A method for predicting time-series concentration of atmospheric pollutants based on fuzzy model support vector machine

Abstract

The application provides an atmospheric pollutant time series concentration prediction method based on a fuzzy model support vector machine, and the method comprises the following steps: generating an initial membership function weighted defuzzifier; obtaining an initial fuzzy granulation data set; substituting into a support vector machine model to initialize training, simultaneously adjusting the hyperparameters of the defuzzifier and the support vector machine, and continuously adjusting and iterating the model to obtain an atmospheric pollutant time series concentration prediction model; and finally obtaining a predicted AQI air pollution level result. The method can not only solve the overfitting problem, realize automatic extraction and optimization of learning samples, and reduce the number of support vectors, but also can automatically construct various pollutant concentration prediction models according to historical data, adapt through an algorithm, and effectively improve the prediction accuracy and practicability.

Description

Technical Field

[0001] The present invention relates to the field of air pollution technology, and in particular to a method for predicting the time series concentration of air pollutants based on a fuzzy model support vector machine. Background Art

[0002] Common atmospheric pollutant prediction methods typically assume that pollutant concentrations are correlated with a series of specific influencing factors, including meteorological conditions and pollution sources. Predictions are based on the correlation between these factors and pollutant concentrations. Specifically, these influencing factors are used as independent variables, and pollutant concentrations as the dependent variable, using regression analysis techniques to predict future changes.

[0003] However, there are many uncertainties when using this method to predict PM2.5 mass concentrations. First, there is still a consensus on the sources of PM2.5 and the factors that influence it. It is generally believed that the main anthropogenic sources of PM2.5 include ground dust, construction dust, coal burning, biomass combustion, motor vehicle emissions, industrial processes, and the formation of secondary pollutants. Its influencing factors mainly include meteorological factors such as wind speed, temperature, and air pressure, as well as geographical factors such as topography. However, in reality, the sources and influencing factors of PM2.5 are far more complex than this, including many factors that have not yet been identified or are difficult to accurately define. Second, even if a relatively complete and accurate system of PM2.5 sources and influencing factors were established, it is currently difficult to obtain complete relevant data to accurately analyze their correlations.

[0004] In the prior art, the prediction methods provided are:

[0005] Artificial Neural Network (ANN), Deep Neural Network (DNN), Support Vector Machine (SVM), Autoregressive Integrated Moving Average (ARIMA), etc. However, the above methods still have some defects.

[0006] For example: (1) ANN and DNN rely on high-dimensional feature input and a large amount of data, and are prone to overfitting for datasets with small data volume and low dimension; in addition, the design of network structure and the selection of hyperparameters are relatively complex, which increases the difficulty of model training.

[0007] (2) SVM has limited effectiveness in dealing with nonlinear problems. This is because it does not have a universal solution for nonlinear fitting and it is sometimes difficult to select a suitable kernel function. At the same time, as the sample size increases, the number of support vectors required for the model will also increase accordingly, which may increase the computational burden.

[0008] (3) The ARIMA model requires that time series data must be stable or can be transformed into a stable state through differentiation, which limits its scope of application; in addition, the ARIMA model cannot effectively capture the nonlinear characteristics in the time series, which may affect the accuracy of the prediction.

[0009] In view of the above problems, it is necessary to provide a method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine to solve the above problems. Summary of the Invention

[0010] The present invention provides a method for predicting the time series concentration of atmospheric pollutants based on a fuzzy model support vector machine. This method can not only solve the overfitting problem, realize the automatic extraction and optimization of learning samples, and reduce the number of support vectors, but also automatically construct various pollutant concentration prediction models based on historical data. Through algorithm adaptation, the accuracy and practicality of the prediction are effectively improved.

[0011] The specific technical solutions are as follows:

[0012] The present invention provides a method for predicting the time series concentration of atmospheric pollutants based on a fuzzy model support vector machine, comprising:

[0013] S1: Collect historical atmospheric pollutant concentration data to obtain continuous time series pollutant concentration data;

[0014] S2: For different AQI air pollution levels, the atmospheric pollutant concentration is divided into different threshold intervals, and the threshold intervals are used as fuzzy sets. These threshold intervals are substituted into the Gaussian fuzzy membership function to generate the initial membership function weighted defuzzifier for each threshold interval of the set.

