Regional air pollutant traceability monitoring and forecasting model and regulation and control method

Abstract

The invention discloses a regional air pollutant traceability monitoring and forecasting model and a regulation and control method. Multi-source data such as meteorology, pollutant concentration and pollution source emission in an area are collected and preprocessed, then a model is constructed and trained based on a machine learning or deep learning algorithm, and performance is improved through feature engineering, model selection and training optimization. The trained model is utilized to forecast pollutant concentration, traceability analysis is carried out, results are visualized, emission reduction targets and regulation and control measures are formulated for a target area according to traceability and forecast simulation data, for example, industrial structure optimization and traffic control measures are implemented for different pollution sources, and the implementation effect is evaluated. Meanwhile, the model is evaluated regularly, the model is updated according to a result, parameters are adjusted, data are increased or an algorithm is improved, so that the accuracy and reliability of the model are continuously improved, accurate monitoring, effective source tracing, accurate forecasting and scientific regulation and control of regional air pollutants are achieved, and air quality improvement is assisted.

Description

Technical Field

[0001] The present invention relates to the field of regional air pollutant source tracing monitoring and forecasting models and regulation, and specifically to a regional air pollutant source tracing monitoring and forecasting model and regulation method. Background Art

[0003] Currently, there are many limitations in air pollutant monitoring. On the one hand, the distribution of monitoring stations is uneven, and there is a lack of monitoring data in remote areas or complex terrain regions, making it difficult to comprehensively understand the true situation of regional pollutants. On the other hand, traditional monitoring methods mainly focus on the monitoring of single pollutants or a few pollutants, and have insufficient ability to co-monitor multiple pollutants, unable to accurately capture the complex interaction relationships between pollutants.

[0004] In terms of source tracing technology, the accuracy and reliability of existing methods need to be improved. Some source tracing methods based on simple diffusion models do not fully consider the dynamic changes of meteorological conditions and the complex emission characteristics of pollution sources, resulting in large deviations in source tracing results. And some source tracing technologies relying on a large amount of prior knowledge have poor adaptability in the face of new pollution sources or complex environmental changes.

[0005] In the field of forecasting models, traditional statistical forecasting models are difficult to accurately depict the spatio-temporal evolution law of pollutants and cannot effectively cope with the uncertainties of meteorological conditions and pollution source emissions. Although emerging machine learning and deep learning models have improved the forecasting accuracy to a certain extent, there are still problems such as overfitting and poor adaptability to small sample data, and it is difficult to obtain a large amount of data required for model training.

[0006] For regulation measures, there is currently a lack of pertinence and flexibility. The source tracing and forecasting results of pollutants are not fully combined, resulting in poor treatment effects. Summary of the Invention

[0007] The purpose of the present invention is to provide a regional air pollutant source tracing monitoring and forecasting model and regulation method to solve the problems raised in the above background art.

[0008] To achieve the above purpose, the present invention provides the following technical solutions: A regional air pollutant source tracing monitoring and forecasting model and regulation method, including the following steps:

[0009] Multi-source data collection and preprocessing: Using various monitoring devices distributed in the region, comprehensively collect meteorological data, including wind speed, wind direction, temperature, humidity, and air pressure; collect atmospheric pollutant concentration data, including but not limited to PM 2.5 、PM 10, SO2, NO2, CO, O3; collect pollution source emission data, detailed emission information of various pollution sources in industry, transportation, and life; collect topographic data, land use data and related environmental data; clean the collected data, remove outliers and erroneous data; use appropriate algorithms to fill missing values; normalize the data through standardized methods to ensure data quality and consistency, and provide a reliable data basis for subsequent analysis;

[0010] Model construction and training: Combine machine learning and deep learning algorithms, convolutional neural networks, and recurrent neural networks to build a comprehensive regional air pollutant source tracing monitoring and forecasting model; perform feature engineering on the collected data, including feature extraction, selection, and combination, to mine potential information in the data; divide the data into training sets, validation sets, and test sets, use the training sets to train the model, optimize the model parameters through cross-validation and gradient descent methods, and continuously adjust the model structure and hyperparameters based on the accuracy and recall performance indicators to improve the model performance;

[0011] Pollutant source tracing and forecasting: Use the trained model to forecast the concentration of air pollutants in the region, combine wind direction, pollution source emission data and topographic information, and accurately identify the main sources of pollutants through reverse tracing algorithms and pollution source contribution analysis; Use geographic information system (GIS) technology to visualize the pollutant source tracing results to intuitively present the source of pollutants and diffusion paths;

[0012] Formulate and implement control measures: formulate targeted pollutant emission reduction measures based on pollutant source tracing results and forecast data; formulate stricter emission standards and optimize production processes for industrial pollution sources; implement traffic control and promote new energy vehicles for traffic pollution sources; strengthen garbage classification and control dust for domestic pollution sources; during the implementation process, continuously track and evaluate the effectiveness of measures, and make timely adjustments and optimizations based on actual conditions to ensure effective improvement of air quality;

[0013] Model evaluation and dynamic update: Regularly use new monitoring data to evaluate the model, and analyze the accuracy and stability of the model in tracing and forecasting; if the evaluation results do not meet the requirements, update the model in a timely manner. By adding new training data, adjusting the model algorithm structure, and optimizing parameter settings, the model performance can be continuously improved to better adapt to the characteristics of regional air quality changes.

