban atmospheric PM2.5 concentration inversion and dynamic early warning system that integrates multi-source remote sensing data according to the present invention. Figure 2 This is a schematic diagram of the working process of the atmospheric PM2.5 concentration inversion and dynamic early warning system of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1-2 A system for retrieving and dynamically warning urban atmospheric PM2.5 concentration by integrating multi-source remote sensing data includes a main control module, a three-dimensional monitoring network module (sky, land, and sea), a multi-source data fusion module, a digital twin simulation module, a dynamic emission reduction strategy generation module, a safety warning module, and an auxiliary module. The main control module provides signal control to the three-dimensional monitoring network module, the multi-source data fusion module, the digital twin simulation module, the dynamic emission reduction strategy generation module, the safety warning module, and the auxiliary module, and coordinates the operation and data scheduling of each module. The aforementioned three-dimensional monitoring network module, encompassing the sky, land, and sea, enables full-space PM2.5 data collection through satellites, drones, ground sensors, and ocean buoys. The multi-source data fusion module uses a spatiotemporal dual-domain feature decoupling algorithm (STFD) and an asymmetric graph convolutional network (AGCN) to integrate heterogeneous data; The digital twin simulation module combines the WRF-Chem physical model with the GCN-GRU spatiotemporal recursive network to achieve minute-level diffusion simulation. The dynamic emission reduction strategy generation module outputs optimized control schemes based on multi-objective reinforcement learning (MORL) and causal inference engine; The safety early warning module achieves closed-loop early warning through dynamic threshold algorithm, pollution source tracing model and emergency response simulation; The auxiliary module provides data visualization, user interaction, and system management functions.

[0019] Example 1 In the aforementioned three-dimensional monitoring network module: Satellite data inversion was performed using the ENVI 2018 aerosol inversion plugin, which achieved hourly PM2.5 inversion at a resolution of 1km×1km based on MODIS / Sentinel data. Low-altitude UAV monitoring uses the LSTM trajectory optimization algorithm to dynamically plan scanning paths, covering blind spots in urban streets and valleys; The marine monitoring node integrates SO2 / NO2 sensors and lidar to establish a land-sea pollution transmission coefficient model.

[0020] Specific algorithm steps for the air-ground-sea integrated monitoring network module: Step 1: Satellite Data Inversion Algorithm 6S model radiative transfer calculation 101. Set geometric parameters (solar zenith angle, azimuth angle, satellite observation angle, etc.), input atmospheric parameters (aerosol type, visibility, water vapor content, etc.), run the 6S model to generate radiation correction factors, and output surface reflectance data; 102 Machine Learning Correction Process: Preparing Training Data: Pairing raw satellite radiation values ​​with measured PM2.5 concentrations on the ground, constructing a random forest / XGBoost model, with inputs including 6S output values, meteorological data, terrain elevation, etc. Step 2: Low-altitude UAV monitoring algorithm 201 LSTM Track Optimization: Data Preprocessing: Cleaning historical track data for position, velocity, and attitude angles; Network Construction: Input layer (time step = 10), LSTM layer (hidden units = 64), fully connected layer; Dynamic Programming: Real-time track updates to avoid obstacles, coverage increased to 95%; 202 Path Planning and Execution Initialize the open / close list, add the starting point to the OPEN queue, calculate the minimum f(n) node in the main loop, expand adjacent nodes and update the cost, generate the optimal scan path through path backtracking, and support street valley blind zone coverage. Step 3: Ocean Monitoring Node Algorithm 301 Land-Sea Transmission Coefficient Modeling and Data Gridding: Kriging interpolation is used to process sparse monitoring point data; Establish the transmission equation: Where α, β, and γ are fitting coefficients. 302 LiDAR data analysis, signal preprocessing: background subtraction, height correction, normalization inversion calculation: Klett algorithm is used to solve for aerosol extinction coefficient, output SO2 / NO2 concentration and transmission flux. Step 4: Data Synchronization: Satellite updates hourly, drones scan every 15 minutes; Real-time transmission quality control of ocean buoys: Dynamic weighted fusion of multi-source data using the STFD algorithm; Edge computing: Deploying lightweight models on monitoring nodes to achieve data assimilation within 90 seconds. Example 2 In the multi-source data fusion module: The STFD algorithm dynamically adjusts the fusion weights of satellite and ground data through an attention mechanism, resulting in an inversion error of <5μg / m³. AGCN network processes the spatial correlation between UAV flight paths and fixed monitoring station data, improving modeling accuracy by 40%.

[0021] The specific process of the dual-domain feature decoupling algorithm (STFD) and the asymmetric graph convolutional network (AGCN): Step 1: Data Preprocessing Stage Satellite data standardization Radiometric calibration and atmospheric correction were performed on the aerosol optical depth (AOD) data from the Gaofen satellite to eliminate the influence of the solar elevation angle. A 1km×1km grid was used for resampling to match the spatial scale of the ground monitoring station. Ground sensor calibration involves outlier detection (3σ principle) and time alignment (15-minute interval interpolation) of PM2.5 monitoring station data. Drone trajectory optimization, based on LSTM to plan scanning paths, ensures that the coverage rate of blind spots in street valleys reaches more than 95%; Step 2: Spatiotemporal Dual-Domain Feature Decoupling (STFD) Algorithm Spatiotemporal Feature Extraction: Spatial Domain: Spatial distribution features of satellite data are extracted using 3D convolution with a window size of 5×5×3. Temporal Domain: A bidirectional GRU is used to capture the diurnal variation cycle of PM2.5 and the characteristics of sudden pollution events. Attention-weighted fusion: Dynamically calculating the weights of satellite and ground data. in, .

