Air pollution prediction system and method based on meteorological data

Abstract

The invention relates to the technical field of air pollution prediction, and provides an air pollution prediction system and method based on meteorological data. The data acquisition and fusion module is used for acquiring meteorological data, air pollutant data and traffic data and carrying out fusion processing on the data; the feature extraction module is used for filling missing data, normalizing the data and extracting features; the pollution prediction module is used for carrying out short-term and long-term prediction on air pollution based on the processed data, and carrying out visual display and air pollution early warning notification on a prediction result; and a model integration module. By integrating data of a weather station, an air pollution monitoring station, a traffic monitoring system and satellite remote sensing equipment, the system realizes comprehensive monitoring of air pollution. The spatial-temporal data fusion unit is used for performing data interpolation and consistency processing, so that the coordination of data from different sources is ensured, more comprehensive and accurate input data is provided, and the reliability of a prediction result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of air pollution prediction, and in particular to an air pollution prediction system and method based on meteorological data. Background Art

[0002] Air pollution is becoming increasingly serious and has become one of the important factors affecting public health and the ecological environment worldwide. Air pollutants have had significant negative impacts on human health, climate change and the natural environment, especially exacerbating health problems such as respiratory diseases and cardiovascular diseases. Therefore, how to effectively monitor and predict air pollution and promptly warn of pollution events has become an important research direction in the current environmental protection field.

[0003] Traditional air pollution prediction methods mainly rely on physical models, such as atmospheric diffusion models and chemical transport models, which usually require complex parameter inputs and expensive computational costs. In addition, these methods have limitations in processing large-scale data and real-time prediction. In recent years, with the development of big data and machine learning technology, data-driven air pollution prediction models have begun to gain attention. Such models are able to use historical and real-time data to predict future air quality conditions by learning patterns in the data.

[0004] However, existing data-based prediction systems often fail to fully utilize multi-source data, such as meteorological data, traffic flow data, and satellite remote sensing data. These data sources contain key information that affects the diffusion and concentration changes of air pollution, but are often not comprehensively considered due to technical and methodological limitations. In addition, existing systems often lack flexibility in model updates and parameter optimization, making it difficult to adapt to environmental changes and the challenges of new data.

[0005] Therefore, in view of the shortcomings of current technology in multi-source data fusion, spatiotemporal feature extraction, and dynamic prediction accuracy optimization, an air pollution prediction system and method based on meteorological data was developed. Summary of the invention

[0006] In view of the deficiencies of the prior art, the present invention provides an air pollution prediction system and method based on meteorological data, which solves the problem that the system often lacks flexibility in model updating and parameter optimization and is difficult to adapt to environmental changes.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: An air pollution prediction system based on meteorological data includes: Data collection and fusion module, used to collect meteorological data, air pollutant data and traffic data, and fuse the data; Feature extraction module, used for missing data filling, data normalization and feature extraction; The pollution prediction module makes short-term and long-term predictions of air pollution based on the processed data, and provides visual display of the prediction results and air pollution warning notifications; Model integration module, which integrates short-term and long-term forecast results and dynamically optimizes the weights and parameters of the forecast model; The error evaluation and optimization module is used to evaluate the prediction error of the model and perform hyperparameter optimization.

[0008] Preferably, the data acquisition and fusion module includes: Meteorological data collection unit, weather stations and meteorological sensors collect meteorological data such as temperature, humidity, wind speed, wind direction, atmospheric pressure, etc. in real time; A pollution data collection unit for obtaining air pollutant concentration data from air pollution monitoring stations; Satellite data acquisition unit, used to obtain large-scale air pollution-related data through satellite remote sensing equipment; The spatiotemporal data fusion unit integrates meteorological data, air pollutant data, traffic data and satellite remote sensing data, and completes and ensures consistency of data through spatiotemporal interpolation methods.

[0009] Preferably, the feature extraction module includes: A missing data filling unit is used to process missing values ​​in the collected meteorological data and air pollutant data; A data normalization unit, used to normalize meteorological data and air pollutant data; The feature extraction unit is used to extract important features from meteorological data and air pollutant data.

[0010] Preferably, the pollution prediction module includes: The short-term prediction submodule is used to make short-term predictions of air pollution based on meteorological data and air pollutant data; The long-term prediction submodule is used to predict the long-term trend of air pollution; Data visualization unit to display air pollution forecast results; A pollution warning unit, used to trigger a warning when the prediction results show that the concentration of air pollutants exceeds a predetermined threshold; The meteorological warning submodule is used to analyze future meteorological conditions based on meteorological data and issue early warnings of severe meteorological conditions that may lead to increased air pollution.

