Air quality prediction method and device based on narrowband Internet of Things and sequential network
By constructing an air quality data acquisition and transmission module based on narrowband IoT and time-series networks, and combining it with a multi-dimensional time-series neural network model, the problems of accuracy and efficiency in air quality prediction were solved, and efficient and accurate air quality prediction was achieved.
Patent Information
- Application Number
- CN202511111742.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-08
- Publication Date
- 2025-11-21
AI Technical Summary
Existing technologies struggle to accurately predict air quality, especially the changing patterns of air pollutants, resulting in low prediction accuracy and high computational demands, making it difficult to adapt to the complex formation and diffusion mechanisms of air pollutants.
This paper adopts a method based on narrowband Internet of Things (NIoT) and time-series networks. It predicts the air quality index by constructing an air quality data acquisition module, a data transmission module, feature correlation analysis, and a multi-dimensional time-series neural network model. It uses narrowband IoT to transmit air quality data and performs air quality prediction through collaborative reasoning using the multi-dimensional time-series neural network model.
It improves the accuracy and real-time performance of air quality forecasts, simplifies the handling of complex air pollutant diffusion and chemical reactions, and enhances the accuracy and efficiency of the forecasting system.
Smart Images
Figure CN120996269A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of environmental protection technology, specifically to an air quality prediction method and device based on narrowband Internet of Things and time-series networks. Background Technology
[0002] With the rapid development of the global economy, human lifestyles and production methods have profoundly impacted the surrounding environment. The demand for industrial products is increasing, and their production is causing increasingly severe air pollution. Many harmful substances in air pollution have a significant impact on health, and if not avoided in time, can cause irreversible damage. Studies have shown that fine particulate matter (PM2.5) with an aerodynamic diameter of <2.5 micrometers can trigger cardiovascular and respiratory diseases. This is because PM2.5 can penetrate deep into the lungs and enter the bloodstream unfiltered; its free radicals, metal ions, and organic components not only damage the alveolar walls but also disrupt the integrity of the blood-brain barrier.
[0003] Sulfur dioxide, an air pollutant, is acidic and can react with other substances in the air to form tiny sulfite and sulfate particles. When these particles are inhaled, they accumulate in the lungs, contributing significantly to respiratory illnesses, breathing difficulties, and premature death. Once in the respiratory tract, sulfur dioxide, being highly soluble in water, forms corrosive sulfurous, sulfuric, and sulfate compounds on moist mucous membranes, posing a strong toxicity to the respiratory and cardiovascular systems.
[0004] Nitrogen dioxide is also a significant air pollutant. It is readily soluble in water to form nitric acid, which directly irritates the respiratory mucosa, causing sore throat, cough, and wheezing. Long-term exposure may lead to bronchitis, pulmonary edema, and even damage to alveolar structure, reducing lung gas exchange function.
[0005] Air contains numerous pollutants from various sources, including transportation, industrial production, and daily life. A comprehensive assessment of air quality is made by considering six major pollutants: PM2.5, PM10, SO2, NO2, O3, and CO. The degree of air pollution is determined based on the Air Quality Index (AQI). While the AQI can be calculated from real-time monitoring data, air quality forecasting is lacking. Accurate air quality forecasting allows for adjustments to production activities, reducing current pollutant emissions and improving air quality.
[0006] For predicting air pollutants, some scholars have attempted prediction methods based on the physicochemical changes of pollutants and atmospheric diffusion models, such as AERMOD, CTM, and Gaussian models. These models have been used to predict PM2.5 concentrations. However, these models need to consider multiple pollution sources, multiple meteorological and geographical factors, and the pre-PM2.5 concentration variables resulting from the chemical reactions of other pollutants. Numerous explicit and implicit factors influence air quality. Furthermore, the formation mechanisms and diffusion principles of air pollutants are complex and difficult to describe with deterministic models. Compared to data-driven neural network models, these physics-based prediction methods involve higher computational costs and have limited prediction accuracy. Summary of the Invention
[0007] To address the technical problems mentioned in the background section, this invention provides an air quality prediction method based on narrowband Internet of Things (IoT) and time-series networks, employing the following technical solution:
[0008] S1: Construct an air quality data acquisition module, connecting PM2.5, PM10, SO2, NO2, O3, CO pollutant concentration sensors, etc. The main controller sends the collected air quality data to the data analysis platform.
