Modeling methods, devices, equipment, and storage media for air quality prediction models

Abstract

This invention provides a modeling method, apparatus, device, and storage medium for air quality prediction models, relating to the field of air quality prediction technology. The modeling method includes: S1, acquiring monitoring data for various monitoring objects; S2, preprocessing the monitoring data to eliminate abnormal data; S3, calculating the correlation coefficients between the various monitoring objects based on the preprocessed monitoring data; S4, classifying meteorological conditions based on the correlation coefficients and obtaining meteorological representatives for each classification; S5, obtaining feature sets of various pollutants from the preprocessed monitoring data based on the correlation coefficients and meteorological representatives; and S7, constructing concentration prediction models for various pollutants based on the feature sets of various pollutants and an Informer deep learning model. The prediction models constructed using the modeling method of this invention can more accurately predict the concentrations of various pollutants, thereby calculating a more accurate air quality index.

Description

Technical Field

[0001] This invention relates to the field of air quality prediction technology, and more specifically, to a modeling method, apparatus, device, and storage medium for an air quality prediction model. Background Technology

[0002] When air pollution reaches a certain concentration, it can harm human health and pollute the ecological environment. Therefore, establishing air quality forecasting models to know air quality conditions in advance and take corresponding measures is one of the effective methods to reduce air pollution and improve ambient air quality.

[0003] The commonly used WRF-CMAQ simulation system is limited by simulation results and the formation mechanism of pollutants, resulting in less than ideal forecasts. Therefore, it is necessary to establish reasonable prediction models to improve prediction accuracy and more accurately predict air quality conditions.

[0004] In view of this, the applicant hereby submits this application after studying the existing technology. Summary of the Invention

[0005] The present invention provides a modeling method, apparatus, device and storage medium for an air quality prediction model, in order to improve at least one of the above-mentioned technical problems.

[0006] First aspect

[0007] This invention provides a modeling method for an air quality prediction model, which includes steps S1 to S5 and step S7.

[0008] S1. Acquire monitoring data for multiple monitoring objects; wherein, the multiple monitoring objects include multiple meteorological conditions and multiple pollutants;

[0009] S2. Preprocess the monitoring data to eliminate abnormal data in the monitoring data;

[0010] S3. Calculate the correlation coefficient between each monitoring object based on the preprocessed monitoring data;

[0011] S4. Based on the correlation coefficient, classify the meteorological conditions and obtain a meteorological representative for each classification;

[0012] S5. Based on the correlation coefficient and the meteorological representative, obtain the feature sets of various pollutants from the preprocessed monitoring data;

[0013] S7. Based on the feature sets of the various pollutants, construct concentration prediction models for each pollutant using the Informer deep learning model.

[0014] The second aspect

[0015] This invention provides a modeling apparatus for an air quality prediction model, comprising:

[0016] The data acquisition module is used to acquire monitoring data of various monitoring objects, including various meteorological conditions and various pollutants.

[0017] A preprocessing module is used to preprocess the monitoring data to eliminate abnormal data in the monitoring data;

[0018] The correlation module is used to calculate the correlation coefficient between various monitoring objects based on the preprocessed monitoring data.

[0019] The classification module is used to classify the meteorological conditions according to the correlation coefficient and obtain a meteorological representative for each classification.

[0020] The feature set module is used to obtain feature sets of various pollutants from the preprocessed monitoring data based on the correlation coefficient and the meteorological representative.

[0021] The module is used to construct concentration prediction models for various pollutants based on the feature sets of the various pollutants and the Informer deep learning model.

[0022] Third aspect

[0023] This invention provides a modeling device for an air quality prediction model, which includes a processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement the modeling method for the air quality prediction model as described in any paragraph of the first aspect.

[0024] Fourth aspect

[0025] This invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a modeling method for an air quality prediction model as described in any paragraph of the first aspect.

[0026] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0027] The prediction model constructed by the modeling method of this invention can more accurately predict the concentration of various pollutants, thereby calculating a more accurate air quality index, which has great practical significance.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating the modeling method provided in the first embodiment of the present invention.

[0031] Figure 2 This is a logic diagram for the preprocessing of monitoring data.

[0032] Figure 3 This is the raw data before the Prophet algorithm detection.

[0033] Figure 4 These are abnormal data detected by the Prophet algorithm.

[0034] Figure 5 It is a visualization chart of the correlation coefficients between various monitored objects.

[0035] Figure 6 This is a logic block diagram of the modeling method provided in the first embodiment of the present invention.