[0022] Pollution source separation We constructed the Local Pollution Index (LPI) and the Regional Transport Index (RTI), and separated the two types of features through orthogonal decomposition. Step 3: Construction of Asymmetric Graph Convolutional Network (AGCN) Graph structure definition Nodes: Monitoring stations / UAV waypoints, with feature vectors containing parameters such as PM2.5 concentration, wind speed, and humidity; Edge weights: Spatial correlation is calculated based on a Gaussian kernel function, with a distance attenuation coefficient α = 0.3. Adaptive graph convolution Node feature update formula: in Given an adjacency matrix with self-loops, Degree matrix Learning with animated graphics The connection strength between UAV waypoints and fixed stations can be dynamically adjusted using a trainable parameter matrix. Step 4: Joint Optimization and Output Loss function design The combined loss is L = 0.7L. MAE +0.3L graph L graph For graph structure smoothing constraints; Edge computing deployment Deploying a lightweight model on the Huawei Atlas 500 edge server enables prediction of a 10km×10km area within 90 seconds; Digital twin integration The output layer interfaces with the WRF-Chem model to generate a 72-hour pollution diffusion simulation animation.

[0023] Example 3 In the digital twin inference module: The WRF-Chem physical model simulates atmospheric chemical processes and provides basic diffusion predictions; The AI-enhanced model combines the spatiotemporal recursive networks of GCN and GRU to achieve minute-level inference and updates.

[0024] In the dynamic emission reduction strategy generation module: the MORL algorithm generates a Pareto optimal solution between PM2.5 control and economic losses; the causal inference engine analyzes the correlation between emission anomalies and process parameters through a Bayesian network.

[0025] The dynamic threshold warning algorithm adjusts the warning threshold based on sliding window statistics (such as the 3σ principle); The pollution source tracing model integrates Bayesian networks and graph neural networks (GNNs) to locate pollution sources; The emergency response simulation is integrated with the digital twin module to simulate the impact of control measures on PM2.5 concentration.

[0026] In this embodiment, the model co-inference process Step 1: Initial Prediction Phase WRF-Chem runs a 5-minute simulation and outputs the basic diffusion field. The results are transmitted to the GCN-GRU network via the NetCDF interface. Step 2: AI Correction Phase GCN-GRU completes deviation correction within 30 seconds: Spatial Correction: Capturing Local Anomalies through Graph Convolution Time Correction: Short-term contamination events are remembered via GRU memory. Output the corrected concentration field (update frequency: 1 minute / time) Step 3: Dynamic Feedback Mechanism Real-time monitoring data triggers model weight adjustments: When the deviation between the monitored value and the predicted value exceeds 15%, online learning is activated, and the GCN-GRU parameters are updated using the FTRL algorithm (step size η=0.01). In addition, the auxiliary module includes a user interaction module and a system management module; The data visualization module supports 3D displays such as PM2.5 concentration heat maps and pollution diffusion animations; The user interaction module provides public service interfaces such as pollution event subscription and early warning push; The system management module monitors device status, data quality, and manages access permissions.

[0027] The main control module adopts a distributed architecture, which supports real-time scheduling of data streams and exception handling between modules.

[0028] Assimilation of satellite-to-ground data within 90 seconds via edge computing nodes In summary, the urban atmospheric PM2.5 concentration retrieval and dynamic early warning system, which integrates multi-source remote sensing data, includes the following steps when in use: Step 1: Data collection for a three-dimensional monitoring network covering the sky, land, and sea, including satellite monitoring, drones, and marine monitoring, to ensure full regional coverage; Step 2: Multi-source data fusion processing STFD Algorithm: Spatiotemporal Dual-Domain Feature Decoupling: 3D convolution extracts spatial features, bidirectional GRU captures temporal features, and dynamic weighted fusion of satellite and ground data (error <5μg / m³). AGCN Network: Asymmetric graph convolution processing improves modeling accuracy by 40% for UAV flight paths and fixed monitoring station data; Step 3: Digital Twin Simulation WRF-Chem physics model: simulates atmospheric chemical processes and outputs the basic diffusion field (5 minutes / time). GCN-GRU spatiotemporal recursive network: Graph convolution corrects spatial anomalies, GRU remembers short-term contamination events, completes bias correction within 30 seconds, and updates the concentration field in minutes; Step 4: Generation of Dynamic Emission Reduction Strategies MORL Algorithm: Multi-objective reinforcement learning generates Pareto optimal solutions for PM2.5 control and economic costs; Causal Inference Engine: Bayesian network analyzes the correlation between emission anomalies and process parameters. Step 5: Security Warning and Response Dynamic threshold early warning: sliding window statistics (3σ principle) adjust thresholds; pollution source tracing: Bayesian network + GNN locates pollution sources; emergency simulation: connecting to digital twin module to simulate the effect of control measures. Step 6: Accessibility Function Execution Visualization: PM2.5 heat map, diffusion animation 3D display; User interaction: pollution event subscription, early warning push interface; System management: equipment status monitoring and data quality control.