[0014] Furthermore, in the multi-source data collection and preprocessing step: In the meteorological data collection part, in addition to conventional meteorological parameters, it also includes collecting meteorological data at different altitude levels, using weather balloons carrying sensors or meteorological radar equipment to obtain meteorological information in the vertical direction, which is used to analyze the impact of the atmospheric vertical structure on pollutant diffusion; for the pollution source emission data, the emission time period and emission mode information of each pollution source are recorded in detail to make subsequent analysis more accurate; in the data cleaning link, a statistical-based outlier detection method, the 3σ criterion, is used, combined with domain knowledge for manual review to ensure the accuracy and reliability of the cleaning results.

[0015] Furthermore, in the model construction and training step: In feature engineering, principal component analysis (PCA) is used to reduce the dimension of high-dimensional data, removing redundant information while retaining the main features; for time series data, the change of pollutant concentration over time, Fourier transform is performed to extract frequency domain features; during model training, the adaptive learning rate optimization algorithm, the Adam algorithm, is used to dynamically adjust the learning rate according to the model training process, accelerating the model convergence speed and preventing overfitting; at the same time, the EarlyStopping mechanism is introduced. When the performance of the model on the validation set no longer improves for multiple consecutive training cycles, the training is stopped to avoid overtraining.

[0016] Furthermore, in the pollutant source tracing and forecasting step: Using the clustering analysis algorithm, the main sources of the identified pollutants are classified, and the pollution sources with similar emission characteristics and influence ranges are grouped into one category to formulate targeted control strategies; in terms of pollutant concentration forecasting, an ensemble learning method is adopted, combining the forecasting results of multiple different models, and the results of the neural network model and the time series model are weighted and fused to improve the accuracy and reliability of the forecasting; and during the visual display, not only the pollutant sources and diffusion paths are displayed, but also real-time meteorological information, the dynamic changes of wind speed and wind direction, are superimposed to make the tracing results more intuitive and analyzable.

[0017] Furthermore, in the control measure formulation and implementation step: For industrial pollution sources, phased emission reduction targets are formulated. Combining the actual production situation of the enterprise, the pollutant emissions are gradually reduced on the premise of ensuring production efficiency; for traffic pollution sources, according to the traffic flow and pollutant emission intensity of different sections, a differentiated traffic control plan is formulated, setting tidal lanes or implementing more stringent traffic restrictions in congested sections; when evaluating the effect of the control measures, in addition to monitoring the change of pollutant concentration, the impact on regional economic development and the convenience of residents' lives is also considered, and comprehensive evaluation is carried out through questionnaire surveys and economic data analysis to achieve the balance of environmental benefits and social benefits.

[0018] Furthermore, in the model evaluation and dynamic update step: The model is comprehensively evaluated using multiple evaluation metrics. In addition to accuracy and recall rate, root mean square error (RMSE) and mean absolute error (MAE) are also introduced to evaluate the prediction accuracy of the model, and the F1 value is used to comprehensively measure the performance of the model in the source tracing of different types of pollutants. When updating the model, if it is found that the model has poor prediction effects on specific types of pollutants or specific regions, the training data for this part is specifically increased. At the same time, transfer learning technology is used to use the model parameters obtained from training in other similar regions or related fields as initialization parameters to accelerate the training and optimization process of the model in the current region.

[0019] Furthermore, in the multi-source data collection link, the Internet of Things technology is used to realize the interconnection of monitoring devices and build a real-time data transmission network to ensure that the collected data can be transmitted to the data processing center in a timely and accurate manner. For some parameters that are difficult to directly measure, such as the components of volatile organic compounds (VOCs), indirect measurement methods based on spectral analysis or mass spectrometry analysis are adopted, and the measurement data is converted into actual pollutant concentration data by establishing corresponding conversion models. At the same time, the monitoring devices are regularly calibrated and maintained to ensure the accuracy and reliability of data collection.

[0020] Furthermore, in the model construction and training process, the interaction relationships between different pollutants are considered, and the chemical reaction mechanism between pollutants is incorporated into the model. When studying the process of nitrogen oxides (NOx) and volatile organic compounds (VOCs) generating ozone (O3) under photochemical reactions, a chemical reaction kinetics model is established and combined with machine learning or deep learning models to more accurately simulate the change of pollutant concentration. In addition, in view of the differences in the diffusion and transformation laws of pollutants under different seasons and meteorological conditions, a multi-mode model system is constructed, and the applicable model is automatically switched according to the actual situation to improve the adaptability and accuracy of the model.

[0021] Furthermore, in the pollutant source tracing and prediction step, satellite remote sensing data is introduced for auxiliary analysis. The sensors carried by satellites are used to obtain the large-scale pollutant distribution information in the region to make up for the problem of insufficient spatial coverage of ground monitoring stations. Through the interpretation and analysis of satellite remote sensing images, the macroscopic distribution characteristics and change trends of pollutants are obtained, which are mutually supplemented and verified with ground monitoring data. At the same time, combined with the spatial analysis function of geographic information system (GIS), comprehensive analysis of topographic and geomorphic features and land use type factors is carried out to further improve the accuracy of pollutant source tracing and prediction, especially the application effect in complex terrain and remote areas.

