Urban area VOCs space-time distribution reconstruction and early warning method
By building a hybrid CNN-LSTM model and combining multiple data sources, accurate reconstruction and prediction of VOCs distribution maps are achieved, solving the problems of limited coverage and low prediction accuracy of traditional monitoring networks, and supporting real-time monitoring and early warning in urban areas.
Patent Information
- Application Number
- CN202510866558.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-26
- Publication Date
- 2025-10-10
AI Technical Summary
Existing technologies make it difficult to accurately monitor and predict the spatiotemporal distribution of volatile organic compounds (VOCs). Traditional monitoring networks have limited coverage and cannot monitor in real time. Convolutional neural network-based methods have low prediction accuracy and lack a comprehensive understanding of the spatiotemporal development of VOCs.
A regional VOCs reconstruction and prediction model based on a hybrid CNN-LSTM is constructed. By collecting multiple data sources, including satellite data, monitoring station data, and traffic information, a unified data representation is constructed. The hybrid neural network is used to extract temporal and spatial features to achieve accurate reconstruction and prediction of VOCs distribution maps.
It achieves accurate reconstruction and prediction of VOCs distribution maps in urban areas, supports real-time monitoring and early warning, meets practical application needs, and provides a scientific basis for environmental management decision-making.
Smart Images
Figure CN120766801A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of VOCs spatiotemporal distribution reconstruction and early warning, and in particular to a VOCs spatiotemporal distribution reconstruction and early warning method in urban areas. Background Art
[0002] With the continuous development of my country's economy, environmental pollution, leakage, and hazards are becoming increasingly prominent, and the need for emergency monitoring is becoming increasingly urgent. Volatile organic compounds (VOCs) are major atmospheric pollutants that seriously pollute air quality and threaten human health. Accurate monitoring and prediction of the spatial and temporal distribution of VOCs will facilitate the development of pollution control strategies.
[0003] However, the spatiotemporal distribution of VOCs has the characteristics of long time span and wide coverage, making it difficult to accurately monitor them using existing methods. Although traditional ground monitoring networks (ground monitoring stations and portable gas chromatography-mass spectrometry) can provide accurate point data, they are limited in coverage and cannot conduct extensive monitoring.
[0004] In recent years, with the continuous development of computer technology, methods based on convolutional neural networks have been used to collect time series data from monitoring stations for prediction. However, due to the single feature extraction and limited data coverage, they have the disadvantages of limited application scenarios and low prediction accuracy.
[0005] In practice, due to a lack of VOC-related knowledge and models, relying solely on information collected through monitoring networks cannot fully understand the spatiotemporal evolution of VOCs. This inability to capture the complex interplay between the temporal characteristics of air pollution and the dynamic spatial characteristics of cities severely limits the formulation and implementation of pollution prevention and control decisions. Furthermore, information collection delays within monitoring networks make real-time VOC monitoring difficult. Summary of the Invention
[0006] In order to solve the above technical problems, the present invention provides a method for reconstructing and early warning the spatiotemporal distribution of VOCs in urban areas, which realizes the accurate reconstruction and prediction of regional VOCs distribution maps and can meet the needs of current practical applications. It has important economic and social benefits and meets the needs of current practical applications.
[0007] The urban area VOCs spatiotemporal distribution reconstruction and early warning method of the present invention comprises the following steps: S1. Collect and preprocess the original VOCs dataset in urban areas; S2. Based on the pollution type of VOCs in urban areas, the regional dataset is divided into a training set and a validation set in a ratio of 8:2; S3. Build a regional VOCs reconstruction and prediction model based on hybrid CNN-LSTM; S4. Set the initial model parameters, train the constructed model, and obtain the optimal model parameters; S5. Use the overall validation set in step 2 to verify the accuracy of the 10 DeepAir models generated after training, and select the model with the highest validation accuracy as the best model; S6. Deploy the trained regional VOCs reconstruction and prediction model to an urban area. By processing the target area’s air pollutants, weather, traffic and road network, and building geometry regional morphological data, a unified data representation is formed and input into the model. S7. The regional VOCs reconstruction and prediction model is used to generate a complete VOCs concentration distribution reconstruction map in the region, and predict the VOCs spatial distribution after 1 to 4 hours based on the reconstruction map. Users can evaluate the regional pollution status in the current and future period based on the reconstruction results and prediction data, thereby providing a scientific basis for environmental management and decision-making and early warning. By constructing a regional VOCs reconstruction and prediction model based on a hybrid CNN-LSTM and adopting advanced network structure and data processing methods, accurate reconstruction and prediction of the regional VOCs distribution map are achieved, and the spatiotemporal distribution monitoring and early warning of VOCs in urban areas required by current practical applications can be met, which has important economic and social benefits and meets the needs of current practical applications.