[0029] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for retrieving and dynamically warning urban atmospheric PM2.5 concentration by integrating multi-source remote sensing data, comprising a main control module, a three-dimensional monitoring network module (sky, land, and sea), a multi-source data fusion module, a digital twin simulation module, a dynamic emission reduction strategy generation module, a safety warning module, and an auxiliary module, characterized in that: The main control module provides signal control to the three-dimensional monitoring network module (sky, land, and sea), the multi-source data fusion module, the digital twin simulation module, the dynamic emission reduction strategy generation module, the safety early warning module, and the auxiliary module, and is used to coordinate the operation of each module and data scheduling. The aforementioned three-dimensional monitoring network module, encompassing the sky, land, and sea, enables full-space PM2.5 data collection through satellites, drones, ground sensors, and ocean buoys. The multi-source data fusion module uses a spatiotemporal dual-domain feature decoupling algorithm (STFD) and an asymmetric graph convolutional network (AGCN) to integrate heterogeneous data; The digital twin simulation module combines the WRF-Chem physical model with the GCN-GRU spatiotemporal recursive network to achieve minute-level diffusion simulation. The dynamic emission reduction strategy generation module outputs optimized control schemes based on multi-objective reinforcement learning (MORL) and causal inference engine; The safety early warning module achieves closed-loop early warning through dynamic threshold algorithm, pollution source tracing model and emergency response simulation; The auxiliary module provides data visualization, user interaction, and system management functions.

2. The urban atmospheric PM2.5 concentration inversion and dynamic early warning system integrating multi-source remote sensing data according to claim 1, characterized in that: In the aforementioned three-dimensional monitoring network module: Satellite data inversion was performed using the ENVI 2018 aerosol inversion plugin, which achieved hourly PM2.5 inversion at a resolution of 1km×1km based on MODIS / Sentinel data. Low-altitude UAV monitoring uses the LSTM trajectory optimization algorithm to dynamically plan scanning paths, covering blind spots in urban streets and valleys; The marine monitoring node integrates SO2 / NO2 sensors and lidar to establish a land-sea pollution transmission coefficient model.

3. The urban atmospheric PM2.5 concentration inversion and dynamic early warning system integrating multi-source remote sensing data according to claim 1, characterized in that: In the multi-source data fusion module: The STFD algorithm dynamically adjusts the fusion weights of satellite and ground data through an attention mechanism, resulting in an inversion error of <5μg / m³. AGCN network processes the spatial correlation between UAV flight paths and fixed monitoring station data, improving modeling accuracy by 40%.

4. The urban atmospheric PM2.5 concentration inversion and dynamic early warning system integrating multi-source remote sensing data according to claim 1, characterized in that: In the digital twin inference module: The WRF-Chem physical model simulates atmospheric chemical processes and provides basic diffusion predictions; The AI-enhanced model combines the spatiotemporal recursive networks of GCN and GRU to achieve minute-level inference and updates.

5. The urban atmospheric PM2.5 concentration inversion and dynamic early warning system integrating multi-source remote sensing data according to claim 1, characterized in that: In the dynamic emission reduction strategy generation module: the MORL algorithm generates a Pareto optimal solution between PM2.5 control and economic losses; the causal inference engine analyzes the correlation between emission anomalies and process parameters through a Bayesian network.

6. The urban atmospheric PM2.5 concentration inversion and dynamic early warning system integrating multi-source remote sensing data according to claim 1, characterized in that: The dynamic threshold warning algorithm adjusts the warning threshold based on sliding window statistics (such as the 3σ principle); The pollution source tracing model integrates Bayesian networks and graph neural networks (GNNs) to locate pollution sources; The emergency response simulation is integrated with the digital twin module to simulate the impact of control measures on PM2.5 concentration.

7. The urban atmospheric PM2.5 concentration inversion and dynamic early warning system integrating multi-source remote sensing data according to claim 1, characterized in that: The auxiliary module includes a user interaction module and a system management module; The data visualization module supports 3D displays such as PM2.5 concentration heat maps and pollution diffusion animations; The user interaction module provides public service interfaces such as pollution event subscription and early warning push; The system management module monitors device status, data quality, and manages access permissions.

8. The urban atmospheric PM2.5 concentration inversion and dynamic early warning system integrating multi-source remote sensing data according to claim 1, characterized in that: The main control module adopts a distributed architecture, which supports real-time scheduling of data streams and exception handling between modules.

9. A system for inverting and dynamically warning urban atmospheric PM2.5 concentration by integrating multi-source remote sensing data as described in claim 1, characterized in that: Satellite-to-ground data assimilation can be achieved within 90 seconds using edge computing nodes.

Images