Intelligent algorithm for adaptively optimizing urban environment air quality forecast

Through adaptive optimization intelligent algorithms, using historical data and machine learning models, the problems of high data dependence and lag in forecast results in urban-level applications are solved, and high accuracy and timely air quality forecasts are achieved. Urban environmental protection and sustainable development.

CN118917710APending Publication Date: 2024-11-08ANHUI UNIVERSITY OF ARCHITECTURE
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410896864.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-05
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In urban-level applications, air quality models face problems such as high data dependence, low model resolution and large demand for computing resources, resulting in insufficient credibility and timeliness of forecast results.

Method used

Adaptively optimized intelligent algorithms, and use statistical methods and machine learning algorithms of 1-90 days of historical data (such as clustering algorithms and long-term memory network LSTM models), to identify pollution characteristics and make immediate updates to achieve air quality pollutants. High accuracy forecast of concentration.

Benefits of technology

It reduces data dependence and forecast costs, improves forecast accuracy and timeliness, can quickly revise air quality indicators, and provides more reliable information to support urban environmental protection and sustainable development.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917710A_ABST
    Figure CN118917710A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of air quality forecasting, and particularly discloses an intelligent algorithm for adaptively optimizing urban environment air quality forecasting, which comprises the following steps of: calculating to obtain a standardized data column for an actually measured pollutant concentration hour value 1-90 days before a forecasting day; recording a sample serial number i belonging to the main type as a new set N; carrying out standardized calculation on the forecast data of the current day; calculating to obtain a new forecast value; according to a rolling forecasting principle, through the steps, M days of complete forecasting and one-time revision are completed, linear regression fitting is adopted to solve a coefficient, a secondary correction result is obtained through calculation, and six pollutant concentrations of air quality are updated and released in real time by integrating an air quality model calculation result and a correction mode, so that the aim of improving forecasting accuracy is achieved. The implementation cost is low, the operation effect is good, and the requirements of urban air quality forecasting are completely met.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of air quality forecasting, and in particular to an intelligent algorithm for adaptively optimizing urban environmental air quality forecasting. Background Art

[0002] Air quality models are a set of mathematical and computational frameworks used to simulate and predict the concentration of pollutants in the atmosphere. These models combine atmospheric chemistry, physical processes, meteorological conditions, and emissions from human activities to predict air quality conditions in specific regions and time periods. The forecasting method usually includes collecting pollutant emission data and meteorological data, using these basic data to drive the model operation, and predicting future air quality indicators by simulating chemical reactions and pollutant diffusion processes in the atmosphere. The forecast results help governments, businesses, and the public understand future air quality conditions and provide a basis for taking appropriate emission reduction measures and protecting public health.

[0003] Air quality models usually require a large amount of accurate input data, and the lack or inaccuracy of data will directly affect the credibility of the forecast results. In addition, the simplified treatment of complex atmospheric chemical reactions and physical processes by the model may not fully reflect the differences between cities, especially under small-scale or complex terrain conditions. Therefore, at the spatial scale of the city, considering directly modifying the air quality model to improve the forecast level will face problems such as missing basic data, low model resolution, high computing resource requirements and high costs. These difficulties limit the promotion and application of air quality models in the field of urban environment and meteorology. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent algorithm for adaptively optimizing urban ambient air quality forecasting to solve the above-mentioned background problems.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] An intelligent algorithm for adaptively optimizing urban ambient air quality forecasting includes the following steps:

[0007] The hourly value of a pollutant concentration measured 1-90 days before the forecast date is recorded as X i,j , where i = 0, 1, 2 ... 90, j = 0, 1, 2 ... 23, i represents the date ordinal, j represents the hour ordinal; the standardized data column is calculated;

[0008] X' obtained by pre-processing the historical concentration of a pollutant for 1-90 days i , where the obtained X' i 'Calculate the average value at the same time every day and get And record the sample number i belonging to the main type as a new set N;

[0009] For the forecast data Y j After normalization calculation, we get Y' j ;

[0010] Assuming that the predicted value and the measured value satisfy the linear equation f(x)=ax+b, a new forecast value is calculated;

[0011] According to the rolling forecast principle, after the above steps, the complete forecast for M days and one revision have been completed. The coefficient μ is calculated by linear regression fitting. i , the predicted μ of the day is M , and then calculate the secondary correction result Y' M '=μ M ×Y' M .