[0036] Figure 7 This is the Informer model framework diagram.

[0037] Figure 8 This is a flowchart for calculating the primary pollutant and AQI.

[0038] Figure 9 This is a flowchart illustrating another modeling method provided in the first embodiment of the present invention.

[0039] Figure 10 It is a coordinate graph between the nearby monitoring station and the current monitoring station.

[0040] Figure 11 This is a schematic diagram of the modeling device provided in the second embodiment of the present invention. Detailed Implementation

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below. 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.

[0042] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0044] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0046] Example 1:

[0047] Please see Figures 1 to 10 The first embodiment of the present invention provides a modeling method for an air quality prediction model, which can be executed by an air quality prediction model modeling device (hereinafter referred to as: modeling device). In particular, it is executed by one or more processors in the modeling device to implement steps S1 to S5 and step S7.

[0048] S1. Acquire monitoring data for multiple monitoring targets. These multiple monitoring targets include various meteorological conditions and various pollutants.

[0049] Specifically, the meteorological monitoring station can periodically monitor various meteorological conditions and concentrations of various meteorological pollutants. In this embodiment, meteorological conditions include 16 meteorological indicators: near-ground 2-meter temperature, surface temperature, specific humidity, near-ground 10-meter wind speed, near-ground 10-meter wind direction, rainfall, cloud cover, marginal layer height, atmospheric pressure, sensible heat flux, latent heat flux, longwave radiation, shortwave radiation, and ground solar radiation. Pollutants include SO2 concentration, H2O concentration, and PM2.5 concentration. 10 Concentration, PM 2.5 The monitoring covers six common pollutants: concentration of oxygen, concentration of oxygen (O3), and concentration of carbon dioxide (CO). Other meteorological conditions and pollutants may also be included in other embodiments; this invention does not specifically limit this. Each monitoring cycle may be 1 hour, 15 minutes, or other durations; this invention also does not limit this.

[0050] It is understandable that forecast data for the nearest date is often quite accurate. Therefore, in this embodiment, the monitoring data includes not only the actual measured data but also the forecast data for the nearest date. In other embodiments, the monitoring data may include only the measured data or forecast data from several days prior to the current date; this invention does not impose specific limitations on this.

[0051] It is understood that the modeling device can be an electronic device with computing power, such as a portable laptop computer, desktop computer, server, smartphone, or tablet computer.

[0052] S2. Preprocess the monitoring data to eliminate abnormal data in the monitoring data.

[0053] Specifically, due to equipment debugging and maintenance at monitoring stations, equipment differences, and the influence of certain accidental factors at and near the monitoring stations, the measured data suffers from missing data and abnormal distribution. Furthermore, in forecasting operations, server outages, such as prolonged power failures, can lead to missing forecast data for certain operating dates. Therefore, to improve the accuracy of subsequent forecasts, data preprocessing is necessary to remove outlier data and repair missing data.

[0054] like Figure 2 As shown, based on the above embodiments, in an optional embodiment of the present invention, step S2 includes steps S21 to S25.

[0055] S21. Detect whether there are missing data in the monitoring data with adjacent time differences not exceeding two hours or two days. When missing data with adjacent time differences not exceeding two hours or two days is detected, fill in the missing data based on the average value of the two time periods before and after the missing value.

[0056] S22. Detect whether there is data missing of no more than 80% in the monitoring data. When data missing of no more than 80% is detected, fill in the missing data based on the average value of the two time periods before and after the missing value.

[0057] Specifically, for cases where data contains partial or intermittent missing variable values ​​due to equipment debugging or maintenance at monitoring stations, and where the time difference between adjacent data does not exceed two hours (two days) or the missing value of a certain record does not exceed 20%, the primary approach is to fill the missing value using the average of the values ​​from the two time periods before and after it.

[0058] S23. The Prophet algorithm is used to detect outliers in the monitoring data. When an outlier is detected, it is set to null, and then the K-nearest neighbor filling method is used to fill in the outlier. The confidence interval of the Prophet algorithm is 98%.

[0059] Specifically, for data deviating from an abnormal distribution, records with null values ​​are removed to detect outliers using the Prophet algorithm. That is, the confidence interval of the fitted value of the data fluctuation (the default confidence interval is 98%) is used as the upper and lower limits for determining whether a value is an outlier. In other embodiments, the confidence interval can be set to other ranges, and this invention does not specifically limit this. Figure 3 and Figure 4 For example, Figure 3 The original data distribution of PM10, after anomaly detection ( Figure 4 Approximately 10 abnormal data points can be found.