[0015] S3: Substitute the time series pollutant concentration data of step S1 into the membership function weighted fuzzifier of step S2 to perform fuzzy granulation to obtain an initial fuzzy granulated data set of each pollutant concentration data in the time series;

[0016] S4: The data set is divided into several groups of time-sequential data sets according to the preset time periods. These data sets are used as training and validation data sets, and are substituted into the support vector machine model for initial training. After initial training, the parameters are adjusted and the model is iterated continuously. Finally, a time-series concentration prediction model for atmospheric pollutants is obtained that can predict the AQI air pollution level result at the next moment based on the data of the previous time period.

[0017] S5: Collect the time series pollutant concentration data of the previous time period, input it into the prediction model, and obtain the AQI air pollution level result at the next moment.

[0018] Furthermore, in step S1, after collecting historical atmospheric pollutant concentration data, data preprocessing is first performed to fill in missing values ​​and remove outliers, and then the time series pollutant concentration data after data cleaning is obtained;

[0019] The method for filling missing values ​​is as follows: the gaps in the atmospheric pollutant concentration data are filled according to the median of the pollutant concentration value on the day of the data detection, or according to the mean of the pollutant concentration value in the previous hour;

[0020] The method for eliminating outliers is: using the outlier detection method to detect abnormal values ​​in the historical atmospheric pollutant concentration data, and then filling them according to the median of the pollutant concentration value on the day of the abnormal value, or filling them according to the mean of the pollutant concentration value in the previous hour of the data.

[0021] Furthermore, in step S1, the air pollutants are sulfur dioxide (SO2), nitrogen dioxide (NO2), PM 10 、PM 2.5 , ozone (O3) or carbon monoxide (CO).

[0022] Furthermore, in step S1, the atmospheric pollutant is PM2.5; in step S2, the AQI air pollution level and the threshold ranges corresponding to different AQI levels are: excellent corresponds to 0-50, good corresponds to 51-100, light pollution corresponds to 101-150, moderate pollution corresponds to 151-200, heavy pollution corresponds to 201-300, and severe pollution corresponds to greater than 300;

[0023] The AQI air pollution level and the corresponding threshold range of PM2.5 concentration at different levels are: Excellent corresponds to 0~35μm / m 3 , good corresponds to 35~75μm / m 3 , light pollution corresponds to 75~115μm / m 3 , moderate pollution corresponds to 115~150μm / m 3 , severe pollution corresponds to 150~250μm / m 3 and severe pollution corresponds to greater than 250μm / m 3 .

[0024] Furthermore, in step S2, the formula of the Gaussian fuzzy membership function is as shown in formula (1):

[0025]

[0026] In formula (1), x represents the concentration value of atmospheric pollutants; represents the Gaussian fuzzy membership function; is the center of the membership function, the mean of the interval endpoints is taken, σ is the standard deviation of the membership function, and the initial value is 1 / 6 of the interval difference.

[0027] Furthermore, in step S4, the time period is divided into hours, days, weeks, months or years, the length of the time period is recorded as m, and x represents a vector composed of pollutant concentration data of the length of the time period m; the support vector machine model uses the radial basis function RBF in the nonlinear mapping function as the kernel function;

[0028] The data set substituted into the support vector machine model is

[0029] Among them, k represents the number of sequences of pollutant data, N represents the total number of sequences of pollutant data, and z k represents the vector Z, y, which is composed of the results of the defuzzifier after the input of the kth group of pollutant concentration data obtained by actual monitoring. k represents the pollutant concentration value predicted by the kth group, R represents the real number domain, m represents the dimension of the input x vector, R m represents an m-dimensional real vector space, where m represents the modulus of the input pollutant concentration Z vector;

[0030] The fitting function in the model is

[0031] Among them, k represents the sequence of pollutant data, N represents the total number of sequences of pollutant data, k(z k ,y) is the kernel function, a k represents the kth support vector, and b represents the bias of the support vector machine model.