[0012] As a further solution of the present invention: for the data column X of the i-th day i,0 , X i,1 , …, X i,23 , calculate its mean Xm i and standard deviation Xs i , get the standardized data column:

[0013] As a further embodiment of the present invention: i The calculation methods include:

[0014] First, a clustering algorithm is used to divide the samples in the data set into K (K ≥ 3) different categories:

[0015] Then, count the K clusters separately;

[0016] If there is a main cluster (the number of samples accounts for more than 50% of the total samples), extract and save the cluster samples as a data set;

[0017] If the sample size ratio cannot reach >50%, resampling and clustering are performed.

[0018] As a further solution of the present invention: the clustering algorithm includes the following steps: Taking the K-means algorithm as an example, it is regarded as a multidimensional space vector, and the specific steps are as follows:

[0019] V1: Randomly select K initial cluster centers, which will serve as representatives of the final clusters;

[0020] V2: For each data sample, calculate its distance from each cluster center point and classify it into the cluster represented by the nearest center point;

[0021] V3: For each cluster, calculate the average value of all samples in the cluster and use it as the new cluster center point;

[0022] V4: Repeat steps V2 and V3 until the cluster center point no longer changes or the preset number of iterations is reached;

[0023] V5: Algorithm: Output K clusters, each cluster contains a group of similar data samples.

[0024] As a further solution of the present invention: the first-order coefficient

[0025] As a further solution of the present invention: the calculation method of the new forecast value is:

[0026] For i∈N, the maximum likelihood estimation method is used to calculate the parameters Introducing the inverse distance weight coefficient At this point, the revised equation for the forecast data is obtained, and the new forecast value is calculated

[0027] As a further solution of the present invention: i The calculation method of μ is: i =regress(X i,j , Y' i,j ), i=1,2,…,M-1.

[0028] As a further solution of the present invention: the first revision includes: for i∈M, there is a measured sample set X i-1,j , a revised forecast sample set Y' i,j .

[0029] As a further solution of the present invention: i Divided into: training set, validation set, and test set.

[0030] As a further solution of the present invention: predicting the μ of the forecast day M , using the Long Short-Term Memory (LSTM) machine learning model.

[0031] Beneficial effects of the present invention:

[0032] (1) In the present invention, the intelligent algorithm for adaptively optimizing the urban ambient air quality forecast is based on the calculation results of the air quality model. According to the characteristics of the forecast target city, the 1-90-day historical data is used to adopt a statistical method to invert the temporal variation pattern of the concentrations of six pollutants in the local air quality. Then, the machine learning algorithm is used to identify the main pollution characteristics and calculate the correction mode of the pollutant concentration forecast results for the next 0-48 hours. Then, the six pollutant concentrations of the air quality are updated and published in real time by combining the calculation results of the air quality model and the correction mode, so as to achieve the purpose of improving the forecast accuracy.

[0033] (2) This method has the following obvious advantages when applied to urban-scale air quality forecasting: first, it has low data dependence and only requires 1-90 days of historical measured data; second, it can restore the local pollution status characteristics to the greatest extent, and does not require reconstruction of atmospheric physical and chemical parameters, topographical parameters, etc.; third, it uses a machine learning algorithm model for rolling calculations, and the entire process is automatically executed without human intervention and has high accuracy; therefore, this method has low implementation costs and good operating effects, and fully meets the requirements of urban-level air quality forecasting;

[0034] (3) In the present invention, the new algorithm realizes the automatic optimization process of the existing forecast results by introducing real-time data sources and learning and modeling historical data sets, which can achieve more timely forecasts and quickly revise the six pollutant concentration indicators of SO2, NO2, O3, CO, PM2.5 and PM10 in the air quality forecast in real time, so that decision makers and the public can obtain the latest air quality information more quickly, with good real-time and timeliness;

[0035] Compared with traditional methods, the model has low implementation cost. It only requires short-term historical measured data and does not require reconstruction of atmospheric physical and chemical parameters, thus reducing the forecast cost;

[0036] The application of machine learning algorithm models improves forecast accuracy. Through automatic rolling calculations, it can more accurately restore the current characteristics of local pollution and provide more reliable air quality forecast information for urban residents. The entire model process is automatically executed without manual intervention, which improves operational efficiency. This means that forecast results are updated more timely, forecast information is more reliable, and helps the public make timely decisions.

[0037] The low cost and high efficiency of this method make it a potential for sustainable development. By providing reliable air quality management support for cities, it helps promote environmental protection and sustainable development in cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The present invention will be further described below in conjunction with the accompanying drawings.