[0008] Preferably, in S1, the specific steps of collecting the original urban area VOCs dataset and preprocessing it are as follows: S1.1. Obtain the traffic road network and building geometry and distribution in the target area through satellite maps. Use third-party map software APIs and weather APIs to obtain urban road congestion and regional climate data by time, respectively. Combine fixed monitoring stations with a mobile mass spectrometer to construct time series data on VOCs concentration distribution in the target area. Combine multiple data sources to form raw VOCs data for the urban area. S1.2. Divide the city map into thousands of non-intersecting grids by longitude and latitude using a 100mx100m grid. Each grid is associated with air pollutant data and other data. The data types include individual pollutant concentrations at each time step, climate, traffic congestion and road density, building density, and height regional morphology. Therefore, the input data structure is like an image with n channels, where each pixel in the image data corresponds to a grid on the map, and each channel corresponds to a data type, resulting in a uniformly represented dataset. S1.3. The singular value soft threshold algorithm and Kriging interpolation method are used to fill missing values in the time and space dimensions respectively to reduce the noise caused by missing data, make up for the sparsity of the data, and obtain fixed-form input data to ensure model training. Finally, after time and space interpolation, each input feature is normalized by subtracting its mean and dividing by its standard deviation to generate a complete normalized grid structure dataset.
[0009] Preferably, in S3, the specific steps of constructing a regional VOCs reconstruction and prediction model based on a hybrid CNN-LSTM are: S3.1. Considering that the input data is similar to an image data with n channels, a 1x1 convolution layer is inserted between each adjacent convolution layer to promote the fusion of information from multiple channels and extract the relationship between multiple features. S3.2. The main body of the network uses a hybrid neural network of CNN and LSTM, aiming to simultaneously focus on temporal and spatial features to improve the accuracy of reconstruction and prediction. A deep residual network based on CNN is used to extract spatial features, and then LSTM is used to capture temporal features to improve the ability to extract various features. Finally, a fully connected layer is used to reconstruct and predict the VOC distribution map of the entire area. S3.3. Build a new regional VOCs reconstruction and prediction model through S3.1 and S3.2.
[0010] Preferably, in said S4, the initial model parameters are set and the constructed regional VOCs reconstruction and prediction model is trained in the following specific steps: S4.1. Input the entire training set in S2 into the established regional VOCs reconstruction and prediction model; S4.2. Set the training parameters, including the initial learning rate to 0.1, the optimizer momentum to 0.9, the weight decay to 0.0001, the batch size to 16, and the number of training epochs to 200. S4.3. Save a model after every 20 epochs of training. After 200 epochs of training, 10 models will be obtained, and 10 trained regional VOCs reconstruction and prediction models will be obtained.
[0011] Preferably, in S6, the trained regional VOCs reconstruction and prediction model is deployed to the actual verification area for testing, and the specific steps are as follows: S6.1. Obtain the original VOCs dataset for urban areas using satellite maps and third-party data acquisition; S6.2. Preprocess the data to generate a complete normalized grid structure dataset; S6.3. Input the regional VOCs reconstruction and prediction model into the data on pollutant concentrations, climate, traffic congestion, road density, building density, and height regional morphology at each time step; S6.4. The regional VOCs reconstruction and prediction model will output regional VOCs distribution maps at present and after a certain period of time in the future for management personnel to view.
[0012] Preferably, in S7, the specific steps of the early warning mechanism based on the constructed model are: S7.1. Based on the real-time monitoring data of the target area, a regional VOCs reconstruction and prediction model is used to generate a complete VOCs concentration distribution map at the current moment. This map can intuitively reflect the pollution status of each area within the region and assist management personnel in quickly identifying key areas where pollutant concentrations exceed the standard. S7.2. Using the VOCs concentration distribution map reconstructed within the past 24 hours and the VOCs map at the current moment as the input data window, the regional VOCs reconstruction and prediction model is used to predict the spatial distribution of VOCs 1 to 4 hours later to obtain a future pollution diffusion trend map. Relevant management parties can analyze the pollution diffusion path in a timely manner based on the prediction results and take corresponding prevention, control and early warning measures.
[0013] Preferably, in S2, if the total number of data sets of a certain type of pollutant cannot be divided by 10, the excess data is placed in the training set of that type of pollutant, the training sets of all pollutants are combined together to obtain an overall training set, and the verification sets of all types are combined together to obtain an overall verification set.