[0022] Furthermore, during the formulation and implementation of control measures, a public participation mechanism should be established. By developing a mobile application for air quality monitoring and control, real-time air quality information, source tracing results, and the progress of control measures should be released to the public. Public feedback channels should be set up to encourage the public to report illegal emissions and provide clues about pollution sources. At the same time, environmental protection volunteers should be organized to participate in air quality monitoring activities to enhance public environmental awareness, forming a regional air pollutant control model jointly participated by the government, enterprises, and the public, and improving the control efficiency and effect.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] Through multi-source data collection, covering rich information such as meteorology, pollutant concentration, and pollution source emissions, and combined with advanced algorithms for data preprocessing and feature engineering, the present invention can comprehensively and accurately grasp the pollutant situation in the region. By adopting an improved source tracing analysis method, fully considering factors such as meteorological transmission and pollution source emission characteristics, the main sources of pollutants and their contribution degrees can be accurately determined, providing a strong basis for subsequent control. Compared with traditional source tracing technologies, the accuracy and reliability of source tracing are greatly improved.

[0025] By comprehensively applying machine learning and deep learning algorithms to construct a multi-modal forecasting model and combining with the atmospheric diffusion mechanism, the spatio-temporal evolution law of pollutants can be effectively captured. Through the online learning mechanism, the model parameters are dynamically updated, enabling it to adapt to the dynamic changes of meteorological conditions and pollution source emissions, and improving the timeliness and accuracy of forecasting. Compared with traditional forecasting models, it can more accurately predict the future changes in pollutant concentration and the impact range, providing reliable information support for taking preventive measures in advance.

[0026] Based on accurate source tracing and accurate forecasting results, targeted pollutant emission reduction targets and control measures are formulated. For different types of pollution sources, such as industry, transportation, and domestic use, personalized control plans are respectively formulated, which not only improves the control effect but also minimizes the negative impact on economic development to the greatest extent. At the same time, through the effect evaluation and dynamic adjustment during the implementation process, it is ensured that the control measures always maintain the optimal state, achieving the balance between environmental benefits and economic benefits.

[0027] The model is evaluated regularly, and the model is updated in a timely manner according to the evaluation results, such as adjusting model parameters, increasing training data, and improving algorithms. This enables the model to continuously adapt to the regional environmental changes, continuously improve the accuracy and reliability of the model, ensure the effectiveness of source tracing, monitoring, forecasting, and control measures, and provide stable technical support for the long-term improvement of regional air quality.

[0028] The regional air pollutant source tracing monitoring and forecasting model and control method of the present invention have good generality and scalability, and can be applied to regions with different scales and environmental characteristics. Whether it is a city, an industrial park or a remote area, it can provide strong support for local air quality governance and contribute to promoting air pollution prevention and control work in China and even the world. Brief Description of the Drawings

[0029] Figure 1 It is a schematic diagram of the method framework of the present invention. Detailed Embodiments

[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] Please refer to Figure 1 , the present invention aims to provide a regional air pollutant source tracing monitoring and forecasting model and control method to meet the urgent needs of current environmental governance. The following will elaborate on the implementation manner of the invention in detail.

[0032] Data Collection and Preprocessing

[0033] Monitoring Site Layout: In the target area, multiple monitoring sites are reasonably planned according to factors such as topography, population distribution, industrial layout, and traffic flow. For example, monitoring points are set at key locations such as the city center, industrial parks, transportation hubs, and upwind and downwind directions to ensure comprehensive coverage of areas with different pollution characteristics within the region.

[0034] Data Type Collection: Using high-precision monitoring equipment, real-time concentration data of various air pollutants are collected, including but not limited to PM 2.5 , PM 10 , SO2, NO2, CO, O3, etc. For example, the β-ray absorption method is used to monitor the concentration of PM 2.5 and PM 10 , the ultraviolet fluorescence method is used to measure the SO2 concentration, and the chemiluminescence method is used to detect the NO2 concentration, etc.

[0035] Through devices such as weather stations, meteorological satellites, and meteorological radars, rich meteorological data are obtained, covering wind speed, wind direction, temperature, humidity, air pressure, precipitation, solar radiation, etc. For example, wind speed and wind direction can be measured using a three-cup anemometer and a wind vane, temperature and humidity are obtained using a temperature and humidity sensor, and air pressure is measured by a barometer.

[0036] Conduct a detailed investigation and monitoring of various pollution sources within the region, and collect data such as industrial emissions, traffic emissions, and domestic emissions. For industrial pollution sources, record information such as the production process of enterprises, the usage of raw materials, the types and amounts of pollutant emissions, etc.; for traffic emissions, count data such as traffic flow, vehicle type distribution, and motor vehicle exhaust emission standards on different road sections; for domestic emissions, focus on residential energy use, waste incineration treatment, etc.

[0037] Also collect topographic and geomorphic data, land use type data, population density data, etc. These data help analyze the diffusion and transmission laws of pollutants.

[0038] Data preprocessing

[0039] Conduct a preliminary check on the collected data to remove obvious errors and abnormal data points. For example, when the monitoring equipment fails and causes data mutations or exceeds a reasonable range, mark and eliminate these outliers. Judgment can be made by setting reasonable data thresholds. For example, the PM 2.5 concentration generally does not exceed 1000 μg / m 3 under normal circumstances. If data greater than this value appears without special circumstances (such as severe sandstorms), it is regarded as an outlier.

[0040] For data with missing values, use appropriate methods to fill them. For time series data, for example, linear interpolation can be used to estimate the missing values by linearly fitting according to the data at adjacent time points; for spatially distributed data, inverse distance weighted interpolation can be used to fill the missing values based on the data at surrounding monitoring stations.

[0041] To reduce the impact of data noise, use the moving average method or Savitzky-Golay filtering method to smooth the data. For example, the moving average method calculates the average value of data within a certain time window to replace the original data points, making the data curve smoother and better reflecting the true trend of the data.