[0060] Outliers are detected using the Prophet algorithm, which can be implemented using the Prophet library in Python. After an outlier is detected, it is set to null, and the data is imputed using the K-nearest neighbor method. This method uses the combined results of multiple nearest neighbors of the sample with the missing value to impute the missing value. Specifically, it calculates the K nearest neighbors to the sample with the missing value based on Euclidean distance (Equation 1), and finally imputes the missing value using the mean of the K nearest neighbors. This method is implemented using the KNNImputer function from the Sklearn library in Python.

[0061]

[0062] S24. Detect whether there is data missing of more than 80% in the monitoring data. When data missing of more than 80% is detected, the K-nearest neighbor imputation method is used to imput the missing data.

[0063] Specifically, this addresses situations where some indicators are significantly missing or where data for consecutive time periods is significantly missing, with a missing data ratio exceeding 20%. Directly deleting rows containing missing values ​​would result in the loss of a large amount of useful information. From a statistical perspective, as the number of independent data points decreases, the degrees of freedom decrease, inevitably reducing the model's prediction accuracy. To ensure the accuracy of subsequent predictions, K-nearest neighbor imputation is also used.

[0064] S25. The filled monitoring data is normalized using Z-Score standardization to obtain preprocessed monitoring data.

[0065] Specifically, Z-Score standardization can transform data of different magnitudes into a unified Z-Score metric for comparison. Its calculation formula is as follows:

[0066]

[0067] In the formula, x is the original data, μ is the sample mean, and σ is the standard deviation of the data (Equation 3).

[0068]

[0069] After normalizing the data, the mean becomes 0 and the variance becomes 1. This method is used to process all data, thus achieving dimensionlessness and avoiding errors caused by differences in units.

[0070] S3. Calculate the correlation coefficient between each monitoring object based on the preprocessed monitoring data.

[0071] Specifically, statistical methods are used to statistically summarize and organize the various indicators in the data. This embodiment of the invention combines correlation calculation methods with visualization analysis to perform correlation calculations and visualizations on the correlations between pollutants, between various meteorological condition indicators, and on the correlations between meteorological conditions and various pollutants.

[0072] Based on the above embodiments, in an optional embodiment of the present invention, step S3 includes calculating the correlation coefficients between various detection objects using Pearson, Spearman, and Kendall methods, respectively, based on the preprocessed monitoring data.

[0073] Specifically, to more scientifically explore and elucidate the correlations between meteorological conditions and pollutant concentrations, as well as between pollutants and between meteorological conditions, this invention employs three correlation coefficient calculation methods: Pearson (Equation 4), Spearman (Equation 5), and Kendall (Equation 6). These three formulas are used to perform correlation analysis on the meteorological characteristics and pollutant factors at monitoring point A. The visualization results are shown below. Figure 5 As shown. In other embodiments, other correlation coefficients can also be used to calculate the correlation between the various monitored objects, and there are no specific limitations on this.

[0074]

[0075]

[0076]

[0077] It is understandable that actual meteorological conditions have a significant impact on pollutant concentrations (for example, reduced humidity helps reduce ozone formation). That is, under the condition that pollutant emissions remain unchanged, the diffusion of pollutants is significantly correlated with meteorological conditions. Calculating the correlation coefficients between various monitoring objects is of great practical significance in order to assess the degree of influence of meteorological conditions on pollutant concentrations from both quantitative and qualitative perspectives.

[0078] S4. Based on the correlation coefficient, classify the meteorological conditions and obtain the meteorological representative for each classification.

[0079] Specifically, various meteorological conditions may have strong correlations. Eliminating similar meteorological conditions can effectively reduce the size of the original data for modeling.

[0080] Based on the above embodiments, in an optional embodiment of the present invention, step S4 includes steps S41 to S42.

[0081] S41. Based on the correlation coefficient, meteorological conditions with high correlation are grouped into one category, thereby classifying various meteorological conditions.

[0082] S42. Select one meteorological condition from each category to obtain a meteorological representative for each category.

[0083] In this embodiment, as Figure 5 As shown, the correlations among various meteorological conditions indicate that near-surface 2-meter temperature, surface temperature, specific humidity, and longwave radiation exhibit similar strong positive correlations; therefore, these meteorological conditions are grouped together. Marginal layer height, surface temperature, sensible heat flux, latent heat flux, shortwave radiation, and ground solar radiation also show similar strong positive correlations; therefore, these meteorological conditions are also grouped together. The final meteorological representations include: specific humidity, surface temperature, humidity, near-surface 10-meter wind speed, near-surface 10-meter wind direction, rainfall, cloud cover, and atmospheric pressure.