[0014] Preferably, in S5, if there are multiple models with high accuracy and the same accuracy, one of the multiple models is randomly selected as the best DeepAir model, the weight of the best model is saved, and the training of the regional VOCs reconstruction and prediction model DeepAir is completed.
[0015] Compared with the existing technology, the beneficial effects of the present invention are: by constructing a regional VOCs reconstruction and prediction model based on a hybrid CNN-LSTM, and adopting advanced network structure and data processing methods, the accurate reconstruction and prediction of the regional VOCs distribution map is achieved, and the spatiotemporal distribution monitoring and early warning of VOCs in urban areas required by current practical applications can be met, which has important economic and social benefits and meets the needs of current practical applications. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 This is a flowchart of the regional VOCs reconstruction and prediction method based on hybrid CNN-LSTM of the present invention; Figure 2This is a structural diagram of the model in the regional VOCs reconstruction and prediction method based on the hybrid CNN-LSTM of the present invention. DETAILED DESCRIPTION
[0017] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.
[0018] Example 1 like Figures 1 to 2 As shown, the urban area VOCs spatiotemporal distribution reconstruction and early warning method of the present invention includes the following steps: S1. Collect and preprocess the original VOCs dataset in urban areas; S2. Based on the pollution type of VOCs in urban areas, the regional dataset is divided into a training set and a validation set in a ratio of 8:2; S3. Build a regional VOCs reconstruction and prediction model based on hybrid CNN-LSTM; S4. Set the initial model parameters, train the constructed model, and obtain the optimal model parameters; S5. Use the overall validation set in step 2 to verify the accuracy of the 10 DeepAir models generated after training, and select the model with the highest validation accuracy as the best model; S6. Deploy the trained regional VOCs reconstruction and prediction model to an urban area. By processing the target area’s air pollutants, weather, traffic and road network, and building geometry regional morphological data, a unified data representation is formed and input into the model. S7. The regional VOCs reconstruction and prediction model is used to generate a complete VOCs concentration distribution reconstruction map within the region and predict the VOCs spatial distribution 1 to 4 hours later based on the reconstruction map. Users can evaluate the regional pollution status in the current and future periods based on the reconstruction results and prediction data, thereby providing a scientific basis for environmental management and decision-making and early warning; In this embodiment, by constructing a regional VOCs reconstruction and prediction model based on a hybrid CNN-LSTM and adopting advanced network structure and data processing methods, accurate reconstruction and prediction of regional VOCs distribution maps are achieved, and the spatiotemporal distribution monitoring and early warning of VOCs in urban areas required by current practical applications can be met, which has important economic and social benefits and meets the needs of current practical applications.
[0019] Example 2 Based on Example 1, the urban area VOCs spatiotemporal distribution reconstruction and early warning method of the present invention, in S1, the urban area VOCs original data set is collected and preprocessed in the following specific steps: S1.1. Obtain the traffic road network and building geometry and distribution in the target area through satellite maps. Use third-party map software APIs and weather APIs to obtain urban road congestion and regional climate data by time, respectively. Combine fixed monitoring stations with a mobile mass spectrometer to construct time series data on VOCs concentration distribution in the target area. Combine multiple data sources to form raw VOCs data for the urban area. S1.2. Divide the city map into thousands of non-intersecting grids by longitude and latitude using a 100mx100m grid. Each grid is associated with air pollutant data and other data. The data types include individual pollutant concentrations at each time step, climate, traffic congestion and road density, building density, and height regional morphology. Therefore, the input data structure is like an image with n channels, where each pixel in the image data corresponds to a grid on the map, and each channel corresponds to a data type, resulting in a uniformly represented dataset. S1.3. Use the singular value soft thresholding algorithm and Kriging interpolation method to fill missing values in the time and space dimensions, respectively, to reduce the noise caused by missing data, compensate for data sparsity, and obtain fixed-form input data to ensure model training. Finally, after temporal and spatial interpolation, each input feature is normalized by subtracting its mean and dividing by its standard deviation, generating a complete normalized grid-structured dataset. In S3, the specific steps of constructing a regional VOCs reconstruction and prediction model based on a hybrid CNN-LSTM are as follows: S3.1. Considering that the input data is similar to an image data with n channels, a 1x1 convolution layer is inserted between each adjacent convolution layer to promote the fusion of information from multiple channels and extract the relationship between multiple features. S3.2. The main body of the network uses a hybrid neural network of CNN and LSTM, aiming to simultaneously focus on temporal and spatial features to improve the accuracy of reconstruction and prediction. A deep residual network based on CNN is used to