[0009] S2: Construct a data transmission module based on narrowband IoT. The data transmission module based on narrowband IoT includes a Cat1 mobile communication module. The air quality data acquisition module is connected to the Cat1 mobile communication module. The collected air quality data is transmitted through the mobile communication network in the form of TCP packets using a transparent transmission method to send the data to the designated data platform.
[0010] S3: On the data processing platform, Pearson correlation analysis is used to analyze the correlation between air data characteristics and air quality index pollutant concentrations. For each air quality index pollutant concentration, the correlation with the original air quality data characteristics is calculated, and the characteristics with high correlation are used as the input of the pollution concentration prediction model.
[0011] S4: For the concentration prediction of each pollutant related to air quality, a temporal neural network model containing a Long Short-Term Memory (LSTM) network module and a Fully Connected Network (FCN) module was constructed. For the concentration prediction of the six pollutants in S1, six temporal neural network models were constructed and trained with different data to fit the model of the concentration change of the six pollutants. During model inference, a multi-dimensional temporal neural network model collaborative inference method was adopted to solve the multi-objective fitting problem of air quality index prediction.
[0012] S5: The prediction result output module finds the adjacent inflection point in the corresponding classification table based on the predicted output value of each pollutant concentration, calculates the corresponding sub-index value (IAQI) through the linear interpolation formula, and then takes the maximum value in the IAQI as the final AQI value.
[0013] Furthermore, the air quality data acquisition module in S1 also includes:
[0014] S1-1: Construct an air quality data acquisition module, connect the STM32 main controller's GPIO interface to PM2.5, PM10, SO2, NO2, O3, CO pollutant concentration sensors, etc., and perform data communication according to the relevant sensor communication protocols. The sensors periodically send the detection information to the main controller.
[0015] Furthermore, the narrowband IoT-based data transmission module in S2 also includes:
[0016] S2-1: Construct a data transmission path based on narrowband IoT. Connect the STM32 main controller to the Cat1 mobile 4G communication module via a serial port. Configure network channel parameters and basic information on the communication module support platform. Set the 4G communication module data sender as the data processing platform. Use the TCP protocol to send data packets and bind the communication serial port to realize the transparent transmission of data from the STM32 main controller to the 4G communication module.
[0017] Furthermore, the meteorological characteristic correlation analysis module in S3 also includes:
[0018] S3-1: Air quality data feature selection. Meteorological data contains various data formats and attributes. The time-series neural network model mainly processes time-series numerical information. Meteorological data shows the concentrations of PM2.5, PM10, SO2, NO2, O3 and CO, as well as temperature, wind speed and wind direction. The prediction results of the time-series neural network model are the concentration values of PM2.5, PM10, SO2, NO2, O3 and CO.
[0019] S3-2: Feature selection based on Pearson correlation analysis. By performing Pearson correlation analysis between the raw meteorological data and the data of each air quality influencing factor, the raw meteorological data with a high correlation to changes in air quality factors are identified.
[0020]
[0021] Where ρ(X,Y) represents the Pearson correlation coefficient between meteorological characteristic variable X and air quality factor variable Y. The larger the absolute value of the correlation coefficient ρ, the higher the correlation between meteorological characteristic variable X and air quality factor variable Y, indicating that X is an important factor affecting the change of Y. X should be the input characteristic factor of the Y prediction model.
[0022] S3-3: Feature encoding is the process of converting categorical features into numerical features so that the model can process them.
[0023] The following encoding methods were mainly used: fixed numerical encoding and one-hot encoding;
[0024] S3-4: Feature standardization is the process of scaling the values of continuous features to a specific range in order to improve the training efficiency and performance of the model. Standard Scaler is used to standardize continuous features.
[0025] Furthermore, the temporal neural network model in S4 is a temporal neural network model that integrates a long short-term memory network module and a fully connected network module (FCN) to meet the requirements of multiple feature inputs and multiple feature outputs, specifically including the following:
[0026] S4-1: Predicting air quality influencing factors is a multi-input, multi-output, multi-objective optimization problem. The input data is meteorological characteristics, and the output data is the concentration values of pollutants that affect air quality. Multi-output problems cannot be handled by traditional artificial intelligence methods, which can only optimize single-objective loss functions.