[0039] Figure 1 It is a schematic flow chart of the method of the present invention;

[0040] Figure 2 It is a schematic diagram of the measured data preprocessing method in the present invention;

[0041] Figure 3 It is a schematic diagram of a method for outputting a standardized data column in the present invention;

[0042] Figure 4 It is a schematic diagram of the simulation data preprocessing method in the present invention;

[0043] Figure 5 It is a schematic diagram of the fitting and parameter optimization method in the present invention;

[0044] Figure 6 It is a schematic diagram of the secondary revision algorithm in the present invention;

[0045] Figure 7 This is the second embodiment of the present invention. Figure 1 ;

[0046] Figure 8 This is the second embodiment of the present invention. Figure 2 ;

[0047] Fig. 9 This is the second embodiment of the present invention. Figure 3 . DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Embodiment 1

[0050] See also Figure 1-Figure 6 As shown, the present invention is an intelligent algorithm for adaptively optimizing urban ambient air quality forecasting, comprising the following steps:

[0051] Step 1: For the hourly value of a pollutant concentration measured 1-90 days before the prediction date, record it as X i,j , where i = 0, 1, 2 ... 90, j = 0, 1, 2 ... 23, i represents the date number, j represents the hour number; for the data column X of the i-th day i,0 , X i,1 , …, X i,23 , calculate its mean Xm i and standard deviation Xs i , get the standardized data column:

[0052] Step 2: X' obtained by pre-processing the historical concentration of a pollutant for 1-90 days i ,Using clustering algorithm, the samples in the data set are divided into K (K ≥ 3) different categories: K clusters are output, each cluster contains a group of similar data samples, if there is a main cluster (the number of samples accounts for more than 50% of the total samples), then the cluster samples are extracted and saved as the data set; finally, the average value is calculated at the same time every day, and the And record the sample number i belonging to the main type as a new set N;

[0053] Step 3: For the forecast data Y j After normalization calculation, we get Y' j , j=0,1,2…23.

[0054] Step 4: Assume that the predicted value and the measured value satisfy the linear equation f(x) = ax + b, where the coefficient of the first-order term is For i∈N, the maximum likelihood estimation method is used to calculate the parameters Introducing the inverse distance weight coefficient At this point, the revised equation for the forecast data is obtained, and the new forecast value is calculated

[0055] Step 5: According to the rolling forecast principle, after the above steps, the M-day complete forecast and one revision have been completed. For i∈M, there is a measured sample set X i-1,j , a revised forecast sample set Y' i,j , use linear regression fitting to find the coefficient μ i =regress(X i,j , Y' i,j ), i = 1, 2, ..., M-1; the obtained μ i Divided into training set, validation set, and test set, the long short-term memory network (LSTM) machine learning model is used to predict the μ of the forecast day. M , and then calculate the secondary correction result Y" M =μ M ×Y' M .

[0056] The intelligent algorithm for adaptively optimizing urban ambient air quality forecast is based on the calculation results of the air quality model. According to the characteristics of the forecast target city, it uses 1-90 days of historical data and adopts statistical methods to invert the temporal variation patterns of the concentrations of six local air quality pollutants. Then, it uses a machine learning algorithm to identify the main pollution characteristics and calculate the correction pattern for the pollutant concentration forecast results for the next 0-48 hours. Then, by combining the calculation results of the air quality model and the correction pattern, it instantly updates and publishes the six air quality pollutant concentrations, thereby achieving the purpose of improving the forecast accuracy.

[0057] This method has the following obvious advantages when applied to urban-scale air quality forecasting: first, it has low data dependence and only requires 1-90 days of historical measured data to be traced back; second, it can restore the local pollution status characteristics to the greatest extent, and does not require reconstruction of atmospheric physical and chemical parameters, topographical parameters, etc.; third, it uses a machine learning algorithm model for rolling calculations, and the entire process is automatically executed without human intervention and has high accuracy; therefore, this method has low implementation costs and good operating effects, and fully meets the requirements of urban-level air quality forecasting.

[0058] Embodiment 2

[0059] S01: The hourly value of a pollutant concentration measured 1-90 days before the forecast date is recorded as X i,j , where i = 0, 1, 2 ... 90, j = 0, 1, 2 ... 23, i represents the date number, j represents the hour number; for the data column X of the i-th day i,0 , X i,1 , …, X i,23 , calculate its mean Xm i and standard deviation Xs i , get the standardized data column:

[0060] S02: X' obtained after pre-processing the historical concentration of a pollutant for 1-90 days i , where i = 0, 1, 2…90;

[0061] S021: Use a clustering algorithm to divide the samples in the data set into K (K ≥ 3) different categories: Taking the K-means algorithm as an example, it is regarded as a multidimensional space vector. The specific steps are as follows:

[0062] S0211: Randomly select K initial cluster center points, which will serve as representatives of the final clusters;

[0063] S0212: For each data sample, calculate its distance from each cluster center point, and classify it into the cluster represented by the nearest center point;