[0011] Preferably, the short-term prediction submodule includes: The convolutional neural network unit captures the dynamic diffusion pattern of pollutants in space and time through spatiotemporal convolution. The spatiotemporal convolution calculation formula is:

[0012] in, For the Layer in time Time, location The convolution output of is the weight of the convolution kernel, is the activation function The gated recurrent unit,performs short-term prediction by capturing the time series dependencies of,meteorological data and air pollutant data.,The state update formula of GRU is;

[0013]

[0014] in, To update the gate, To reset the gate, and is the corresponding weight matrix, is the hidden state at the previous moment, Enter data for the current

[0015]

[0016] in, is a candidate hidden state, is the final hidden state, is the weight matrix, is the activation function.

[0017] Preferably, the long-term prediction submodule includes: The pollutant diffusion simulation unit is used to simulate and predict the diffusion process of pollutants in the environment. This unit uses numerical simulation methods to calculate by establishing a physical model of pollutant diffusion. The model formula is:

[0018] in, Indicates at time and distance The pollutant concentration at Initial concentration, i.e. the concentration of the pollution source at time The concentration at The diffusion coefficient is used to characterize the rate at which pollutants diffuse and is usually determined by environmental conditions and the nature of the pollutants. Distance from pollution source, time.

[0019] Deep forest unit, used for long-term air pollution prediction, deep forest uses a multi-layer random forest structure to model input meteorological data and historical pollutant data.

[0020] Preferably, the model integration module includes: The weather warning submodule includes: A high wind warning unit is used to monitor and predict high wind weather conditions and issue a high wind warning when the wind speed exceeds a predetermined threshold; The sandstorm warning unit is used to detect the possibility of sandstorm weather and issue an early warning when it is detected that a sandstorm is about to arrive; The haze warning unit is used to predict possible haze weather conditions and issue a haze warning when haze weather is predicted to occur.

[0021] Preferably, the model integration module includes: The model integration unit is used to weightedly fuse the output results of the short-term prediction module and the long-term prediction module, and dynamically adjust the weights of each model according to the historical prediction accuracy of the model; The model integration formula is:

[0022] The final prediction result is The output of the short-term prediction module, The output of the long-term prediction module, The dynamic weight of the short-term forecast results represents the contribution of the short-term forecast to the final result. The dynamic weight of the long-term forecast result represents the contribution of the long-term forecast to the final result.

[0023] Dynamic adaptive unit, used to adjust the model's weight coefficients in real time according to the model's performance at different prediction time scales; The online learning module is used to update the model online as new data arrives. The online learning module optimizes the parameters of the model based on the gradient descent method.

[0024] Preferably, the error assessment and optimization module includes: The error evaluation unit is used to evaluate the prediction error of the model and quantify the performance of the model based on multiple evaluation indicators; The hyperparameter optimization unit is used to automatically adjust the hyperparameters of the model so that the model can maintain optimal performance under different data conditions.

[0025] The method for air pollution prediction based on meteorological data comprises the following steps: Collect meteorological data, air pollutant data and traffic data from meteorological stations, air pollution monitoring stations, traffic monitoring systems and satellite remote sensing equipment, and fill missing data, normalize data and extract features from the collected data through data preprocessing and feature extraction modules; S2, predicts future air pollution conditions through the short-term prediction module using a spatiotemporal convolutional neural network (ST-CNN) and a gated recurrent unit (GRU). The ST-CNN captures the pollution diffusion pattern in the spatial and temporal dimensions through convolution operations, and the GRU captures the short-term dependencies in the time series. The long-term prediction module simulates the long-term diffusion process of pollutants and predicts future air pollution trends through a deep forest model; S3, weighted fusion of short-term prediction results and long-term prediction results through the model integration module. The weights of short-term and long-term models are dynamically adjusted according to historical prediction errors to ensure that the prediction accuracy of the model is optimized in different time periods. The online learning module updates the model parameters according to the new data arriving in real time to improve the adaptability of the system; S4, evaluates the prediction results of the model through the error evaluation and optimization module, and uses multiple indicators such as mean square error (MSE), mean absolute error (MAE), and determination coefficient (R²) to evaluate the model performance. The hyperparameter optimization unit automatically adjusts the hyperparameters of the model according to the evaluation results to ensure that the prediction system is always in the best state under different scenarios. It analyzes future meteorological conditions based on meteorological data and warns in advance of severe meteorological conditions that may cause increased air pollution due to strong winds, sandstorms, and haze. When extreme weather events are predicted, the system will issue a meteorological warning in advance to remind relevant departments and users to take protective measures. The air pollution prediction results are displayed to users through the data visualization unit. When the concentration of pollutants exceeds the predetermined threshold, the system will issue an alarm to relevant departments or the public through the pollution warning unit to help respond to air pollution events in a timely manner.