[0027] S4-2: A multidimensional temporal neural network model and a multidimensional loss function are used to predict the concentration values of various pollutants affecting air quality. Each dimension of the multidimensional temporal neural network is a temporal neural network model, corresponding to the change pattern of a pollutant concentration. During model training, each temporal neural network model in the multidimensional temporal neural network is trained independently, and the minimum variance function corresponding to that pollutant is used as the optimization function to fit the change pattern of the pollutant concentration.
[0028] S4-3: A multidimensional temporal neural network model using collaborative inference is used to predict air quality pollutant concentrations. This multidimensional model comprises multiple single-objective optimized temporal neural network models described in S4-2. Air quality pollutant concentration prediction is a time-dependent problem involving the prediction of multiple pollutant concentrations. To ensure prediction accuracy across multiple time periods, an iterative approach is used, employing the values predicted by the temporal neural network model as the input for the next time step. In the multidimensional temporal neural network model described in S4-2, the air quality characteristics at time t are input into the single-objective output temporal neural network model for each dimension, with the output value X... t The output values of the time series neural network in other dimensions form the air quality feature prediction vector at time t+1. At the same time, it can be used as the input value of the time series neural network in each dimension of the multidimensional time series neural network model when predicting air quality at time t+2. The multidimensional time series neural network model using collaborative inference can effectively reduce prediction error.
[0029] S4-4: The fully connected network module is a multilayer perceptron (MLP) used for nonlinear transformation and feature learning. Through multiple fully connected layers and activation functions, it enables multi-feature interaction, allowing the model to automatically weight features. The activation function helps the model fit the nonlinear relationship between data.
[0030] Furthermore, S5 uses the forecast results to calculate the air quality index, specifically including the following:
[0031] S5-1: Predicted concentrations C of various pollutants related to the air quality index based on a collaborative inference neural network. i In the interval position of the classification table, when the pollutant concentration C i,j ≤C i ≤C i,j+1 At that time, calculate the pollutant concentration sub-index (IAQI);
[0032] IAQI i =(C i -C i,j (IAQI) i,j+1 -IAQI i,j ) / (C i,j+1 -C i,j )+IAQI i,j
[0033] Among them, IAQI i It is the pollution sub-index of the i-th pollutant;
[0034] C i This is the concentration value of pollutant i;
[0035] C i,j It is in Table 1 and C i The lower value of similar pollutant concentration limits;
[0036] C i,j+1 It is in Table 1 and C i The higher end of similar pollutant concentration limits;
[0037] IAQI i，j It is C in Table 1 i,j The corresponding air quality sub-index;
[0038] IAQI i，j+1 It is C in Table 1 i,j+1 The corresponding air quality sub-index;
[0039] S5-2: After calculating the pollution sub-indices of various pollutant parameters, the maximum value is taken as the Air Quality Index (AQI) for the region.
[0040] Furthermore, the STM32 processor in the air quality data acquisition module is used to connect to PM2.5, PM10, SO2, NO2, O3, CO pollutant concentration sensors, etc., and the STM32 controller is connected to the Cat1 mobile communication module via a serial port.
[0041] This invention offers the following advantages: The device utilizes Cat1 narrowband IoT technology to transmit air quality data collected by remote sensors to a data processing platform via a mobile communication network. It incorporates a multi-dimensional temporal neural network to accurately capture the patterns of air quality data changes, eliminating the need to consider complex atmospheric diffusion and chemical reactions between pollutants, thus improving the accuracy of the prediction system. Furthermore, the theoretical and technological achievements of this algorithm can provide reference and guidance for remote air pollutant prediction and enterprise operating environment monitoring, promoting the safety of enterprise and social operations and supporting the construction of a green, healthy, and safe working and living environment in my country.
[0042] Meeting domain requirements: The proposed air quality prediction method and device based on narrowband Internet of Things and time-series networks can meet the complex requirements of remote air quality data acquisition and transmission and air quality prediction modeling, and are applicable to algorithms in this field.
[0043] Improving recommendation accuracy: This method introduces Cat1 narrowband IoT technology for remote air quality data acquisition and transmission. The data processing platform uses a multi-dimensional temporal neural network model to accurately capture changes in air pollutants, thereby improving the real-time performance and accuracy of air quality prediction.