[0084] In other embodiments, the monitored objects may also include other meteorological conditions and pollutants, so the final classification and meteorological representation may not be the same as those in the embodiments of the present invention, and the present invention does not specifically limit them.

[0085] In this embodiment, the correlation between the pollutants indicates that PM 2. and PM 10 There is a very strong positive correlation, with a correlation coefficient of nearly 0.99. This indicates that PM... 2.5 and PM 10 The formation mechanisms of PM are similar. However, the reality is that PM... 2.5 PM refers to particulate matter with an aerodynamic diameter of 2.5 micrometers or less. 10 This indicates that the size is less than or equal to 10 micrometers. This demonstrates that the types of pollution and possible sources of pollutants are similar, and the actual situation is completely consistent with the correlation analysis results. Furthermore, SO2 has a significant impact on NO2 and PM2.5. 2.5 PM 10 NO2 has a strong correlation with CO, and its correlation with SO2 and PM2.5 is also strong. 2.5 PM 10 PM is strongly correlated with CO. 2.5 PM 10 CO also has a strong correlation with SO2 and NO2.

[0086] In this embodiment, the correlation between various meteorological conditions and pollutants shows that: near-ground 2-meter temperature, surface temperature, specific humidity, and longwave radiation affect SO2, NO2, and PM2.5. 2.5 and PM 10 They exhibit a strong negative correlation with atmospheric pressure on SO2, NO2, and PM2.5. 2.5 and PM 10 The marginal layer height, sensible heat flux, latent heat flux, shortwave radiation, and ground solar radiation all show a positive correlation with ozone, while the marginal layer height, sensible heat flux, latent heat flux, shortwave radiation, and ground solar radiation all show a similar strong positive correlation with ozone. These meteorological characteristics have similar effects on changes in pollutant concentrations, which precisely illustrates the strong similarity among these meteorological conditions in the meteorological condition classification.

[0087] S5. Based on the correlation coefficient and meteorological representative, obtain the characteristic sets of various pollutants from the preprocessed monitoring data.

[0088] Specifically, as shown in step S4, the influencing factors of the six pollutants are different. Therefore, the strongly correlated factors of each pollutant are combined into a feature set for that pollutant, thus providing a more reliable data foundation for subsequent predictions of the pollutant.

[0089] Based on the above embodiments, in an optional embodiment of the present invention, step S5 includes steps S51 to S52.

[0090] S51. Based on the correlation coefficient, obtain the meteorological representatives of strong correlations for various pollutants and other pollutants.

[0091] Table 1 Strong correlation factors of six pollutants

[0092]

[0093]

[0094] S52. From the preprocessed monitoring data, obtain the monitoring data of strongly correlated meteorological representatives and other pollutants, thereby obtaining the feature sets of various pollutants.

[0095] Specifically, due to the different formation mechanisms of the six pollutants and the varying degrees to which different meteorological conditions influence their formation, a feature set for each monitored object is constructed using the correlation coefficients calculated in step S4. In general, the prediction of pollutants requires consideration of two characteristics: the influence of meteorological conditions on pollutant concentrations and the interactions between pollutants.

[0096] S7. Based on the feature sets of various pollutants, construct concentration prediction models for various pollutants using the Informer deep learning model.

[0097] Specifically, as shown in step S4, the influencing factors of the concentrations of the six pollutants are different, therefore it is necessary to establish separate prediction models for each of the six pollutants. Given the various unknown nonlinear factors and complex formation mechanisms, the correlation calculation results from step S4 are used in conjunction with a deep learning model to establish prediction models for the concentrations of the six pollutants.

[0098] The Transformer deep learning model performs well in long time series data, and its variant, the Informer, effectively captures the precise long-range correlation coupling between the output and input, significantly reducing network parameters and processing dimensions. This improves the speed of long-term prediction compared to previous step-by-step methods and reduces computational time and space complexity. In this pollutant concentration prediction, the prediction error was reduced to the decimal level. The Informer network training algorithm consists of two stages: a forward propagation stage and a backpropagation stage.

[0099] The forward propagation phase is as follows:

[0100] Select training samples (X) i ,Y i The input is fed into the network to calculate the corresponding output. i .

[0101] Output i =F n (,…,(F2(F1(X i W (1) W (2) ),…,)W (n) (7)

[0102] During the backpropagation phase, it is necessary to construct the model output. i Compared with the true result value Y i The objective function (Equation 8) is used to adjust the parameters with the goal of minimizing the objective function.