[0026] Working principle: First, the system obtains real-time data from multiple data sources through the data acquisition and fusion module. The meteorological data acquisition unit collects meteorological parameters such as temperature, humidity, wind speed, and wind direction that affect the diffusion of air pollution. The pollutant data acquisition unit obtains air pollutant concentrations (such as PM2.5, PM10, NO2, etc.) from monitoring stations. The satellite data acquisition unit provides a wide range of pollutant distribution information through remote sensing equipment. The traffic data acquisition unit provides information such as traffic flow. The spatiotemporal data fusion unit uses interpolation algorithms (such as Kriging interpolation) to fuse data of different time and space scales to ensure the spatial and temporal consistency of the data. The collected data is preprocessed through the feature extraction module, and missing data is first filled to ensure the integrity of the input data. Then, the future meteorological conditions are analyzed based on the meteorological data, and the severe meteorological conditions that may cause air pollution to increase, such as strong winds, sandstorms, and haze, are warned in advance. The system normalizes meteorological data, pollutant data, etc. to avoid deviations in model training due to differences in feature magnitude. Next, the system extracts key features from the original data, such as temperature, humidity, wind speed and direction, and generates new interactive features to further improve the prediction effect of the model. After data preprocessing, the system performs short-term and long-term predictions through the pollution prediction module. The short-term prediction submodule uses convolutional neural networks (CNN) and gated recurrent units (GRU) to predict air pollution in the next 1 to 2 days. CNN is responsible for capturing the diffusion pattern of pollutants in space and time, and GRU is used to process the time series dependencies in the data. The long-term prediction submodule uses a pollutant diffusion simulation model combined with a deep forest model to simulate the diffusion behavior of pollutants over a longer period of time to predict the long-term trend of air pollution. After the prediction is completed, the error evaluation and optimization module will evaluate the model prediction results, using multiple evaluation indicators such as mean square error (MSE) and mean absolute error (MAE) to calculate the prediction error of the model. Based on the evaluation results, the system automatically adjusts the model's hyperparameters through the hyperparameter optimization module to ensure the best performance of the model under different scenarios and data conditions. This can effectively improve the robustness and generalization ability of the model and reduce the decline in prediction performance during long-term use.

[0027] The present invention provides an air pollution prediction system and method based on meteorological data. It has the following beneficial effects: 1. The present invention integrates data from meteorological stations, air pollution monitoring stations, traffic monitoring systems and satellite remote sensing equipment to achieve comprehensive monitoring of air pollution. The spatiotemporal data fusion unit is used to interpolate and process data consistency, ensuring the coordination of data from different sources and providing more comprehensive and accurate input data, thereby improving the reliability of the prediction results.

[0028] 2. The present invention combines short-term and long-term prediction modules. The system can effectively predict the air pollution situation within one to two days in the short term, and accurately predict the long-term trend through the deep forest and pollutant diffusion model. The short-term prediction uses convolutional neural networks and gated recurrent units to effectively capture dynamic changes in time and space, while the long-term prediction captures the pollution trend over a long period of time by simulating pollutant diffusion and deep learning models, which greatly improves the accuracy of the prediction.

[0029] 3. The model integration module realizes the weighted fusion of short-term and long-term prediction results, and the dynamic adaptive unit adjusts the weight according to the historical error to ensure the accuracy of predictions in different time ranges. At the same time, the online learning module continuously optimizes the system performance by updating the model parameters in real time, so that the system has good adaptability, can quickly respond to changes in environmental data, and improve the accuracy and stability of the overall prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A framework diagram of the system of the present invention; Figure 2 It is a schematic diagram of the data acquisition and fusion module of the present invention; Figure 3 It is a schematic diagram of a feature extraction module of the present invention; Figure 4 This is a schematic diagram of a pollution pretreatment module of the present invention; Figure 5 It is the weather warning submodule of the present invention; Figure 6 It is a schematic diagram of a short-term prediction module of the present invention; Figure 7 It is a schematic diagram of a long-term prediction module of the present invention; Figure 8 It is a schematic diagram of the model integration module of the present invention; Fig. 9 It is a schematic diagram of the error evaluation and optimization module of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Please see attached Figure 1 - Attachment Fig. 9 , an embodiment of the present invention provides an air pollution prediction system based on meteorological data, including: Data collection and fusion module, used to collect meteorological data, air pollutant data and traffic data, and fuse the data; Feature extraction module, used for missing data filling, data normalization and feature extraction; The pollution prediction module makes short-term and long-term predictions of air pollution based on the processed data, and provides visual display of the prediction results and air pollution warning notifications; Model integration module, which integrates short-term and long-term forecast results and dynamically optimizes the weights and parameters of the forecast model; The error evaluation and optimization module is used to evaluate the prediction error of the model and perform hyperparameter optimization.