[0042] Standardize different types of data so that they have a unified dimension and scale. Commonly used standardization methods include Z-score standardization, and the formula is where x is the original data, μ is the mean of the data, and σ is the standard deviation of the data. After standardization, data with different characteristics can be compared and analyzed on the same scale, which helps improve the effect of model training.

[0043] Model construction and training

[0044] Extract valuable features from the collected multi-source data. For example, for meteorological data, extract features such as the daily variation range of wind speed and the stability of wind direction; for pollutant concentration data, calculate the correlation coefficients between different pollutants as new features. In addition, use wavelet transform to decompose time series data to obtain feature components at different frequencies to capture the complex variation patterns of the data.

[0045] Adopt feature selection algorithms to screen out the features that contribute most to model prediction from the numerous extracted features. Common methods include information gain, mutual information, recursive feature elimination, etc. For example, by calculating the information gain between each feature and the target variable, select the features with larger information gain, remove redundant and irrelevant features, thereby reducing the complexity of the model, improving the training efficiency and prediction accuracy of the model.

[0046] Combine different types of features to create new features. For example, combine temperature and humidity in meteorological data into a temperature-humidity index, and combine industrial emissions and traffic flow in pollution source emission data into a comprehensive emission intensity index. These new features can more comprehensively reflect the internal relationships between data and provide richer information for the model.

[0047] Select appropriate machine learning or deep learning algorithms to build a model according to the characteristics of the data and the research purpose.

[0048] It is suitable for small sample and non-linear data classification and regression problems. For the problem of pollutant source tracing, if the number of pollutant source categories is relatively small and the data distribution is relatively complex, SVM can be considered. By selecting an appropriate kernel function (such as the radial basis kernel function), map the low-dimensional data to a high-dimensional space, so as to achieve linear separability of the data, and then determine the source of the pollutant.

[0049] It has good anti-overfitting ability and the ability to process high-dimensional data. When building a pollutant source tracing and forecasting model, the random forest can improve the stability and accuracy of the model by integrating the prediction results of multiple decision trees. For multi-source data containing a large number of features, the random forest can automatically select important features and reduce the workload of feature selection.

[0050] Especially, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) and their variants (such as LSTMs, GRUs) in deep learning have unique advantages in processing spatio-temporal sequence data. CNNs can effectively extract the spatial features of the data and are suitable for analyzing the spatial distribution patterns of pollutants in a region; RNNs and their variants can handle time series data well and capture the changing trends of pollutant concentrations over time for pollutant concentration forecasting. For example, use an LSTM network to learn historical pollutant concentration data to predict the changes in pollutant concentrations in the future for a period of time.

[0051] Divide the preprocessed data into a training set, a validation set, and a test set. Generally, the division is made in the ratio of 70%, 15%, and 15%. The training set is used for training the model, the validation set is used to adjust the hyperparameters of the model, and the test set is used to evaluate the final performance of the model.

[0052] Train the model using a suitable training algorithm. For example, for neural networks, common training algorithms include Stochastic Gradient Descent (SGD) and its variants (such as Adagrad, Adadelta, Adam, etc.). The Adam algorithm combines the advantages of Adagrad and RMSProp, can adaptively adjust the learning rate, accelerate the convergence speed of the model during training, and at the same time avoid the model falling into local optima.

[0053] During the training process, optimize the model through methods such as cross-validation and grid search. Cross-validation can effectively evaluate the generalization ability of the model. For example, using 5-fold cross-validation, divide the training set into 5 subsets, use 4 subsets for training each time, and 1 subset for validation, repeat 5 times, and finally take the average validation result as the performance index of the model. Grid search is used to find the optimal model hyperparameters, such as the number of hidden layer nodes, learning rate, number of iterations, etc. of a neural network. By searching within the preset range of hyperparameters, select the combination of hyperparameters that makes the model performance optimal.

[0054] Pollutant Source Tracing and Forecasting

[0055] Use the trained model to predict the air pollutant concentration in a region for a period of time in the future. For example, for a time series prediction model (such as LSTM), input historical pollutant concentration data, meteorological data, and other relevant features, and the model outputs pollutant concentration prediction values at different time scales such as 1 hour, 3 hours, and 6 hours in the future.

[0056] To evaluate the reliability of the prediction results, perform uncertainty analysis. The Monte Carlo simulation method can be used to randomly sample the training data multiple times for model training and prediction, obtain multiple prediction results, and calculate the confidence interval of the predicted values by analyzing the distribution of these results. For example, calculate the 95% confidence interval to reflect the uncertainty range of the prediction results.

[0057] The main sources of pollutants are determined by using the trained source tracing model in combination with real-time pollutant concentration data, meteorological data and pollution source emission data. For example, based on the positive definite matrix factorization model (PMF), the monitored pollutant concentration data is decomposed into different pollution source contribution factors, and the source of the pollutants is determined by analyzing the correlation between each factor and the known pollution source. In the present invention, the PMF model is improved, and the meteorological transmission matrix is introduced as a constraint condition to more accurately reflect the impact of meteorological conditions on pollutant transmission and improve the accuracy of source tracing.

[0058] In addition to model-based tracing, a comprehensive tracing method is also used. Combined with the wind rose diagram, the dominant wind direction when the pollutant concentration is high is analyzed to preliminarily determine the location of the pollution source; based on the pollution source emission list, the emission characteristics of different pollution sources are compared with the monitored pollutant components to further determine the source of the pollutant. For example, if PM2.5 is detected in a certain area, the source of the pollutant can be determined. 2.5 The organic carbon content in the medium is high, and there are many biomass burning sources in the upwind direction of the area. Combined with the wind direction information, it can be inferred that biomass burning may be the main source of PM in this area. 2.5 An important source of.