[0032] In step S4, the objects of parameter adjustment are the penalty parameter C and the kernel function parameter Gamma;

[0033] Among them, the kernel function parameter Gamma is adjusted using formula (2);

[0034] Formula (2) is as follows:

[0035]

[0036] Among them, G represents the Gamma parameter of the kernel function, Gamma is the kernel function parameter of the radial basis function RBF, σ is the reciprocal of the kernel function Gamma, k(z k ,y) represents the Gaussian kernel function, ||z k -y|| is the two-norm distance, which represents the atmospheric pollutant concentration value z kThe Euclidean distance from the predicted atmospheric pollutant concentration value y;

[0037] The parameter adjustment method is:

[0038] (4-1) Set the initial variation range and step size of the penalty parameter C and the kernel function parameter Gamma, and search for the optimal parameters of the penalty parameter C and the kernel function parameter Gamma within a certain range through grid division. Figure 3 ;

[0039] (4-2) Select the smallest penalty parameter C and its corresponding kernel function parameter Gamma that can achieve the highest verification regression accuracy as the optimal parameters;

[0040] (4-3) If there are multiple sets of minimum penalty parameters C and corresponding kernel function parameters Gamma, the first set of penalty parameters C and kernel function parameters Gamma found is selected as the optimal parameters.

[0041] Furthermore, the model iteration method in step S4 is:

[0042] (4s-1) The inverse of the kernel function parameter Gamma obtained after parameter adjustment is used as the σ parameter in the Gaussian fuzzy membership function to reconstruct the iterative weighted defuzzifier;

[0043] (4s-2) Substituting the time series pollutant concentration data of step S1 into the iterative weighted defuzzifier of step (4s-1) for fuzzy granulation, an iterative data set consisting of the time series pollutant concentration data and the corresponding AQI air pollution level is obtained;

[0044] (4s-3) The iterative data set is then divided into several groups of time-sequential data sets according to the preset time periods. These data sets are used as training and validation data sets, and are substituted into the support vector machine model for iterative training to obtain the iterative model and the iterative kernel function parameter Gamma after iterative parameter adjustment.

[0045] (4s-4) Repeat steps (4s-1) to (4s-3) and calculate the threshold θ until the threshold θ is less than the preset value or greater than the preset number of iterations, then stop the iteration; where the threshold i represents the i-th iteration, σ i represents the σ of the i-th iteration, σ i-1 represents σ at the i-1th iteration.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] (1) The present invention combines the advantages of fuzzy logic and support vector machines, and cooperates with model iteration to effectively overcome the "overfitting" problem of traditional neural networks. At the same time, it can also make it easy for people to incorporate prior system knowledge into fuzzy rules, automatically extract typical data from learning samples as final learning samples, and provide the extracted features to the support vector machine for modeling, so that the model can automatically extract data with inherent regularities in the time series as support vectors, reducing the number of model support vectors. At the same time, the adjustment parameters of the support vector machine can provide support for the selection of hyperparameters of the fuzzy logic model.

[0048] (2) Solving the overfitting problem: Traditional neural network models are prone to overfitting when processing small-scale or low-dimensional data; the present invention effectively overcomes the overfitting problem by combining the advantages of fuzzy logic and support vector machines.

[0049] (3) Automatic extraction and optimization of learning samples: This approach automatically identifies and extracts typical data from learning samples as the final learning samples through a fuzzy system, which not only improves learning efficiency but also reduces the complexity of data processing. In addition, this approach provides high-quality input data for support vector machine modeling by extracting representative features.

[0050] (4) Reducing the number of support vectors: By preprocessing and extracting features from data using fuzzy logic, support vector machines can more accurately identify and utilize the inherent patterns in time series data, thereby reducing the number of support vectors required. This not only improves the model's operational efficiency but also reduces computational costs.

[0051] (5) Automatic parameter adjustment and model optimization: In traditional methods, parameter adjustment of support vector machines usually requires a lot of manual intervention, which is time-consuming and inefficient. The model of the present invention supports the selection of hyperparameters through fuzzy logic and realizes automatic adjustment of parameters, thereby reducing the burden of manual parameter adjustment.