extract spatial features, and then LSTM is used to capture temporal features to improve the ability to extract various features. Finally, a fully connected layer is used to reconstruct and predict the VOC distribution map of the entire area. S3.3. Build a new regional VOCs reconstruction and prediction model through S3.1 and S3.2; In S4, the initial model parameters are set and the constructed regional VOCs reconstruction and prediction model is trained. The specific steps are as follows: S4.1. Input the entire training set in S2 into the established regional VOCs reconstruction and prediction model; S4.2. Set the training parameters, including the initial learning rate to 0.1, the optimizer momentum to 0.9, the weight decay to 0.0001, the batch size to 16, and the number of training epochs to 200. S4.3. Save a model after every 20 epochs of training. After 200 epochs of training, 10 models will be obtained, and 10 trained regional VOCs reconstruction and prediction models will be obtained. In S6, the trained regional VOCs reconstruction and prediction model is deployed to the actual verification area for testing. The specific steps are as follows: S6.1. Obtain the original VOCs dataset for urban areas using satellite maps and third-party data acquisition; S6.2. Preprocess the data to generate a complete normalized grid structure dataset; S6.3. Input the regional VOCs reconstruction and prediction model into the data on pollutant concentrations, climate, traffic congestion, road density, building density, and height regional morphology at each time step; S6.4. The regional VOCs reconstruction and prediction model will output regional VOCs distribution maps at present and at a certain time in the future for management personnel to review; In S7, the specific steps of the early warning mechanism based on the constructed model are: S7.1. Based on the real-time monitoring data of the target area, a regional VOCs reconstruction and prediction model is used to generate a complete VOCs concentration distribution map at the current moment. This map can intuitively reflect the pollution status of each area within the region and assist management personnel in quickly identifying key areas where pollutant concentrations exceed the standard. S7.2. Using the VOCs concentration distribution map reconstructed within the past 24 hours and the VOCs map at the current moment as the input data window, the regional VOCs reconstruction and prediction model is used to predict the spatial distribution of VOCs 1 to 4 hours later to obtain a future pollution diffusion trend map. Relevant management parties can analyze the pollution diffusion path in a timely manner based on the prediction results and take corresponding prevention, control and early warning measures.
[0020] Example 3 Based on Example 1, the urban area VOCs spatiotemporal distribution reconstruction and early warning method of the present invention, in said S2, if the total number of a certain type of pollutant data set is not divisible by 10, then the excess data is added to the training set of the pollutant of that type, and the training sets of all pollutants are combined together to obtain an overall training set, and the validation sets of all types are combined together to obtain an overall validation set; In S5, if there are multiple models with high accuracy and the same accuracy, one of the multiple models is randomly selected as the best DeepAir model, the weight of the best model is saved, and the training of the regional VOCs reconstruction and prediction model DeepAir is completed.
[0021] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for reconstructing the spatiotemporal distribution of VOCs in urban areas and for early warning, characterized by: The following steps are involved: S1. Collect and preprocess the original VOCs dataset in urban areas; S2. Based on the pollution type of VOCs in urban areas, the regional dataset is divided into a training set and a validation set in a ratio of 8:2; S3. Build a regional VOCs reconstruction and prediction model based on hybrid CNN-LSTM; S4. Set the initial model parameters, train the constructed model, and obtain the optimal model parameters; S5. Use the overall validation set in step 2 to verify the accuracy of the 10 DeepAir models generated after training, and select the model with the highest validation accuracy as the best model; S6. Deploy the trained regional VOCs reconstruction and prediction model to an urban area. By processing the target area’s air pollutants, weather, traffic and road network, and building geometry regional morphological data, a unified data representation is formed and input into the model. S7. The regional VOCs reconstruction and prediction model is used to generate a complete VOCs concentration distribution reconstruction map within the region, and based on the reconstruction map, predict the VOCs spatial distribution 1 to 4 hours later. Users can evaluate the regional pollution status in the current and future period based on the reconstruction results and prediction data, thereby providing a scientific basis for environmental management and decision-making warnings.
2. The urban area VOCs spatiotemporal distribution reconstruction and early warning method according to claim 1, characterized in that: In S1, the specific steps of collecting the original urban area VOCs dataset and preprocessing it are as follows: S1.
1. Obtain the traffic road network and building geometry and distribution in the target area through satellite maps. Use third-party map software APIs and weather APIs to obtain urban road congestion and regional climate data by time, respectively. Combine fixed monitoring stations with a mobile mass spectrometer to construct time series data on VOCs concentration distribution in the target area. Combine multiple data sources to form raw VOCs data for the urban area. S1.