[0059] Using geographic information system (GIS) technology, the pollutant source tracing results are visualized on the map. The locations of different pollution sources are marked on the map, and icons of different colors and sizes are used to represent the type and contribution of the pollution source. At the same time, the diffusion path of pollutants is drawn, and the diffusion of pollutants at different times is displayed through dynamic maps, intuitively presenting the transmission process of pollutants from the source to the monitoring point.

[0060] Use bar graphs, pie charts and other charts to show the contribution of different pollution sources to pollutant concentrations. For example, use a bar graph to compare the contribution of industrial sources, traffic sources, and daily life sources to PM. 2.5 The contribution of a certain pollutant (such as NO2) to the concentration is shown in a pie chart, making the traceability results clearer and easier to understand, and convenient for decision makers and researchers to analyze.

[0061] Refer to the pollutant source tracing results and set differentiated emission reduction targets according to the contribution of different pollution sources. For pollution sources with greater contribution, set a higher emission reduction ratio. For example, if industrial sources contribute more to PM 2.5 If the contribution of industrial sources reaches 50%, industrial sources may be required to reduce emissions by 30%-50% within a certain period of time, while for domestic sources with smaller contributions, relatively lower emission reduction targets can be set.

[0062] Encourage enterprises to adopt advanced production processes and cleaner production technologies to reduce pollutant generation. For example, promote new blast furnace ironmaking processes in the steel industry to improve energy utilization efficiency and reduce dust and waste gas emissions; in the chemical industry, adopt green chemical synthesis methods to reduce the generation of toxic and harmful intermediate products.

[0063] Upgrade and transform the end-treatment equipment of existing industrial pollution sources to improve pollutant removal efficiency. For example, install efficient desulfurization, denitrification, and dust removal equipment to ensure the up-to-standard discharge of industrial waste gas. For some enterprises with relatively high emission concentrations, require them to install on-line monitoring equipment to monitor emissions in real time, facilitating the regulatory authorities to discover problems in a timely manner and take measures.

[0064] Eliminate backward production capacity and promote the optimization and upgrading of the industrial structure. For highly polluting and energy-consuming enterprises, through policy guidance and economic means, encourage them to transform or close down. At the same time, encourage the development of high-tech industries and industries with low pollution and low energy consumption to reduce pollutant emissions from the source.

[0065] Through intelligent transportation systems, reasonably plan traffic flow to reduce the idling and congestion time of motor vehicles. For example, implement intelligent control of traffic lights and adjust the signal duration according to real-time road conditions; promote tidal lanes to relieve traffic congestion during peak hours, thereby reducing motor vehicle exhaust emissions.

[0066] Increase the promotion of new energy vehicles, introduce preferential policies such as purchase subsidies and charging facility construction subsidies, and encourage consumers to purchase and use new energy vehicles. At the same time, accelerate the construction of infrastructure such as charging piles and battery swapping stations to improve the convenience of using new energy vehicles.

[0067] Strictly enforce motor vehicle exhaust emission standards and strengthen the exhaust gas detection of in-use vehicles. For vehicles that do not meet the emission standards, require them to be repaired or compulsorily scrapped. Conduct irregular road inspections and spot checks to severely crack down on excessive emissions.

[0068] In rural areas and areas at the urban-rural fringe, promote the replacement of domestic coal with clean energy, such as using clean energy such as natural gas and electricity. At the same time, carry out clean treatment of domestic coal and promote the use of clean-type coal to reduce pollutant emissions during the coal combustion process.

[0069] Strictly manage construction sites, road dust, etc. Require construction sites to set up enclosures, sprinkler dust suppression facilities, and wash construction vehicles to prevent construction dust from spreading; strengthen road cleaning and sanitation, increase the frequency of sprinkling, especially in dry and windy seasons, and intensify the cleaning of urban primary and secondary roads.

[0070] Strengthen the classified treatment and resource utilization of domestic waste and construction waste. Reasonably plan the layout of landfills and incinerators, adopt advanced waste treatment technologies, and reduce the emissions of pollutants such as malodorous gases and dioxins during the waste treatment process.

[0071] Establish a special environmental supervision department to be responsible for supervising and inspecting the implementation of emission reduction measures. Formulate detailed supervision plans and standards, and regularly inspect industrial enterprises, transportation facilities, domestic pollution sources, etc. to ensure the effective implementation of various emission reduction measures.

[0072] For enterprises and individuals who violate the emission reduction regulations, impose severe penalties in accordance with the law. Increase the penalty intensity, raise the cost of violations, and form an effective deterrence. For example, for industrial enterprises with excessive emissions, in addition to imposing fines, they can also be ordered to suspend production for rectification; for high-emission motor vehicles driving illegally, impose fines and point deductions.

[0073] Regularly evaluate the implementation effect of emission reduction measures. By comparing the changes in pollutant concentrations before and after implementation, judge the effectiveness of emission reduction measures. If it is found that the effect of certain measures is not obvious, adjust and optimize them in a timely manner. For example, if the pollutant concentration does not decrease significantly after implementing traffic restrictions in a certain area, further analyze the reasons, such as whether the restricted area is reasonable, whether there is inadequate implementation, etc., and adjust the traffic restriction plan accordingly.