[0052] (6) Enhance the adaptive ability of the model: This technical solution can automatically construct various pollutant concentration prediction models based on historical data, and effectively improve the accuracy and practicality of the prediction through algorithm adaptation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of the method for predicting the time series concentration of atmospheric pollutants based on the fuzzy model support vector machine provided in Example 1.

[0054] Figure 2 This is a graph of the Gaussian fuzzy membership function drawn using AQI intervals in Application Example 1.

[0055] Figure 3This figure shows the results of adjusting the parameters of the support vector machine model using the grid method in Application Example 1.

[0056] Figure 4 This is the result chart of the 24-hour concentration meter prediction trend in Application Example 1. DETAILED DESCRIPTION

[0057] The present invention will be further described below in conjunction with specific embodiments. The following are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto.

[0058] Example 1

[0059] A method for predicting time series concentrations of atmospheric pollutants based on a fuzzy model support vector machine, comprising:

[0060] S1: Collect historical atmospheric pollutant concentration data to obtain continuous time series pollutant concentration data;

[0061] In step S1, after collecting historical atmospheric pollutant concentration data, data preprocessing is first performed to fill in missing values ​​and remove outliers, and then the time series pollutant concentration data after data cleaning is obtained;

[0062] The method for filling missing values ​​is as follows: the gaps in the atmospheric pollutant concentration data are filled according to the median of the pollutant concentration value on the day of the data detection, or according to the mean of the pollutant concentration value in the previous hour;

[0063] The method for eliminating outliers is: using the outlier detection method to detect abnormal values ​​in the historical atmospheric pollutant concentration data, and then filling them according to the median of the pollutant concentration value on the day of the abnormal value, or filling them according to the mean of the pollutant concentration value in the previous hour of the data.

[0064] S2: For different AQI air pollution levels, the atmospheric pollutant concentration is divided into different threshold intervals, and the threshold intervals are used as fuzzy sets. These threshold intervals are substituted into the Gaussian fuzzy membership function to generate the initial membership function weighted defuzzifier for each threshold interval of the set.

[0065] If the air pollutant is PM2.5, in step S2, the AQI air pollution level and the threshold range corresponding to the AQI at different levels are:

[0066] Excellent corresponds to 0-50, good corresponds to 51-100, light pollution corresponds to 101-150, moderate pollution corresponds to 151-200, heavy pollution corresponds to 201-300, and severe pollution corresponds to greater than 300;

[0067] The AQI air pollution level and the corresponding threshold range of PM2.5 concentration at different levels are: Excellent corresponds to 0~35μm / m3 , good corresponds to 35~75μm / m 3 , light pollution corresponds to 75~115μm / m 3 , moderate pollution corresponds to 115~150μm / m 3 , severe pollution corresponds to 150~250μm / m 3 and severe pollution corresponds to greater than 250μm / m 3 .

[0068] In step S2, the formula of the Gaussian fuzzy membership function is as shown in formula (1):

[0069]

[0070] In formula (1), x represents the concentration value of atmospheric pollutants; represents the Gaussian fuzzy membership function; is the center of the membership function, the mean of the interval endpoints is taken, σ is the standard deviation of the membership function, and the initial value is 1 / 6 of the interval difference.

[0071] S3: Substitute the time series pollutant concentration data of step S1 into the membership function weighted fuzzifier of step S2 to perform fuzzy granulation to obtain an initial fuzzy granulated data set of each pollutant concentration data in the time series;

[0072] S4: The data set is divided into several groups of time-sequential data sets according to the preset time periods. These data sets are used as training and validation data sets, and are substituted into the support vector machine model for initial training. After initial training, the parameters are adjusted and the model is iterated continuously. Finally, a time-series concentration prediction model for atmospheric pollutants is obtained that can predict the AQI air pollution level result at the next moment based on the data of the previous time period.