[0044] It has reference value: The theoretical and technical achievements of this algorithm can provide reference and guidance for remote air pollutant prediction and enterprise operating environment monitoring, promote the safety level of enterprise and social operation, and support the construction of a green, healthy and safe working and living environment in my country. Attached Figure Description
[0045] Figure 1 This is a diagram illustrating the air quality prediction method and device architecture based on narrowband Internet of Things and time-series networks of the present invention.
[0046] Figure 2 This is a diagram of the Cat1 mobile network data transmission module of the present invention;
[0047] Figure 3 This is a graph showing the model training loss function of the present invention.
[0048] Figure 4 This is a comparison chart of the actual AQI and the predicted AQI of this invention. Detailed Implementation
[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0050] Data Acquisition and Transmission: Meteorological sensors are used to collect concentration information of PM2.5, PM10, CO, SO2, NO2, and O3, and then... Figure 3 The mobile network data model shown uses a transparent transmission model to transmit data to a designated network data processing platform as input information for real-time air quality prediction.
[0051] Dataset: The dataset in this paper mainly comes from actual weather and air quality monitoring. Web scraping software was used to crawl historical weather data and air quality pollutant datasets for a specific weather station in a certain region over several consecutive months from meteorological websites such as the China Weather Network and environmental monitoring station websites. The granularity is daily, meeting the requirements of time series models. Typical characteristic variables include: PM2.5, PM10, CO, SO2, NO2, and O3 concentration characteristics. After the raw data was sampled, the Pandas package was used to perform unified preprocessing of the data, including data imputation, outlier handling, and time alignment.
[0052] Experimental environment configuration: The experimental environment configuration for this paper is as follows.
[0053] Table 1 Experimental Environment
[0054]
[0055] When training an LSTM model on data, plotting the relationship between training loss, validation loss, and the number of training epochs makes it easy to observe the model's convergence and generalization ability. For example... Figure 4 As shown, after approximately 20 iterations, the losses on the training and validation sets tend to stabilize, and in subsequent iterations, the curves tend to overlap without significant oscillations or divergence, indicating that the model has converged. Furthermore, the trend shows that the distance between the two curves is very small, indicating that the difference between the training and validation set losses is minimal.
[0056] Depend on Figure 1(Based on real-time and predicted AQI values in the test set) it can be seen that, overall, the temporal neural network model can capture the long-term trend of AQI changes. Within the 1500-7000 dataset, the curves of the predicted and actual values show the same trend, indicating that the temporal network has a good learning ability for AQI change trends. However, within the 1000-2000 dataset, when the actual AQI drops rapidly, the predicted real-time AQI does not react quickly and has a large error. This is because there are sudden increases in natural environmental events, such as strong winds or rainfall leading to increased pollutant emissions, which the model fails to capture, thus failing to learn the non-linear trends of sudden increases and decreases in air quality index. Under high pollution conditions (AQI = 300-400), the peak value of the predicted AQI is lower than the actual AQI value, indicating that the model has certain limitations in predicting sudden events. This is because the training data contains relatively few high-pollution air quality index samples, leading to over-regularization of the model. Further enhancement with high-pollution air sample data could be implemented later. The prediction results for low pollution levels are more stable than those for high pollution levels, indicating that the model's predictive performance is stronger in low-pollution scenarios with a large number of samples, demonstrating a greater ability to learn from the data. Improving prediction accuracy under high pollution conditions requires increasing the sample size and providing more real-time data to enhance model performance.
[0057] In summary, this invention proposes an air quality prediction method and device based on narrowband Internet of Things (IoT) and time-series networks. Previous research has focused on establishing pollutant distribution models based on the physicochemical interactions between atmospheric diffusion and pollutants. These models involve a wide variety of complex physicochemical changes, and atmospheric diffusion is influenced by multiple factors, exhibiting different change patterns, making it difficult to establish a standard model. This invention, based on data-driven artificial intelligence methods combined with a real-time data acquisition system based on IoT, can effectively solve the problem of air pollutant modeling and prediction. The collected air quality data is remotely transmitted to a data platform using Cat1 narrowband IoT technology. On the data processing platform, a time-series neural network is used to construct multiple pollutant concentration models. Combined with the real-time acquired data, a multi-model collaborative inference approach is employed to predict air quality. This device effectively handles the complexity of air quality prediction and improves the predictive accuracy of meteorological systems.