[0103]

[0104] like Figure 7 As shown, the Informer model is mainly divided into three parts: the data input part, the encoder part, and the decoder part.

[0105] Data Input: The ability to capture long-term independence requires global information, such as hierarchical timestamps (week, month, and year) and agnostic timestamps (holidays, events). These are difficult to utilize in classic self-attention, so Informer addresses this issue by using a unified input representation. The input embedding consists of three independent parts: scalar projection, local timestamps (location), and global timestamp embeddings (minutes, hours, weeks, months, holidays, etc.).

[0106] Encoder section: The encoder receives a large number of long sequence inputs. Among them, ProbSparse Self-attention replaces the classic self-attention. The self-attention distilling part solves the attention score problem by selecting the most important self-attention, reducing the network size, and the stacked copies also increase the robustness of the copies.

[0107] Decoder section: The decoder receives a long sequence input, fills the target element with zeros, measures the weighted attention component of the feature map, and immediately predicts the output element in a generative manner. That is, the generative decoder obtains the long sequence output using a forward step.

[0108] It should be noted that, based on the concentration prediction model, the air quality sub-indices of various pollutants, the primary pollutant, and the overall air quality index (AQI) can be predicted and calculated. Specifically, the AQI of each pollutant is calculated using Formula 9 and the predicted concentration values ​​of each pollutant, and finally, Formula 10 is used to calculate the primary pollutant and the AQI. The calculation process is as follows: Figure 8 As shown.

[0109]

[0110]

[0111] The prediction model constructed by the modeling method of this invention can more accurately predict the concentration of various pollutants, thereby calculating a more accurate air quality index, which has great practical significance.

[0112] like Figure 9 As shown, in an optional embodiment of the present invention, based on the above embodiments, step S6 is further included before step S7.

[0113] S6. Obtain the geographical location relationship between the current monitoring point and adjacent monitoring points, and construct the regional cross-correlation factor between adjacent monitoring points and the current monitoring point based on the geographical location relationship and the preprocessed monitoring data.

[0114] Specifically, pollutant concentrations in adjacent areas may influence each other, so regional collaborative forecasting may improve the accuracy of air quality forecasts. In this embodiment of the invention, a collaborative forecasting model is established for the current monitoring point and monitoring points in its neighboring areas. Preferably, step S6 includes steps S61 to S63.

[0115] First, it is necessary to identify the factors that will have an impact. In this embodiment, the main factors considered are geographical location, wind speed, and wind direction, which will interact with each other.

[0116] S61. Obtain the geographical location relationship between the current monitoring point and adjacent monitoring points.

[0117] S62. Based on geographical location relationships, obtain the distance and angle of maximum influence of adjacent monitoring points on the current monitoring point.

[0118] Specifically, in this embodiment of the invention, wind direction is represented by angles. The wind direction from due north to the detection point is defined as 0° wind direction, and the wind direction is recorded with a positive value of clockwise rotation angle (unit: °). When the wind reaches the monitoring point from due east, the wind direction is recorded as 90°.

[0119] First, establish a relative coordinate system between the current monitoring station and nearby monitoring stations. Wind direction will cause pollutants to disperse between the two monitoring points. Specifically, for example... Figure 10 As shown, if the wind blows from point A to A1, A2, and A3, the influence of these three monitoring points on point A is 0. If the wind blows from A1 to A, A2 to A, and A3 to A, the influence of these three points on A is the greatest. Therefore, based on the actual wind direction influence, it is necessary to connect AA1, AA2, and AA3 respectively to establish their respective relative coordinate systems. The direction from A to A1 is defined as 0°, and the direction perpendicular to line AA1 upwards is defined as 90°. When the wind direction is 0° and 90°, the influence of monitoring point A1 on monitoring point A is 0; when the wind direction is 180°, the influence of A1 on A is the greatest; when the wind direction is between 180° and 90° and between 180° and 270°, the influence of A1 on monitoring point A gradually decreases, therefore the influential wind direction range is [90°, 270°]. The process of establishing the new wind direction coordinates for AA2 and AA3 is the same as for AA1.

[0120] Then, a standard coordinate system is established with the current monitoring point as the origin. The wind direction from due north to the monitoring point is 0°, and the wind direction is recorded with a positive value of clockwise rotation angle (unit: °). When the wind reaches the monitoring point from due east, the wind direction is 90°. Using this defined wind direction coordinate as the standard, the wind direction angles that affect monitoring point A in the relative coordinate system are converted into angle values ​​in the standard coordinate system. The influential wind direction angle values ​​are shown in Table 2.