[0073] In step S4, the time period is divided into hours, days, weeks, months or years, the length of the time period is recorded as m, and x represents a vector composed of pollutant concentration data of the length of the time period m; the support vector machine model uses the radial basis function RBF in the nonlinear mapping function as the kernel function;

[0074] The data set substituted into the support vector machine model is

[0075] Among them, k represents the number of sequences of pollutant data, N represents the total number of sequences of pollutant data, and z k represents the vector Z, y, which is composed of the results of the defuzzifier after the input of the kth group of pollutant concentration data obtained by actual monitoring. k represents the pollutant concentration value predicted by the kth group, R represents the real number domain, m represents the dimension of the input x vector, R mrepresents an m-dimensional real vector space, where m represents the modulus of the input pollutant concentration Z vector;

[0076] The fitting function in the model is

[0077] Among them, k represents the sequence of pollutant data, N represents the total number of sequences of pollutant data, k(z k ,y) is the kernel function, a k represents the kth support vector, and b represents the bias of the support vector machine model.

[0078] The objects of parameter adjustment are penalty parameter C and kernel function parameter Gamma;

[0079] Among them, the kernel function parameter Gamma is adjusted using formula (2);

[0080] Formula (2) is as follows:

[0081]

[0082] Among them, G represents the Gamma parameter of the kernel function, Gamma is the kernel function parameter of the radial basis function RBF, σ is the reciprocal of the kernel function Gamma, k(z k ,y) represents the Gaussian kernel function, ||z k -y|| is the two-norm distance, which represents the atmospheric pollutant concentration value z k The Euclidean distance from the predicted atmospheric pollutant concentration value y;

[0083] The parameter adjustment method is:

[0084] (4-1) Set the initial variation range and step size of the penalty parameter C and the kernel function parameter Gamma, and search for the optimal parameters of the penalty parameter C and the kernel function parameter Gamma within a certain range through grid division. Figure 3 ;

[0085] (4-2) Select the smallest penalty parameter C and its corresponding kernel function parameter Gamma that can achieve the highest verification regression accuracy as the optimal parameters;

[0086] (4-3) If there are multiple sets of minimum penalty parameters C and corresponding kernel function parameters Gamma, the first set of penalty parameters C and kernel function parameters Gamma found is selected as the optimal parameters.

[0087] The model iteration method is:

[0088] (4s-1) The inverse of the kernel function parameter Gamma obtained after parameter adjustment is used as the σ parameter in the Gaussian fuzzy membership function to reconstruct the iterative weighted defuzzifier;

[0089] (4s-2) Substituting the time series pollutant concentration data of step S1 into the iterative weighted defuzzifier of step (4s-1) for fuzzy granulation, an iterative data set consisting of the time series pollutant concentration data and the corresponding AQI air pollution level is obtained;

[0090] (4s-3) The iterative data set is then divided into several groups of time-sequential data sets according to the preset time periods. These data sets are used as training and validation data sets, and are substituted into the support vector machine model for iterative training to obtain the iterative model and the iterative kernel function parameter Gamma after iterative parameter adjustment.

[0091] (4s-4) Repeat steps (4s-1) to (4s-3) and calculate the threshold θ until the threshold θ is less than a preset value or greater than a preset number of iterations, then stop the iteration;

[0092] Among them, the threshold i represents the i-th iteration, σ i represents the σ of the i-th iteration, σ i-1 represents σ at the i-1th iteration.

[0093] S5 collects the time series pollutant concentration data of the previous time period, inputs it into the prediction model, and obtains the AQI air pollution level result at the next moment.

[0094] Application Example 1

[0095] This application example takes PM2.5 as an example and uses the prediction methods of Example 1 and Comparative Example 1 to make predictions, as follows:

[0096] The study collected hourly average PM2.5 mass concentrations from a monitoring station over a year. Because pollutant concentration data may be missing due to equipment failure, maintenance, or other external factors, addressing these missing values ​​is essential. Considering the temporal volatility and clustering of pollutant concentrations, data preprocessing, such as missing value and outlier handling, was performed.

[0097] Missing value processing: The median of the pollutant concentration value of the monitoring point on the day or the concentration value of the previous hour is used to fill in the missing values.

[0098] Outlier processing: Use the outlier detection method (either the interquartile range (IQR) detection or the percentile method detection) to fill the missing values ​​of the outliers with the median of the pollutant concentration value of the monitoring point on that day or the concentration value of the previous hour.