[0074] Model Evaluation and Update

[0075] Adopt a variety of evaluation indicators to comprehensively evaluate the performance of the model. For the pollutant concentration prediction model, commonly used evaluation indicators include Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R 2 ) etc. RMSE reflects the average error degree between the predicted value and the true value, MAE measures the average absolute value of the error between the predicted value and the true value, and R 2 is used to evaluate the goodness of fit of the model to the data. The closer it is to 1, the better the fitting effect of the model. For the pollutant source tracing model, indicators such as accuracy rate, recall rate, and F1 value can be used for evaluation. These indicators can reflect the accuracy and reliability of the model in identifying the sources of pollutants.

[0076] At regular intervals (such as weekly, monthly or quarterly), use newly collected data to evaluate the model. Compare the prediction results of the model with the actual monitoring data and calculate the values of various evaluation indicators. Through regular evaluation, timely discover the changes in the performance of the model so as to take corresponding measures for adjustment.

[0077] If the performance metrics of the model show a decline, but the decline is relatively small, parameter adjustment of the model can be considered. For example, for a neural network model, parameters such as the learning rate and the number of hidden layer nodes can be adjusted. By retraining the model, observe whether the performance metrics are improved. Use optimization algorithms such as gradient descent to find the optimal parameter combination to make the model perform better on new data.

[0078] When it is found that the model is not adapted to some newly emerging pollution situations or environmental changes, collect more relevant data and add it to the training set, and retrain the model. For example, if a large industrial park is newly built in a certain area, resulting in changes in pollutant emissions, the monitoring data around the industrial park and relevant pollution source information can be collected to expand the training data, enabling the model to learn new pollution characteristics and improve the adaptability of the model.

[0079] If the performance of the model drops significantly and cannot be effectively improved through parameter adjustment and data addition, consider improving the model algorithm. For example, upgrade the original simple neural network model to a more complex deep learning model, such as the Transformer model, and use its powerful self-attention mechanism to better capture complex relationships in the data; or improve the existing pollution source analysis model by introducing new constraints or algorithms to improve the accuracy of source tracing.

[0080] Taking an industrial city as an example, the air quality problem in this city has been relatively prominent in recent years, and the main pollutants are PM 2.5 and NO2.

[0081] Data collection and preprocessing: 50 monitoring stations were set up in this city and its surrounding areas, including different areas such as the city center, industrial parks, near traffic arteries, and suburbs. Meteorological data, atmospheric pollutant concentration data, and pollution source emission data were continuously collected for one year. After preprocessing steps such as data cleaning, missing value filling, smoothing, and standardization, a high-quality data set was obtained.

[0082] Model construction and training: Select a model structure that combines a convolutional neural network (CNN) and a long short-term memory network (LSTM) for pollutant source tracing and forecasting. In the feature engineering stage, features such as wind speed, wind direction, temperature, and humidity in the meteorological data, as well as the historical change trends of pollutant concentration data and the correlations between different pollutants, were extracted. Through cross-validation and grid search, the optimal hyperparameters of the model were determined. After several months of training, the model achieved good performance.

[0083] Pollutant source tracing and forecasting: Use the trained model to conduct source tracing analysis of the pollutants in this city and find that industrial sources contribute to PM 2.5The contribution rate reaches 45%, the contribution rate of traffic sources is 30%, and the contribution rate of living sources is 25%; for NO2, the contribution rate of traffic sources is as high as 60%, the contribution rate of industrial sources is 30%, and the remaining 10% comes from living sources and other scattered pollution sources. In terms of pollutant concentration forecasting, the model accurately predicts that after a strong cold air mass passes through, due to the increased wind speed being conducive to pollutant diffusion, the concentrations of urban PM 2.5 and NO2 will significantly decrease within 24 hours; however, under the stable weather conditions after the cold air passes, the pollutant concentrations will gradually rise again. By comparing the prediction results with the actual monitoring data, the RMSE of the PM 2.5 concentration forecast is 15 μg / m 3 , the MAE is 12 μg / m 3 , and R 2 reaches 0.85; the RMSE of the NO2 concentration forecast is 10 μg / m 3 , the MAE is 8 μg / m 3 , and R 2 is 0.88, indicating that the model has a high forecasting accuracy.

[0084] Formulation and implementation of control measures: Based on the source tracing results and forecasting information, targeted control measures were formulated. For industrial sources, key industrial enterprises are required to complete the upgrade of production processes within the next year to ensure that the emissions of particulate matter and nitrogen oxides are reduced by 30% and 25% respectively. At the same time, the supervision of the operation of enterprise pollution control facilities is strengthened, and on-line monitoring equipment is installed to achieve 24-hour real-time monitoring. For traffic sources, on the one hand, the investment in public transportation is increased, 10 new bus lines are added, and the sharing rate of bus trips is improved; on the other hand, the motor vehicle restriction policy is implemented. During the peak hours on weekdays, vehicles with odd and even license plate numbers are restricted respectively to reduce road congestion and exhaust emissions. In the treatment of living sources, the use of clean energy is promoted. It is planned to replace all the coal used by urban residents for living with natural gas or electricity within half a year. At the same time, the dust management of construction sites is strengthened. It is required that construction sites must be equipped with enclosures and sprinkler dust reduction equipment, and transport vehicles must be sealed for transportation. Within half a year after the implementation of the control measures, through comparison, it is found that the average concentration of PM 2.5 in this city has decreased by 18%, and the average concentration of NO2 has decreased by 15%, and the air quality has been significantly improved.