[0033] The data acquisition and fusion modules include: Meteorological data collection unit, weather stations and meteorological sensors collect meteorological data such as temperature, humidity, wind speed, wind direction, atmospheric pressure, etc. in real time; A pollution data collection unit for obtaining air pollutant concentration data from air pollution monitoring stations; Satellite data acquisition unit, used to obtain large-scale air pollution-related data through satellite remote sensing equipment; The spatiotemporal data fusion unit integrates meteorological data, air pollutant data, traffic data and satellite remote sensing data, and performs data completion and consistency processing through spatiotemporal interpolation methods; Specifically, the meteorological data acquisition unit is used to obtain environmental related meteorological parameters from meteorological stations and meteorological sensors in real time. The collected data include temperature, humidity, wind speed, wind direction, atmospheric pressure, etc. These meteorological parameters have a direct impact on the diffusion and sedimentation of air pollution. The meteorological data acquisition unit communicates with multiple local meteorological stations in real time to ensure the timeliness and accuracy of the data. The collected data is imported into the system database through a standardized interface as input for subsequent prediction models; The pollutant data collection unit obtains real-time pollutant concentration information from air pollution monitoring stations. Pollutant data includes the concentrations of major pollutants such as PM2.5, PM10, SO2, NO2, CO, and O3. The unit can collect pollutant distribution in different areas according to different locations and times. These data are derived from ground stations and transmitted to the central database through the network system. Pollutant data is crucial for short-term and long-term predictions of air quality. The system combines this data with meteorological data for the training and prediction process of the prediction model; The satellite data acquisition unit obtains a wide range of air pollution-related data from satellite equipment through remote sensing technology. In particular, it monitors atmospheric particulate matter such as aerosols and dust, and captures their distribution and diffusion in the atmosphere. This data covers a wide spatial range, which can make up for the local limitations of ground monitoring, provide the system with pollutant concentration distribution data in a large area, and ensure the spatial integrity of the model input data; The spatiotemporal data fusion unit is the core link of the system data processing, responsible for the spatiotemporal fusion of data from different sources such as meteorology, pollutants, satellites, and traffic. Different data sources may have differences in collection frequency, spatial coverage, and resolution. For this reason, the spatiotemporal data fusion unit uses Kriging Interpolation to perform data interpolation processing to ensure data consistency and integrity. The core formula of Kriging interpolation is as follows;

[0034] in, Location to be predicted The predicted value at Known location The observation data at The weighting coefficient represents the Known points to predicted points The weights are based on the distance and correlation between locations (usually determined by the semivariogram). The constraints on the weighting coefficients ensure that the interpolation results are linear and unbiased; Through this method, the spatiotemporal data fusion unit can effectively solve the spatiotemporal inconsistency problem of collected data, fill the spatial gaps in the data, and ensure that meteorological data, pollutant data, and traffic data are analyzed and processed in a unified time and space framework.

[0035] The feature extraction module includes: A missing data filling unit is used to process missing values ​​in the collected meteorological data and air pollutant data; A data normalization unit, used to normalize meteorological data and air pollutant data; A feature extraction unit, used to extract important features from meteorological data and air pollutant data; Specifically, the missing data filling unit is responsible for processing the missing value problem in meteorological data and air pollutant data. Since meteorological monitoring equipment or air pollution monitoring stations may lose some data due to hardware failure or network problems, in order to ensure the integrity of the model input data, the interpolation method is used to fill these missing data; The K nearest neighbor interpolation algorithm is used to process missing values ​​in time series. The missing values ​​are inferred by using the similarity between data. The interpolation formula is as follows;

[0036] in, Indicates The prediction results of missing values, The number of neighboring samples, that is, the number of most similar data points selected, Indicates The set of neighboring samples of missing data points, that is, The closest observation data points, Neighboring samples The actual observed value of Pollutant data and traffic data have different dimensions. Directly using these data for training may cause the model to favor one feature and ignore other important features. Therefore, before feature extraction, the data needs to be normalized to keep the value range of each feature consistent.