[0058] The theoretical and technological achievements of this algorithm can provide reference and guidance for remote air pollutant prediction and enterprise operating environment monitoring, promote the safety level of enterprise and social operations, and support the construction of a green, healthy and safe working and living environment in my country.
[0059] Table 2 Air Quality Sub-Indices and Corresponding Pollutant Concentration Limits
[0060]
[0061] This invention is simple to operate, convenient to use, and suitable for widespread promotion and application. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An air quality prediction method based on narrowband Internet of Things and time-series networks, characterized in that, The specific steps are as follows: S1: Construct an air quality data acquisition module, connecting PM2.5, PM10, SO2, NO2, O3, CO pollutant concentration sensors, etc. The main controller sends the collected air quality data to the data analysis platform. S2: Construct a data transmission module based on narrowband IoT. The data transmission module based on narrowband IoT includes a Cat1 mobile communication module. The air quality data acquisition module is connected to the Cat1 mobile communication module. The collected air quality data is transmitted through the mobile communication network in the form of TCP packets using a transparent transmission method to send the data to the designated data platform. S3: On the data processing platform, Pearson correlation analysis is used to analyze the correlation between air data characteristics and air quality index pollutant concentrations. For each air quality index pollutant concentration, the correlation with the original air quality data characteristics is calculated, and the characteristics with high correlation are used as the input of the pollution concentration prediction model. S4: For the concentration prediction of each pollutant related to air quality, a temporal neural network model containing a Long Short-Term Memory (LSTM) network module and a Fully Connected Network (FCN) module was constructed. For the concentration prediction of the six pollutants in S1, six temporal neural network models were constructed and trained with different data to fit the model of the concentration change of the six pollutants. During model inference, a multi-dimensional temporal neural network model collaborative inference method was adopted to solve the multi-objective fitting problem of air quality index prediction. S5: The prediction result output module finds the adjacent inflection point in the corresponding classification table based on the predicted output value of each pollutant concentration, calculates the corresponding sub-index value (IAQI) through the linear interpolation formula, and then takes the maximum value in the IAQI as the final AQI value.
2. The air quality prediction method based on narrowband Internet of Things and time-series networks according to claim 1, characterized in that, The air quality data acquisition module in S1 also includes: S1-1: Construct an air quality data acquisition module, connect the STM32 main controller's GPIO interface to PM2.5, PM10, SO2, NO2, O3, CO pollutant concentration sensors, etc., and perform data communication according to the relevant sensor communication protocols. The sensors periodically send the detection information to the main controller.
3. The air quality prediction method based on narrowband Internet of Things and time-series networks according to claim 1, characterized in that, The narrowband IoT-based data transmission module in S2 also includes: S2-1: Construct a data transmission path based on narrowband IoT. Connect the STM32 main controller to the Cat1 mobile 4G communication module via a serial port. Configure network channel parameters and basic information on the communication module support platform. Set the 4G communication module data sender as the data processing platform. Use the TCP protocol to send data packets and bind the communication serial port to realize the transparent transmission of data from the STM32 main controller to the 4G communication module.
4. The air quality prediction method based on narrowband Internet of Things and time-series networks according to claim 1, characterized in that, The meteorological feature correlation analysis module in S3 also includes: S3-1: Air quality data feature selection. Meteorological data contains various data formats and attributes. The time-series neural network model mainly processes time-series numerical information. Meteorological data shows the concentrations of PM2.5, PM10, SO2, NO2, O3 and CO, as well as temperature, wind speed and wind direction. The prediction results of the time-series neural network model are the concentration values of PM2.5, PM10, SO2, NO2, O3 and CO. S3-2: Feature selection based on Pearson correlation analysis. By performing Pearson correlation analysis between the raw meteorological data and the data of each air quality influencing factor, the raw meteorological data with a high correlation to changes in air quality factors are identified. Where ρ(X,Y) represents the Pearson correlation coefficient between meteorological characteristic variable X and air quality factor variable Y. The larger the absolute value of the correlation coefficient ρ, the higher the correlation between meteorological characteristic variable X and air quality factor variable Y, indicating that X is an important factor affecting the change of Y. X should be the input characteristic factor of the Y prediction model. S3-3: Feature encoding is the process of converting categorical features into numerical features so that the model can process them. The following encoding methods were mainly used: fixed numerical encoding and one-hot encoding; S3-4: Feature standardization is the process of scaling the values of continuous features to a specific range in order to improve the training efficiency and performance of the model. Standard Scaler is used to standardize continuous features.