[0099] The Air Quality Index (AQI) is a nonlinear, dimensionless index that quantitatively describes air quality. The larger the value, the higher the grade and category, and the darker the color, the more severe the air pollution and the greater the health risk. The current AQI is categorized into six levels: excellent, good, lightly polluted, moderately polluted, heavily polluted, and severely polluted, with values ​​corresponding to 0-50, 51-100, 101-150, 151-200, 201-300, and greater than 300, respectively. The initial fuzzy membership function is constructed using the concentration of a pollutant, such as PM2.5, as categorized into values ​​of 0-35, 35-75, 75-115, 115-150, 150-250, and greater than 250 micrograms per cubic meter.

[0100] Taking the PM2.5 grade as an example, its concentration is divided into 0 to 35. The other registered fuzzy functions are processed separately to obtain the initialized Gaussian weighting function. The results are as follows Figure 2 shown.

[0101] The prediction time window is selected according to the specific project requirements. For example, 24 hours, 7 days, 30 days, and 90 days can be used as the window width to form prediction models of different time scales.

[0102] If 24 hours is used as a window width (time period), the number of windows is the original data length divided by 24, rounded to the nearest integer. If 48 hours of data are available, 24 sets of training data can be formed, with x1 representing the first 24 hours of data and y1 representing the 25th hour of data; x2 representing the data from hours 1-25 and y2 representing the 26th hour of data. The number of iterations is determined by the convergence of the model calculation threshold or by the number of iterations exceeding the preset number.

[0103] The final results are expressed as root mean square error. The root mean square error (RMSE) of the parameter adjustment method is 0.045, and the root mean square error (RMSE) of the direct modeling is 0.0716. The final model prediction results and the actual PM2.5 results are shown in Figure 4 , it can be found that the model can fit the PM2.5 concentration value and trend very well.

Claims

1. A method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine, characterized in that: include: S1: Collect historical atmospheric pollutant concentration data to obtain continuous time series pollutant concentration data; S2: For different AQI air pollution levels, the atmospheric pollutant concentration is divided into different threshold intervals, and the threshold intervals are used as fuzzy sets. These threshold intervals are substituted into the Gaussian fuzzy membership function to generate the initial membership function weighted defuzzifier for each threshold interval of the set. S3: Substitute the time series pollutant concentration data of step S1 into the membership function weighted fuzzifier of step S2 to perform fuzzy granulation to obtain an initial fuzzy granulated data set of each pollutant concentration data in the time series; S4: The data set is divided into several groups of time-sequential data sets according to the preset time periods. These data sets are used as training and validation data sets, and are substituted into the support vector machine model for initial training. After initial training, parameter adjustment and model iteration are performed to finally obtain an atmospheric pollutant time series concentration prediction model that can predict the AQI air pollution level result at the next moment based on the data of the previous time period; S5: Collect the time series pollutant concentration data of the previous time period, input it into the prediction model, and obtain the AQI air pollution level result at the next moment.

2. The method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine according to claim 1, characterized in that: In step S1, after collecting historical atmospheric pollutant concentration data, data preprocessing is first performed to fill in missing values ​​and remove outliers, and then the time series pollutant concentration data after data cleaning is obtained; The method for filling missing values ​​is as follows: the gaps in the atmospheric pollutant concentration data are filled according to the median of the pollutant concentration value on the day of the data detection, or according to the mean of the pollutant concentration value in the previous hour; The method for eliminating outliers is: using the outlier detection method to detect abnormal values ​​in the historical atmospheric pollutant concentration data, and then filling them according to the median of the pollutant concentration value on the day of the abnormal value, or filling them according to the mean of the pollutant concentration value in the previous hour of the data.

3. The method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine according to claim 1, characterized in that: In step S1, the air pollutants are sulfur dioxide, nitrogen dioxide, PM 10 、PM 2.5 , ozone or carbon monoxide.

4. The method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine according to claim 1, characterized in that: In step S1, the air pollutant is PM2.5; in step S2, the AQI air pollution level and the threshold ranges corresponding to different AQI levels are: excellent corresponds to 0-50, good corresponds to 51-100, light pollution corresponds to 101-150, moderate pollution corresponds to 151-200, heavy pollution corresponds to 201-300, and severe pollution corresponds to greater than 300; The AQI air pollution level and the corresponding threshold range of PM2.5 concentration at different levels are: Excellent corresponds to 0~35μm / m 3 , good corresponds to 35~75μm / m 3 , light pollution corresponds to 75~115μm / m 3 , moderate pollution corresponds to 115~150μm / m 3 , severe pollution corresponds to 150~250μm / m 3 and severe pollution corresponds to greater than 250μm / m 3 .