[0037] In this embodiment, the data normalization unit uses the Min-Max normalization method to scale all feature values ​​to the interval [0,1] to ensure that the magnitude of the features is consistent during model training. The normalization formula is

[0038] in, The original data value, feature The minimum value of feature The maximum value of The normalized data value ranges from [0,1]; The feature extraction unit is used to construct efficient features that are helpful for prediction from the raw data. Based on factors such as meteorological conditions, pollutant concentrations, and traffic conditions, the feature extraction unit generates a series of key features that can reflect changes in air pollution; In order to better reflect the complex impact of meteorological factors on pollutant diffusion, the feature extraction unit constructs interaction terms between meteorological data. For example, the interaction term between temperature and humidity can reflect the combined effect of air humidity and temperature on the diffusion rate of pollutants, and the interaction term between wind speed and wind direction reflects the impact of airflow on pollutant transmission. The expression of interactive features is:

[0039] in, The interactive feature is a new feature generated by multiplying two original features. and are the original eigenvalues ​​respectively.

[0040] The pollution prediction module includes: The short-term prediction submodule is used to make short-term predictions of air pollution based on meteorological data and air pollutant data; The long-term prediction submodule is used to predict the long-term trend of air pollution; Data visualization unit to display air pollution forecast results; A pollution warning unit, used to trigger a warning when the prediction results show that the concentration of air pollutants exceeds a predetermined threshold; The meteorological warning submodule is used to analyze future meteorological conditions based on meteorological data and to warn in advance of severe meteorological conditions that may lead to increased air pollution; Specifically, the short-term prediction submodule is used to predict air pollution conditions within the next 1 to 2 days. The model mainly uses a combination of convolutional neural networks (CNN) and gated recurrent units (GRU) to comprehensively process meteorological data, pollutant data, and traffic data. The convolutional neural network unit captures the dynamic diffusion pattern of pollutants in space and time through spatiotemporal convolution. The spatiotemporal convolution calculation formula is:

[0041] in, For the Layer in time Time, location The convolution output of is the weight of the convolution kernel, is the activation function; GRU is used to deal with the time series prediction problem of air pollution. GRU is an improved recurrent neural network (RNN) that can effectively capture the time dependency of pollutants and has higher computational efficiency than traditional RNN. GRU controls the flow of information by updating gates and resetting gates to capture short-term changes in pollutant concentrations. The state update formula of GRU is:

[0042]

[0043] in, To update the gate, To reset the gate, and is the corresponding weight matrix, is the hidden state at the previous moment, Enter data for the current

[0044]

[0045] in, is a candidate hidden state, is the final hidden state, is the weight matrix, is the activation function; The long-term forecast uses the pollutant diffusion simulation model and the Deep Forest model to simulate the pollutant diffusion process and predict future pollution trends respectively; The pollutant diffusion simulation unit is used to simulate and predict the diffusion process of pollutants in the environment. This unit uses numerical simulation methods to calculate by establishing a physical model of pollutant diffusion. The model formula is:

[0046] in, Indicates at time and distance The pollutant concentration at Initial concentration, i.e. the concentration of the pollution source at time The concentration at The diffusion coefficient is used to characterize the rate at which pollutants diffuse and is usually determined by environmental conditions and the nature of the pollutants. Distance from pollution source, time; The Deep Forest unit is used to predict the trend of long-term air pollution. Deep Forest is a model based on decision tree ensembles, suitable for modeling complex nonlinear relationships. The model uses a multi-layer cascaded random forest structure to combine historical meteorological data, pollutant concentration data, and traffic data for prediction, and can capture the long-term fluctuation trend of pollutants. The hierarchical structure of Deep Forest can automatically identify complex patterns in the data and provide reliable predictions at different time scales; Meteorological warning submodule: when the meteorological model predicts extreme weather events, the system will issue a meteorological warning in advance to remind relevant departments and users to take protective measures; The gale warning unit is used to monitor and predict gale weather conditions. It conducts real-time analysis based on wind speed and wind direction information in meteorological data, and uses the following formula to predict gale conditions in combination with historical meteorological data and wind speed change trends;

[0047] in, Indicates the past The average wind speed at a time point, For the The actual wind speed at a point in time. Exceeds the set wind speed threshold The system will issue a high wind warning signal. The unit also determines the direction in which wind speed may affect the spread of pollutants based on changes in wind direction. By detecting sudden high wind events, it ensures timely warning before air quality changes. The sandstorm warning unit is used to monitor and predict sandstorm weather. By analyzing multiple meteorological factors, including wind speed and direction, humidity and satellite remote sensing data, a probability prediction system based on the sandstorm model is established. The prediction formula for sandstorms is:

[0048] in, is the probability of sandstorm occurrence, Indicates wind speed, is the wind direction angle, Indicates humidity, is the dust concentration parameter from satellite remote sensing. Function Based on the training model, multiple regression analysis or machine learning methods are used to obtain the possibility of sandstorms. Exceeding the threshold When the system issues a sandstorm warning in advance, relevant agencies can take corresponding emergency measures based on the warning signal; The haze warning unit is used to predict the formation and development of haze weather. This unit conducts a comprehensive analysis based on the humidity, temperature, wind speed and pollutant concentration in the meteorological data, and evaluates the changing trend of the concentration of particles in the air by building a model of haze weather formation conditions. The haze warning uses the following prediction model;

[0049] in, is the haze index, and Respectively represent the concentration of fine particles and coarse particles in the air. Indicates humidity, Indicates temperature, Indicates wind speed, , , is the empirical coefficient, which is used to adjust the influence of each variable. Exceeding the set threshold When the weather is very cold, the system will issue a haze warning and determine the duration and severity of the haze weather based on meteorological conditions.

[0050] The model integration module includes: The model integration unit is used to perform weighted fusion on the output results of the short-term prediction module and the long-term prediction module, and dynamically adjust the weights of each model according to the historical prediction accuracy of the model.

[0051] Dynamic adaptive unit, used to adjust the model's weight coefficients in real time according to the model's performance at different prediction time scales; Online learning module, used to update the model online as new data arrives. The online learning module optimizes the model parameters based on the gradient descent method; Specifically, the model integration unit is mainly responsible for weighted fusion of the output results of the short-term prediction module and the long-term prediction module to improve the overall prediction accuracy. Since the prediction performance of short-term prediction and long-term prediction in different time ranges has its own advantages and disadvantages, it is necessary to use a weighted method to determine the contribution weight of each model based on the historical performance of each model in a specific time period; The model integration formula is:

[0052] The final prediction result is The output of the short-term prediction module, The output of the long-term prediction module, The dynamic weight of the short-term forecast results represents the contribution of the short-term forecast to the final result. The dynamic weight of the long-term forecast results represents the contribution of the long-term forecast to the final result; The dynamic adaptive unit is used to dynamically adjust the weight coefficient after each prediction based on the historical prediction error and model performance of each model. This process automatically assigns model weights by comparing the actual prediction error of each model in the corresponding time period, ensuring that the better performing model receives a higher weight. The specific weight adjustment strategy is as follows; Assume that the errors of the short-term model and the long-term model at the previous moment are and The dynamically adjusted weights can be obtained by Formula calculation;

[0053] Through such weight adjustment, the system can ensure that the short-term prediction model obtains a higher weight in the short-term period when pollution changes rapidly, while the contribution of the long-term model gradually increases in the long-term trend prediction. The dynamic adaptive unit monitors the prediction error in real time and adjusts the weight coefficient of each model in time, so that the model integration process can effectively adapt to different prediction needs; In response to the dynamic changes in air pollution data, in this embodiment, the online learning module is used to update the model parameters in real time. As new data continues to arrive, the model needs to perform incremental learning to maintain prediction accuracy under the latest environmental conditions. The online learning module minimizes the loss function through the gradient descent algorithm, so that the weights and parameters of the model can be automatically adjusted as the data is updated to adapt to the new prediction environment; Assume that the current parameters of the model are The loss function is ,The goal of online learning is to minimize this loss function.,Through the gradient descent method; To update, the formula is as follows;

[0054] in, is the learning rate, is the loss function with respect to the parameter Through this online learning mechanism, the system can continuously update the parameters of the model to ensure that the prediction results of the model can keep up with the data changes, thereby maintaining a high prediction accuracy.

[0055] The error evaluation and optimization module includes; The error evaluation unit is used to evaluate the prediction error of the model and quantify the performance of the model based on multiple evaluation indicators; Hyperparameter optimization unit, used to automatically adjust the model's hyperparameters so that the model can maintain optimal performance under different data conditions; Specifically, the error evaluation unit is responsible for calculating the error based on the model's prediction results and the actual air pollution monitoring data, and measuring the performance of the model through a variety of evaluation indicators. In this embodiment, a variety of common evaluation indicators such as mean square error (MSE), mean absolute error (MAE) and determination coefficient (R²) are used to ensure that the prediction performance of the model is comprehensively evaluated; The mean square error is used to evaluate the deviation between the predicted value and the actual value. The smaller the MSE, the more accurate the model's prediction. Its calculation formula is:

[0056] in, is the predicted value of the model, is the actual monitoring value, is the total number of data. Through this formula, the system can quantify the prediction error and provide a basis for subsequent model optimization; Mean absolute error (MAE); MAE is used to measure the absolute deviation between the predicted value and the actual value. Unlike MSE, it is less sensitive to large errors and better reflects the overall performance of the model. Its calculation formula is:

[0057] Through the calculation of MAE, the system can evaluate the overall error performance of the model in each period and ensure the robustness of the model over a long period of time; Coefficient of determination (R²); The R2 coefficient of determination is used to evaluate the explanatory power of the model. Its value is between 0 and 1. The closer the value is to 1, the better the model fits the data. Its calculation formula is:

[0058] in, is the average value of the actual data. R2 evaluates the model's ability to explain data changes by measuring the deviation of the predicted value relative to the fluctuation of the data itself.

[0059] The method for air pollution prediction based on meteorological data comprises the following steps: S1, collects meteorological data, air pollutant data and traffic data from meteorological stations, air pollution monitoring stations, traffic monitoring systems and satellite remote sensing equipment, and performs missing data filling, data normalization and feature extraction on the collected data through data preprocessing and feature extraction modules; S2, predicts future air pollution using a spatiotemporal convolutional neural network (ST-CNN) and a gated recurrent unit (GRU) through a short-term prediction module. ST-CNN captures pollution diffusion patterns in spatial and temporal dimensions through convolution operations, and GRU captures short-term dependencies in time series. The long-term prediction module simulates the long-term diffusion process of pollutants and predicts future air pollution trends through a deep forest model. S3, weighted fusion of short-term prediction results and long-term prediction results through the model integration module. The weights of short-term and long-term models are dynamically adjusted according to historical prediction errors to ensure that the prediction accuracy of the model is optimized in different time periods. The online learning module updates the model parameters according to the new data arriving in real time to improve the adaptability of the system; S4, evaluates the prediction results of the model through the error evaluation and optimization module, and uses multiple indicators such as mean square error (MSE), mean absolute error (MAE), and determination coefficient (R²) to evaluate the model performance. The hyperparameter optimization unit automatically adjusts the hyperparameters of the model according to the evaluation results to ensure that the prediction system is always in the best state under different scenarios. The air pollution prediction results are displayed to users through the data visualization unit. When the pollutant concentration exceeds the predetermined threshold, the system will issue an alarm to relevant departments or the public through the pollution warning unit to help respond to air pollution incidents in a timely manner.

[0060] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An air pollution prediction system based on meteorological data, characterized in that: include; Data collection and fusion module, used to collect meteorological data, air pollutant data and traffic data, and fuse the data; Feature extraction module, used for missing data filling, data normalization and feature extraction; The pollution prediction module makes short-term and long-term predictions of air pollution based on the processed data, and provides visual display of the prediction results and air pollution warning notifications; Model integration module, which integrates short-term and long-term forecast results and dynamically optimizes the weights and parameters of the forecast model; The error evaluation and optimization module is used to evaluate the prediction error of the model and perform hyperparameter optimization.

2. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The data acquisition and fusion module includes: Meteorological data collection unit, weather stations and meteorological sensors collect meteorological data such as temperature, humidity, wind speed, wind direction, atmospheric pressure, etc. in real time; A pollution data collection unit for obtaining air pollutant concentration data from air pollution monitoring stations; Satellite data acquisition unit, used to obtain large-scale air pollution-related data through satellite remote sensing equipment; The spatiotemporal data fusion unit integrates meteorological data, air pollutant data, traffic data and satellite remote sensing data, and completes and ensures consistency of data through spatiotemporal interpolation methods.

3. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The feature extraction module comprises: A missing data filling unit is used to process missing values ​​in the collected meteorological data and air pollutant data; A data normalization unit, used to normalize meteorological data and air pollutant data; The feature extraction unit is used to extract important features from meteorological data and air pollutant data.

4. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The pollution prediction module includes: The short-term prediction submodule is used to make short-term predictions of air pollution based on meteorological data and air pollutant data; The long-term prediction submodule is used to predict the long-term trend of air pollution; Data visualization unit to display air pollution forecast results; A pollution warning unit, used to trigger a warning when the prediction results show that the concentration of air pollutants exceeds a predetermined threshold; The meteorological warning submodule is used to analyze future meteorological conditions based on meteorological data and issue early warnings of severe meteorological conditions that may lead to increased air pollution.

5. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The short-term prediction submodule includes: The convolutional neural network unit captures the dynamic diffusion pattern of pollutants in space and time through spatiotemporal convolution. The spatiotemporal convolution calculation formula is: in, For the Layer in time Time, location The convolution output of is the weight of the convolution kernel, is the activation function The gated recurrent unit,performs short-term prediction by capturing the time series dependencies of,meteorological data and air pollutant data.,The state update formula of GRU is; in, To update the gate, To reset the gate, and is the corresponding weight matrix, is the hidden state at the previous moment, Enter data for the current in, is a candidate hidden state, is the final hidden state, is the weight matrix, is the activation function.

6. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The long-term prediction submodule includes: The pollutant diffusion simulation unit is used to simulate and predict the diffusion process of pollutants in the environment. This unit uses numerical simulation methods to calculate by establishing a physical model of pollutant diffusion. The model formula is: in, Indicates at time and distance The pollutant concentration at Initial concentration, i.e. the concentration of the pollution source at time The concentration at The diffusion coefficient is used to characterize the rate at which pollutants diffuse and is usually determined by environmental conditions and the nature of the pollutants. Distance from pollution source, Time. Deep forest unit, used for long-term air pollution prediction, deep forest uses a multi-layer random forest structure to model input meteorological data and historical pollutant data.

7. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The weather warning submodule includes: A high wind warning unit is used to monitor and predict high wind weather conditions and issue a high wind warning when the wind speed exceeds a predetermined threshold; The sandstorm warning unit is used to detect the possibility of sandstorm weather and issue an early warning when it is detected that a sandstorm is about to arrive; The haze warning unit is used to predict possible haze weather conditions and issue a haze warning when haze weather is predicted to occur.

8. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The model integration module includes: The model integration unit is used to perform weighted fusion on the output results of the short-term prediction module and the long-term prediction module, and dynamically adjust the weights of each model according to the historical prediction accuracy of the model; The model integration formula is: The final prediction result is The output of the short-term prediction module, The output of the long-term prediction module, The dynamic weight of the short-term forecast results represents the contribution of the short-term forecast to the final result. The dynamic weight of the long-term prediction results represents the contribution of the long-term prediction to the final result. The dynamic adaptive unit is used to adjust the weight coefficient of the model in real time according to the performance of the model at different prediction time scales; The online learning module is used to update the model online as new data arrives. The online learning module optimizes the parameters of the model based on the gradient descent method.

9. The air pollution prediction system based on meteorological data according to claim 1, characterized in that: The error assessment and optimization module includes: The error evaluation unit is used to evaluate the prediction error of the model and quantify the performance of the model based on multiple evaluation indicators; The hyperparameter optimization unit is used to automatically adjust the hyperparameters of the model so that the model can maintain optimal performance under different data conditions.

10. A method for air pollution prediction based on meteorological data, according to the air pollution prediction system based on meteorological data according to any one of claims 1 to 9, characterized in that: The steps include: S1, collects meteorological data, air pollutant data and traffic data from meteorological stations, air pollution monitoring stations, traffic monitoring systems and satellite remote sensing equipment, and performs missing data filling, data normalization and feature extraction on the collected data through data preprocessing and feature extraction modules; S2, predicts future air pollution conditions through the short-term prediction module using a spatiotemporal convolutional neural network (ST-CNN) and a gated recurrent unit (GRU). The ST-CNN captures the pollution diffusion pattern in the spatial and temporal dimensions through convolution operations, and the GRU captures the short-term dependencies in the time series. The long-term prediction module simulates the long-term diffusion process of pollutants and predicts future air pollution trends through a deep forest model; S3, weighted fusion of short-term prediction results and long-term prediction results through the model integration module. The weights of short-term and long-term models are dynamically adjusted according to historical prediction errors to ensure that the prediction accuracy of the model is optimized in different time periods. The online learning module updates the model parameters according to the new data arriving in real time to improve the adaptability of the system; S4, evaluates the prediction results of the model through the error evaluation and optimization module, and uses multiple indicators such as mean square error (MSE), mean absolute error (MAE), and determination coefficient (R²) to evaluate the model performance. The hyperparameter optimization unit automatically adjusts the hyperparameters of the model according to the evaluation results to ensure that the prediction system is always in the best state under different scenarios. It analyzes future meteorological conditions based on meteorological data and warns in advance of severe meteorological conditions that may cause increased air pollution due to strong winds, sandstorms, and haze. When extreme weather events are predicted, the system will issue a meteorological warning in advance to remind relevant departments and users to take protective measures. The air pollution prediction results are displayed to users through the data visualization unit. When the concentration of pollutants exceeds the predetermined threshold, the system will issue an alarm to relevant departments or the public through the pollution warning unit to help respond to air pollution events in a timely manner.

Images