[0085] Model evaluation and update: The model is comprehensively evaluated once every quarter. After the implementation of the control measures, it is found that there are deviations in the prediction of pollutant concentrations in some periods by the model. The main reason is that with the implementation of the emission reduction measures for industrial sources, the emission characteristics of pollution sources have changed, and the model has not adapted in time. In response to this situation, new pollution source emission data after the implementation of the control measures are collected, added to the training set, the model is retrained, and the model parameters are fine-tuned. After the update, the prediction accuracy of the model has been improved, and the PM 2.5The RMSEs of the forecasts for [substance] and NO2 concentrations are respectively reduced to 12 μg / m 3 and 8 μg / m 3 , and the MAEs are respectively reduced to 10 μg / m 3 and 6 μg / m 3 , and the R 2 both increase to above 0.9, which can better support the subsequent pollution prevention and control work.

[0086] Further explore the deep integration of multiple different types of models. In addition to the combination of existing machine learning and deep learning models, the combination of physical models (such as atmospheric chemical transport models) and data-driven models can also be considered. Physical models can accurately describe the processes of pollutant transport, diffusion, and transformation based on the principles of atmospheric science; while data-driven models are good at mining complex non-linear relationships from a large amount of observational data. By integrating the advantages of both, the changing patterns of regional air pollutants can be characterized more comprehensively and accurately, improving the accuracy of source tracing and forecasting. For example, during the forecast of heavy pollution weather, the physical model can provide the transport trend of pollutants in the large-scale space, and the data-driven model can finely simulate the complex changes in local areas. The two complement each other, providing a more reliable basis for decision-making.

[0087] With the development of environmental monitoring technology, the amount of data has grown exponentially. Using big data technologies, such as distributed storage and parallel computing frameworks (such as Hadoop, Spark), can efficiently process massive environmental data, including historical monitoring data, real-time data, and multi-source heterogeneous data from satellite remote sensing, unmanned aerial vehicle monitoring, etc. At the same time, with the powerful computing power of the cloud computing platform, complex tasks such as model training and simulation calculations can be quickly completed, greatly shortening the calculation time and improving the real-time performance and response speed of the model. For example, during large-scale source apportionment, the cloud computing platform can process data from multiple regions simultaneously, accelerating the source tracing analysis process and providing decision support for environmental management departments in a timely manner.

[0088] When formulating control measures, not only environmental benefits should be concerned, but also the dynamic changes of socio-economic factors should be fully considered. Establish an environmental-economic-social comprehensive model, taking into account the regional economic development plan, industrial structure adjustment plan, population growth trend, etc. For example, when formulating emission reduction targets for industrial sources, combined with the economic affordability and development prospects of enterprises, formulate phased and differentiated emission reduction plans to avoid affecting the normal operation of enterprises and regional economic development due to overly strict emission reduction measures. At the same time, according to the changing social demand for environmental quality, dynamically adjust the control strategy to achieve the coordinated progress of environmental quality improvement and socio-economic sustainable development.

Claims

1. A regional air pollutant source tracing, monitoring and forecasting model and regulation method, characterized in that, The following steps are involved: Multi-source data collection and preprocessing: Using various monitoring devices distributed within the region, comprehensively collect meteorological data, covering wind speed, wind direction, temperature, humidity, and air pressure; collect atmospheric pollutant concentration data, including but not limited to PM 2.5 , PM 10 , SO2, NO2, CO, O3; collect pollution source emission data, detailed emission information of various pollution sources in industry, transportation, and life; at the same time, collect topographic and geomorphic data, land use data, and related environmental data; clean the collected data, remove outliers and incorrect data; use appropriate algorithms to fill in missing values; normalize the data through standardization to ensure data quality and consistency, providing a reliable data basis for subsequent analysis; Model construction and training: Combine machine learning and deep learning algorithms, convolutional neural networks, and recurrent neural networks to build a comprehensive regional air pollutant source tracing monitoring and forecasting model; perform feature engineering on the collected data. It includes feature extraction, selection and combination to mine potential information in the data; divide the training set, validation set and test set, use the training set to train the model, optimize the model parameters through cross-validation and gradient descent methods, and continuously adjust the model structure and hyperparameters according to the accuracy and recall performance indicators to improve the model performance; Pollutant source tracing and forecasting: Use the trained model to forecast the concentration of air pollutants in the region, combine wind direction, pollution source emission data and topographic information, and accurately identify the main sources of pollutants through reverse tracing algorithms and pollution source contribution analysis; Use geographic information system (GIS) technology to visualize the pollutant source tracing results to intuitively present the source of pollutants and diffusion paths; Development and implementation of regulatory measures: Based on pollutant source tracing results and forecast data, formulate targeted pollutant emission reduction measures; For industrial pollution sources, formulate stricter emission standards and optimize production processes; for traffic pollution sources, implement traffic control and promote new energy vehicles; for domestic pollution sources, strengthen garbage classification and control dust; during the implementation process, continuously track and evaluate the effectiveness of measures, and make timely adjustments and optimizations based on actual conditions to ensure that air quality is effectively improved; Model evaluation and dynamic update: Regularly use new monitoring data to evaluate the model, and analyze the accuracy and stability of the model in tracing and forecasting; if the evaluation results do not meet the requirements, update the model in a timely manner. By adding new training data, adjusting the model algorithm structure, and optimizing parameter settings, the model performance can be continuously improved to better adapt to the characteristics of regional air quality changes.

2. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that, In the multi-source data collection and preprocessing steps: the meteorological data collection part, in addition to conventional meteorological parameters, also includes collecting meteorological data at different altitudes, using sensors or meteorological radar equipment on meteorological balloons to obtain meteorological information in the vertical direction, which is used to analyze the impact of the vertical structure of the atmosphere on the diffusion of pollutants; for pollution source emission data, the emission time period and emission mode information of each pollution source are recorded in detail to make subsequent analysis more accurate; in the data cleaning stage, a statistically based outlier detection method and the 3σ criterion are used, combined with manual review of domain knowledge to ensure that the cleaning results are accurate and reliable.

3. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that In the model construction and training steps: in feature engineering, principal component analysis PCA is used to reduce the dimensionality of high-dimensional data, removing redundant information while retaining the main features; for time series data, the change of pollutant concentration over time, Fourier transform is performed to extract frequency domain features; when training the model, the adaptive learning rate optimization algorithm Adam algorithm is used to dynamically adjust the learning rate according to the model training process, accelerate the model convergence speed and prevent overfitting; at the same time, the EarlyStopping mechanism is introduced, and when the performance of the model on the validation set does not improve for multiple consecutive training cycles, the training is stopped to avoid overtraining.

4. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that, In the steps of pollutant source tracing and forecasting: Using the clustering analysis algorithm, classify the main sources of pollutants identified, and group the pollution sources with similar emission characteristics and influence ranges into one category, so as to formulate targeted control strategies; In terms of pollutant concentration forecasting, adopt the integrated learning method, combine the forecasting results of multiple different models, and perform weighted fusion on the results of the neural network model and the time series model to improve the accuracy and reliability of the forecasting. And during the visualization display, not only display the pollutant sources and diffusion paths, but also overlay real-time meteorological information and the dynamic changes of wind speed and direction, making the source tracing results more intuitive and analyzable.

5. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that, In the steps of formulating and implementing the control measures: For industrial pollution sources, formulate phased emission reduction targets, and gradually reduce pollutant emissions on the premise of ensuring production benefits in combination with the actual production situation of the enterprise; For traffic pollution sources, formulate differentiated traffic control plans according to the traffic flow and pollutant emission intensity of different sections of the road, set tidal lanes or implement stricter traffic restrictions in congested sections; When evaluating the effectiveness of the control measures, in addition to monitoring the changes in pollutant concentrations, also consider the impact on regional economic development and the convenience of residents' lives, and comprehensively evaluate through methods such as questionnaire surveys and economic data analysis to achieve a balance between environmental benefits and social benefits.

6. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that, In the steps of model evaluation and dynamic update: Use a variety of evaluation indicators to comprehensively evaluate the model. In addition to accuracy and recall rate, the root mean square error RMSE and mean absolute error MAE are also introduced to evaluate the forecasting accuracy of the model, and the F1 value is used to comprehensively measure the performance of the model in tracing different types of pollutants; When updating the model, if it is found that the model has poor prediction effects on specific types of pollutants or specific regions, increase the training data for this part accordingly, and at the same time use transfer learning technology to use the model parameters obtained from training in other similar regions or related fields as initial parameters to accelerate the training and optimization process of the model in the current region.

7. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that, In the multi-source data acquisition link, use the Internet of Things technology to achieve the interconnection of monitoring devices, build a real-time data transmission network, and ensure that the collected data can be transmitted to the data processing center in a timely and accurate manner; For some parameters that are difficult to directly measure, such as the components of volatile organic compounds VOCs, adopt indirect measurement methods based on spectral analysis or mass spectrometry analysis, and establish corresponding conversion models to convert the measurement data into actual pollutant concentration data; At the same time, regularly calibrate and maintain the monitoring devices to ensure the accuracy and reliability of data acquisition.

8. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that During the model construction and training process, the interaction relationships between different pollutants are considered, and the chemical reaction mechanisms between pollutants are incorporated into the model. When studying the process of the formation of ozone O3 from nitrogen oxides NOx and volatile organic compounds VOCs under photochemical reactions, by establishing a chemical reaction kinetics model and combining it with a machine learning or deep learning model, the concentration changes of pollutants can be more accurately simulated. In addition, aiming at the differences in the diffusion and transformation laws of pollutants under different seasons and meteorological conditions, a multi-mode model system is constructed, and the applicable model is automatically switched according to the actual situation to improve the adaptability and accuracy of the model.

9. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that, In the steps of pollutant source tracing and forecasting, satellite remote sensing data is introduced for auxiliary analysis. The sensors carried by satellites are used to obtain the distribution information of pollutants in a large area within the region, making up for the problem of insufficient spatial coverage of ground monitoring stations. Through the interpretation and analysis of satellite remote sensing images, the macroscopic distribution characteristics and change trends of pollutants are obtained, which are mutually supplemented and verified with ground monitoring data. At the same time, combined with the spatial analysis function of the Geographic Information System GIS, comprehensive analysis is carried out on factors such as topography and land use types to further improve the accuracy of pollutant source tracing and forecasting, especially the application effect in complex terrain and remote areas.

10. The regional air pollutant traceability monitoring and forecasting model and regulation method according to claim 1, characterized in that, In the process of formulating and implementing control measures, a public participation mechanism is established. By developing a mobile application for air quality monitoring and control, air quality information, source tracing results, and the progress of control measures are released to the public in real time. Public feedback channels are set up to encourage the public to report illegal emissions and provide clues to pollution sources. At the same time, environmental protection volunteers are organized to participate in air quality monitoring activities to enhance the public's environmental protection awareness, forming a regional air pollutant control model jointly participated by the government, enterprises, and the public, and improving the control efficiency and effect.

Images