5. The method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine according to claim 1, characterized in that: In step S2, the formula of the Gaussian fuzzy membership function is as shown in formula (1): In formula (1), x represents the concentration value of atmospheric pollutants; represents the Gaussian fuzzy membership function; is the center of the membership function, the mean of the interval endpoints is taken, σ is the standard deviation of the membership function, and the initial value is 1 / 6 of the interval difference.

6. The method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine according to claim 1, characterized in that: In step S4, the time period is divided into hours, days, weeks, months or years, the length of the time period is recorded as m, and x represents a vector composed of pollutant concentration data of the length of the time period m; the support vector machine model uses the radial basis function RBF in the nonlinear mapping function as the kernel function; The data set substituted into the support vector machine model is Among them, k represents the number of sequences of pollutant data, N represents the total number of sequences of pollutant data, and z k represents the vector Z, y, which is composed of the results of the defuzzifier after the input of the kth group of pollutant concentration data obtained by actual monitoring. k represents the pollutant concentration value predicted by the kth group, R represents the real number domain, m represents the dimension of the input x vector, R m represents an m-dimensional real vector space, where m represents the modulus of the input pollutant concentration Z vector; The fitting function in the model is Among them, k represents the sequence of pollutant data, N represents the total number of sequences of pollutant data, k(z k ,y) is the kernel function, a k represents the kth support vector, and b represents the bias of the support vector machine model.

7. The method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine according to claim 6, characterized in that: In step S4, the objects of parameter adjustment are the penalty parameter C and the kernel function parameter Gamma; Among them, the kernel function parameter Gamma is adjusted using formula (2); Formula (2) is as follows: Among them, G represents the Gamma parameter of the kernel function, Gamma is the kernel function parameter of the radial basis function RBF, σ is the reciprocal of the kernel function Gamma, k(z k ,y) represents the Gaussian kernel function, ||z k -y|| is the two-norm distance, which represents the atmospheric pollutant concentration value z k The Euclidean distance from the predicted atmospheric pollutant concentration value y; The parameter adjustment method is: (4-1) Setting the initial variation range and step size of the penalty parameter C and the kernel function parameter Gamma, and searching for the optimal parameters of the penalty parameter C and the kernel function parameter Gamma within a certain range through grid division; (4-2) Select the smallest penalty parameter C and its corresponding kernel function parameter Gamma that can achieve the highest verification regression accuracy as the optimal parameters; (4-3) If there are multiple sets of minimum penalty parameters C and corresponding kernel function parameters Gamma, the first set of penalty parameters C and kernel function parameters Gamma found is selected as the optimal parameters.

8. The method for predicting the time series concentration of atmospheric pollutants based on fuzzy model support vector machine according to claim 7, characterized in that: The model iteration method in step S4 is: (4s-1) The inverse of the kernel function parameter Gamma obtained after parameter adjustment is used as the σ parameter in the Gaussian fuzzy membership function to reconstruct the iterative weighted defuzzifier; (4s-2) Substituting the time series pollutant concentration data of step S1 into the iterative weighted defuzzifier of step (4s-1) for fuzzy granulation, an iterative data set consisting of the time series pollutant concentration data and the corresponding AQI air pollution level is obtained; (4s-3) The iterative data set is then divided into several groups of time-sequential data sets according to the preset time periods. These data sets are used as training and validation data sets, and are substituted into the support vector machine model for iterative training to obtain the iterative model and the iterative kernel function parameter Gamma after iterative parameter adjustment. (4s-4) Repeat steps (4s-1) to (4s-3) and calculate the threshold θ until the threshold θ is less than a preset value or greater than a preset number of iterations, then stop the iteration; Among them, the threshold i represents the i-th iteration, σ i represents the σ of the i-th iteration, σ i-1 represents σ at the i-1th iteration.

Images