[0121] Table 2: Wind direction angles affecting nearby monitoring stations

[0122]

[0123] S63. Based on distance, the angle of maximum influence, and preprocessed monitoring data, construct a calculation model for the regional cross-correlation factor between adjacent monitoring points and the current monitoring point. Multiple meteorological conditions include wind direction and wind speed. The calculation model is as follows:

[0124]

[0125] h(x1) = |max - x1|

[0126] In the formula, x0 represents the distance from the adjacent monitoring point to the current monitoring point, x1 represents the wind direction in absolute coordinates, x2 represents the wind speed, e represents the natural base, and max represents the angle with the greatest influence.

[0127] Specifically, wind direction and speed are limited by the distance between two monitoring points. When the distance is short, wind direction and speed have a greater impact on pollutant dispersion. When the distance is long, their impact is smaller. Therefore, it can be concluded that distance is inversely proportional to the degree of influence. Higher wind speeds have a greater impact on pollutant dispersion, while lower wind speeds have a smaller impact. Therefore, it can be concluded that wind speed is directly proportional to the degree of influence.

[0128] Wind direction has a greater impact on pollutant dispersion in areas with influence than in areas without influence, with an impact approaching zero. Therefore, wind direction in areas with influence is directly proportional to the degree of influence.

[0129] by Figure 10 For example: For A1->A, the influence gradually decreases when the influencing angle is within the range of the maximum value (max) to the left endpoint (172.255°) and when it is within the range of the maximum value (max) to the left endpoint (352.255°) (showing symmetry); for A2->A, the influence gradually decreases when it is within the range of the maximum value (max) to the left endpoint (228.704°) and when it is within the range of the maximum value (48.704°) (showing symmetry); for A3->A, the influence also decreases when it is within the range of the maximum value (max) to the left endpoint (123.783°) and when it is within the range of the maximum value (303.783°) (showing symmetry).

[0130] In this embodiment, the influence of wind direction can be calculated using formula 11:

[0131] h(x1)=|max-x1| (11)

[0132] In this embodiment, the overall impact can be summarized by Formula 12:

[0133]

[0134] Preferably, step S7 includes: based on the feature sets of each pollutant and the regional cross-correlation factors of adjacent monitoring points to the current monitoring point, constructing concentration prediction models for each pollutant using the Informer deep learning model.

[0135] Specifically, the current step is the same as the aforementioned step S7. The only difference is that this embodiment of the invention incorporates a regional cross-correlation factor into the feature set, thereby obtaining a concentration prediction model that considers the various pollutants from nearby monitoring stations.

[0136] In this embodiment of the invention, the accuracy of the concentration prediction model is improved by considering the influence of nearby monitoring stations on the current monitoring station.

[0137] Table 3: Symbol Explanation of Embodiments of the Invention

[0138]

[0139]

[0140] Example 2

[0141] Please see Figure 11 This invention provides a modeling apparatus for an air quality prediction model, comprising:

[0142] Data acquisition module 1 is used to acquire monitoring data for various monitoring targets. These targets include various meteorological conditions and various pollutants.

[0143] Preprocessing module 2 is used to preprocess the monitoring data to eliminate abnormal data in the monitoring data.

[0144] The correlation module 3 is used to calculate the correlation coefficient between each monitoring object based on the preprocessed monitoring data.

[0145] Classification module 4 is used to classify meteorological conditions based on correlation coefficients and obtain meteorological representatives for each classification.

[0146] Feature set module 5 is used to obtain feature sets of various pollutants from the preprocessed monitoring data based on correlation coefficients and meteorological representations.

[0147] Module 7 is used to construct concentration prediction models for various pollutants based on the feature sets of various pollutants and the Informer deep learning model.

[0148] In an optional embodiment, the preprocessing module 2 includes:

[0149] The first filling unit is used to detect whether there are missing data in the monitoring data with adjacent time differences not exceeding two hours or two days. When missing data with adjacent time differences not exceeding two hours or two days is detected, the missing data is filled in according to the average value of the two time periods before and after the missing value.

[0150] The second unit is used to detect whether there is data loss of no more than 80% in the monitoring data. When data loss of no more than 80% is detected, the missing data is filled in based on the average value of the two time periods before and after the missing value.