2. Divide the city map into thousands of non-intersecting grids by longitude and latitude using a 100mx100m grid. Each grid is associated with air pollutant data and other data. The data types include individual pollutant concentrations at each time step, climate, traffic congestion and road density, building density, and height regional morphology. Therefore, the input data structure is like an image with n channels, where each pixel in the image data corresponds to a grid on the map, and each channel corresponds to a data type, resulting in a uniformly represented dataset. S1.
3. The singular value soft threshold algorithm and Kriging interpolation method are used to fill missing values in the time and space dimensions respectively to reduce the noise caused by missing data, make up for the sparsity of the data, and obtain fixed-form input data to ensure model training. Finally, after time and space interpolation, each input feature is normalized by subtracting its mean and dividing by its standard deviation to generate a complete normalized grid structure dataset.
3. The urban area VOCs spatiotemporal distribution reconstruction and early warning method according to claim 1, characterized in that: In S3, the specific steps of constructing a regional VOCs reconstruction and prediction model based on a hybrid CNN-LSTM are as follows: S3.
1. Considering that the input data is similar to an image data with n channels, a 1x1 convolution layer is inserted between each adjacent convolution layer to promote the fusion of information from multiple channels and extract the relationship between multiple features. S3.
2. The main body of the network uses a hybrid neural network of CNN and LSTM, aiming to simultaneously focus on temporal and spatial features to improve the accuracy of reconstruction and prediction. A deep residual network based on CNN is used to extract spatial features, and then LSTM is used to capture temporal features to improve the ability to extract various features. Finally, a fully connected layer is used to reconstruct and predict the VOC distribution map of the entire area. S3.
3. Build a new regional VOCs reconstruction and prediction model through S3.1 and S3.
2.
4. The urban area VOCs spatiotemporal distribution reconstruction and early warning method according to claim 1, characterized in that: In S4, the initial model parameters are set and the constructed regional VOCs reconstruction and prediction model is trained. The specific steps are as follows: S4.
1. Input the entire training set in S2 into the established regional VOCs reconstruction and prediction model; S4.
2. Set the training parameters, including the initial learning rate to 0.1, the optimizer momentum to 0.9, the weight decay to 0.0001, the batch size to 16, and the number of training epochs to 200. S4.
3. Save a model after every 20 epochs of training. After 200 epochs of training, 10 models will be obtained, and 10 trained regional VOCs reconstruction and prediction models will be obtained.
5. The urban area VOCs spatiotemporal distribution reconstruction and early warning method according to claim 1, characterized in that: In S6, the trained regional VOCs reconstruction and prediction model is deployed to the actual verification area for testing. The specific steps are as follows: S6.
1. Obtain the original VOCs dataset for urban areas using satellite maps and third-party data acquisition; S6.
2. Preprocess the data to generate a complete normalized grid structure dataset; S6.
3. Input the regional VOCs reconstruction and prediction model into the data on pollutant concentrations, climate, traffic congestion, road density, building density, and height regional morphology at each time step; S6.
4. The regional VOCs reconstruction and prediction model will output regional VOCs distribution maps at present and after a certain period of time in the future for management personnel to view.
6. The urban area VOCs spatiotemporal distribution reconstruction and early warning method according to claim 1, characterized in that: In S7, the specific steps of the early warning mechanism based on the constructed model are: S7.
1. Based on the real-time monitoring data of the target area, a regional VOCs reconstruction and prediction model is used to generate a complete VOCs concentration distribution map at the current moment. This map can intuitively reflect the pollution status of each area within the region and assist management personnel in quickly identifying key areas where pollutant concentrations exceed the standard. S7.
2. Using the VOCs concentration distribution map reconstructed within the past 24 hours and the VOCs map at the current moment as the input data window, the regional VOCs reconstruction and prediction model is used to predict the spatial distribution of VOCs 1 to 4 hours later to obtain a future pollution diffusion trend map. Relevant management parties can analyze the pollution diffusion path in a timely manner based on the prediction results and take corresponding prevention, control and early warning measures.
7. The urban area VOCs spatiotemporal distribution reconstruction and early warning method according to claim 1, characterized in that: In S2, if the total number of data sets of a certain type of pollutant cannot be divided by 10, the excess data is placed in the training set of that type of pollutant, the training sets of all pollutants are combined together to obtain an overall training set, and the validation sets of all types are combined together to obtain an overall validation set.
8. The urban area VOCs spatiotemporal distribution reconstruction and early warning method according to claim 1, characterized in that: In S5, if there are multiple models with high accuracy and the same accuracy, one of the multiple models is randomly selected as the best DeepAir model, the weight of the best model is saved, and the training of the regional VOCs reconstruction and prediction model DeepAir is completed.