5. The air quality prediction method of narrowband Internet of Things and time-series network according to claim 1, characterized in that, The temporal neural network model in S4 is a temporal neural network model that integrates a long short-term memory network module and a fully connected network module (FCN) to meet the requirements of multiple feature inputs and multiple feature outputs. Specifically, it includes the following: S4-1: Predicting air quality influencing factors is a multi-input, multi-output, multi-objective optimization problem. The input data is meteorological characteristics, and the output data is the concentration values of pollutants that affect air quality. Multi-output problems cannot be handled by traditional artificial intelligence methods, which can only optimize single-objective loss functions. S4-2: A multidimensional temporal neural network model and a multidimensional loss function are used to predict the concentration values of various pollutants affecting air quality. Each dimension of the multidimensional temporal neural network is a temporal neural network model, corresponding to the change pattern of a pollutant concentration. During model training, each temporal neural network model in the multidimensional temporal neural network is trained independently, and the minimum variance function corresponding to that pollutant is used as the optimization function to fit the change pattern of the pollutant concentration. S4-3: A multidimensional temporal neural network model using collaborative inference is used to predict air quality pollutant concentrations. This multidimensional model comprises multiple single-objective optimized temporal neural network models described in S4-2. Air quality pollutant concentration prediction is a time-dependent problem involving the prediction of multiple pollutant concentrations. To ensure prediction accuracy across multiple time periods, an iterative approach is used, employing the values predicted by the temporal neural network model as the input for the next time step. In the multidimensional temporal neural network model described in S4-2, the air quality characteristics at time t are input into the single-objective output temporal neural network model for each dimension, with the output value X... t The output values of the time series neural network in other dimensions form the air quality feature prediction vector at time t+1. At the same time, it can be used as the input value of the time series neural network in each dimension of the multidimensional time series neural network model when predicting air quality at time t+2. The multidimensional time series neural network model using collaborative inference can effectively reduce prediction error. S4-4: The fully connected network module is a multilayer perceptron (MLP) used for nonlinear transformation and feature learning. Through multiple fully connected layers and activation functions, it enables multi-feature interaction, allowing the model to automatically weight features. The activation function helps the model fit the nonlinear relationship between data.
6. The air quality prediction method based on narrowband Internet of Things and time-series networks according to claim 1, characterized in that, S5 uses the prediction results to calculate the air quality index, specifically including the following: S5-1: Predicted concentrations C of various pollutants related to the air quality index based on a collaborative inference neural network. i In the interval position of the classification table, when the pollutant concentration C i,j ≤C i ≤C i,j+1 At that time, calculate the pollutant concentration sub-index (IAQI); IAQI i =(C i -C i,j )(IAQI i,j+1 -IAQI i,j ) / (C i,j+1 -C i,j )+IAQI i,j Among them, IAQI i It is the pollution sub-index of the i-th pollutant; C i This is the concentration value of pollutant i; C i,j It is in Table 1 and C i The lower value of similar pollutant concentration limits; C i,j+1 It is in Table 1 and C i The higher end of similar pollutant concentration limits; IAQI i，j It is C in Table 1 i,j The corresponding air quality sub-index; IAQI i，j+1 It is C in Table 1 i,j+1 The corresponding air quality sub-index; S5-2: After calculating the pollution sub-indices of various pollutant parameters, the maximum value is taken as the Air Quality Index (AQI) for the region.
7. An air quality prediction device based on narrowband Internet of Things and time-series networks, characterized in that, The system includes an STM32 processor in the air quality data acquisition module, which connects to PM2.5, PM10, SO2, NO2, O3, and CO pollutant concentration sensors. The STM32 controller is connected to the Cat1 mobile communication module via a serial port.