[0151] The third unit is used to detect outliers in the monitoring data using the Prophet algorithm. When an outlier is detected, it is set to null, and then the K-nearest neighbor filling method is used to fill in the outlier. The confidence interval for the Prophet algorithm is 98%.

[0152] The fourth unit is used to detect whether there are data gaps greater than 80% in the monitoring data. When data gaps greater than 80% are detected, the K-nearest neighbor imputation method is used to fill in the missing data.

[0153] The normalization unit is used to normalize the filled monitoring data using Z-Score standardization to obtain preprocessed monitoring data.

[0154] Correlation module 3 is specifically used to calculate the correlation coefficients between various detection objects based on the preprocessed monitoring data using Pearson, Spearman, and Kendall methods.

[0155] In an optional embodiment, the classification module 4 includes:

[0156] Classification units are used to group meteorological conditions with high correlation into one category based on their correlation coefficients, thereby classifying various meteorological conditions.

[0157] The representative acquisition unit is used to select a meteorological condition from each category to obtain the meteorological representative for each category. The meteorological representative includes: specific humidity, surface temperature, humidity, wind speed at 10 meters above the ground, wind direction at 10 meters above the ground, rainfall, cloud cover, and atmospheric pressure.

[0158] In an optional embodiment, the feature set module 5 includes:

[0159] The strongly correlated monitoring object acquisition unit is used to acquire the strongly correlated meteorological representatives of various pollutants and other pollutants based on the correlation coefficient.

[0160] The feature set acquisition unit is used to acquire strongly correlated meteorological representative data and monitoring data of other pollutants from the preprocessed monitoring data, thereby obtaining feature sets of various pollutants.

[0161] In an optional embodiment, the modeling apparatus further includes a regional cross-correlation factor construction module 7.

[0162] The regional cross-correlation factor construction module 7 is used to obtain the geographical location relationship between the current monitoring point and adjacent monitoring points, and to construct the regional cross-correlation factor between adjacent monitoring points and the current monitoring point based on the geographical location relationship and the preprocessed monitoring data.

[0163] In an optional embodiment, the construction module 7 is specifically used to: construct concentration prediction models for each pollutant based on the feature sets of each pollutant and the regional cross-correlation factors of adjacent monitoring points to the current monitoring point, using the Informer deep learning model.

[0164] In an optional embodiment, the regional cross-correlation factor construction module 7 includes:

[0165] The geographic location acquisition unit is used to obtain the geographic location relationship between the current monitoring point and adjacent monitoring points.

[0166] The influence parameter acquisition unit is used to obtain the distance and the angle of maximum influence of adjacent monitoring points to the current monitoring point based on geographical location relationships.

[0167] The regional cross-correlation factor calculation unit is used to construct a calculation model of the regional cross-correlation factor between adjacent monitoring points and the current monitoring point, based on distance, the angle of maximum influence, and preprocessed monitoring data. Multiple meteorological conditions, including wind direction and wind speed, are considered. The calculation model is as follows:

[0168]

[0169] h(x1) = |max - x1|

[0170] In the formula, x0 represents the distance from the adjacent monitoring point to the current monitoring point, x1 represents the wind direction in absolute coordinates, x2 represents the wind speed, e represents the natural base, and max represents the angle with the greatest influence.

[0171] Example 3

[0172] This invention provides a modeling device for an air quality prediction model, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement the modeling method for the air quality prediction model as described in any paragraph of Embodiment 1.

[0173] Example 4

[0174] This invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the modeling method of the air quality prediction model as described in any paragraph of Embodiment 1.

[0175] In the several embodiments provided in this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0176] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0177] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0178] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A modeling method for an air quality prediction model, characterized in that, Include: Acquire monitoring data for multiple monitoring targets; wherein, the multiple monitoring targets include multiple meteorological conditions and multiple pollutants; The monitoring data is preprocessed to eliminate abnormal data. Based on the preprocessed monitoring data, the correlation coefficient between each monitoring object is calculated; Based on the correlation coefficient, the meteorological conditions are classified, and a meteorological representative for each classification is obtained; Based on the correlation coefficient and the meteorological representative, feature sets of various pollutants are obtained from the preprocessed monitoring data; Based on the feature sets of various pollutants, concentration prediction models for each pollutant are constructed using the Informer deep learning model. Based on the correlation coefficient and the meteorological representative, feature sets of various pollutants are obtained from the preprocessed monitoring data, specifically including: Based on the correlation coefficients, the meteorological representatives of strong correlations for various pollutants and other pollutants are obtained respectively; From the preprocessed monitoring data, obtain the monitoring data of strongly correlated meteorological representatives and other pollutants, thereby obtaining the feature sets of various pollutants; Before constructing concentration prediction models for each pollutant based on the feature sets of the various pollutants and the Informer deep learning model, the following steps are also included: Obtain the geographical location relationship between the current monitoring point and adjacent monitoring points, and construct the regional cross-correlation factor between adjacent monitoring points and the current monitoring point based on the geographical location relationship and the preprocessed monitoring data; Obtain the geographical location relationship between the current monitoring point and adjacent monitoring points, and construct the regional cross-correlation factor between adjacent monitoring points and the current monitoring point based on the geographical location relationship and the preprocessed monitoring data. Specifically, this includes: Obtain the geographical location relationship between the current monitoring point and adjacent monitoring points; Based on the geographical location relationship, obtain the distance and the angle of maximum influence of the adjacent monitoring points on the current monitoring point; Based on the distance, the angle of maximum influence, and the preprocessed monitoring data, a calculation model for the regional cross-correlation factor between adjacent monitoring points and the current monitoring point is constructed; wherein, the various meteorological conditions include wind direction and wind speed; the calculation model is as follows: In the formula, The distance from the adjacent monitoring point to the current monitoring point Indicates wind direction in absolute coordinates; Indicates wind speed, Represents the natural base, This indicates the angle with the greatest impact; Based on the feature sets of the various pollutants, concentration prediction models for each pollutant are constructed using the Informer deep learning model, specifically including: Based on the feature sets of various pollutants and the regional cross-correlation factors between adjacent monitoring points and the current monitoring point, concentration prediction models for various pollutants are constructed using the Informer deep learning model.

2. The modeling method for the air quality prediction model according to claim 1, characterized in that, The monitoring data is preprocessed to eliminate abnormal data, specifically including: The system detects whether there are missing data in the monitoring data with an adjacent time difference of no more than two hours or two days; when it detects that there are missing data with an adjacent time difference of no more than two hours or two days, it fills the missing data with the average value of the two time periods before and after the missing value. The system detects whether there is data loss of no more than 80% in the monitoring data; when data loss of no more than 80% is detected, the missing data is filled in according to the average value of the missing value in the two time periods before and after the missing value. The Prophet algorithm is used to detect outliers in the monitoring data. When an outlier is detected, it is set to null, and then the K-nearest neighbor filling method is used to fill in the outliers. The confidence interval of the Prophet algorithm is 98%. The system detects whether there is data loss greater than 80% in the monitored data; when data loss greater than 80% is detected, the K-nearest neighbor filling method is used to fill the missing data. For the monitoring data after filling, adopt Standardization is performed to normalize the data and obtain preprocessed monitoring data.

3. The modeling method for the air quality prediction model according to claim 1, characterized in that, Based on the preprocessed monitoring data, the correlation coefficients between various monitoring objects are calculated, specifically including: Based on the preprocessed monitoring data, the correlation coefficients between various detection objects were calculated using Pearson, Spearman, and Kendall methods, respectively. Based on the correlation coefficient, the meteorological conditions are classified, and a meteorological representative for each classification is obtained, specifically including: Based on the correlation coefficient, meteorological conditions with high correlation are grouped into one category, thereby classifying the various meteorological conditions. Select one meteorological condition from each category to obtain a meteorological representative for each category; wherein the meteorological representative includes: specific humidity, surface temperature, humidity, wind speed at 10 meters above the ground, wind direction at 10 meters above the ground, rainfall, cloud cover, and atmospheric pressure.

4. A modeling apparatus for an air quality prediction model, suitable for executing the modeling method for an air quality prediction model as described in any one of claims 1 to 3, characterized in that, Include: The data acquisition module is used to acquire monitoring data of various monitoring objects, including various meteorological conditions and various pollutants. A preprocessing module is used to preprocess the monitoring data to eliminate abnormal data in the monitoring data; The correlation module is used to calculate the correlation coefficient between various monitoring objects based on the preprocessed monitoring data. The classification module is used to classify the meteorological conditions according to the correlation coefficient and obtain a meteorological representative for each classification. The feature set module is used to obtain feature sets of various pollutants from the preprocessed monitoring data based on the correlation coefficient and the meteorological representative. The module is used to construct concentration prediction models for various pollutants based on the feature sets of the various pollutants and the Informer deep learning model.

5. A modeling device for an air quality prediction model, characterized in that, It includes a processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement the modeling method of the air quality prediction model as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the modeling method of the air quality prediction model as described in any one of claims 1 to 3.

Images