[0064] S0213: For each cluster, calculate the average value of all samples in the cluster and use it as the new cluster center point;

[0065] S0214: Repeat step S0212 and step S0213 until the cluster center point no longer changes or the preset number of iterations is reached;

[0066] S0215: The algorithm will output K clusters, each of which contains a group of similar data samples;

[0067] S022: Count the K clusters separately;

[0068] If there is a main cluster (the number of samples accounts for more than 50% of the total samples), extract and save the cluster samples as a data set;

[0069] If the sample size ratio cannot be > 50%, resample and cluster;

[0070] S023: The obtained X' i 'Calculate the average value at the same time every day and get And record the sample number i belonging to the main type as a new set N;

[0071] S03: The simulation data preprocessing module has the same algorithm as the measured data preprocessing module. j After normalization calculation, we get Y' j , j = 0, 1, 2…23;

[0072] S04: Assume that the predicted value and the measured value satisfy the linear equation f(x) = ax + b, where the coefficient of the first-order term is For i∈N, the maximum likelihood estimation method is used to calculate the parameters Introducing the inverse distance weight coefficient At this point, the revised equation for the forecast data is obtained, and the new forecast value is calculated

[0073] S05: According to the rolling forecast principle, after the above steps, a complete forecast of M days and one revision have been completed. For i∈M, there is a measured sample set X i-1,j , a revised forecast sample set Y' i,j , use linear regression fitting to find the coefficient μ i =regress(X i,j , Y' i,j ), i = 1, 2, ..., M-1; the obtained μ iDivided into training set, validation set, and test set, the long short-term memory network (LSTM) machine learning model is used to predict the μ of the forecast day. M , and then calculate the secondary correction result Y" M =μ M ×Y' M .

[0074] In existing technologies, air quality forecasts may be limited by the frequency of data updates, resulting in delayed forecast results. The new algorithm introduces real-time data sources and automatically optimizes existing forecast results through learning and modeling of historical data sets. It can achieve more timely forecasts and quickly revise the six pollutant concentration indicators of SO2, NO2, O3, CO, PM2.5 and PM10 in air quality forecasts in real time, such as Figure 7-Figure 9 As shown, decision makers and the public can obtain the latest air quality information more quickly, with good real-time and timely performance;

[0075] Compared with traditional methods, the model has low implementation cost. It only requires short-term historical measured data and does not require reconstruction of atmospheric physical and chemical parameters, thus reducing the forecast cost;

[0076] The application of machine learning algorithm models improves forecast accuracy. Through automatic rolling calculations, it can more accurately restore the current characteristics of local pollution and provide more reliable air quality forecast information for urban residents. The entire model process is automatically executed without manual intervention, which improves operational efficiency. This means that forecast results are updated more timely, forecast information is more reliable, and helps the public make timely decisions.

[0077] The low cost and high efficiency of this method make it a potential for sustainable development. By providing reliable air quality management support for cities, it helps promote environmental protection and sustainable development in cities.

[0078] Embodiment 3

[0079] An intelligent algorithm system for adaptively optimizing urban ambient air quality forecasting, comprising:

[0080] Measured data preprocessing module: The hourly value of a pollutant concentration measured 1-90 days before the forecast date is recorded as X i,j , where i = 0, 1, 2 ... 90, j = 0, 1, 2 ... 23, i represents the date number, j represents the hour number; for the data column X of the i-th day i,0 , X i,1 , …, X i,23 , calculate its mean Xm i and standard deviation Xs i , get the standardized data column:

[0081] Daily variation characteristic analysis module: X' is obtained by pre-processing the historical concentration of a pollutant for 1-90 days i , where i = 0, 1, 2…90;

[0082] First, a clustering algorithm is used to divide the samples in the data set into K (K ≥ 3) different categories: Taking the K-means algorithm as an example, it is regarded as a multidimensional space vector. The specific steps are as follows:

[0083] Randomly select K initial cluster centers, which will serve as representatives of the final clusters;

[0084] For each data sample, calculate its distance from each cluster center point and classify it into the cluster represented by the nearest center point;

[0085] For each cluster, calculate the average value of all samples in the cluster and use it as the new cluster center point;

[0086] Repeat steps 2 and 3 until the cluster center point no longer changes or the preset number of iterations is reached;

[0087] Finally, the algorithm will output K clusters, each of which contains a group of similar data samples;

[0088] Then, count the K clusters separately;

[0089] If there is a main cluster (the number of samples accounts for more than 50% of the total samples), extract and save the cluster samples as a data set;

[0090] If the sample size ratio cannot be > 50%, resample and cluster;

[0091] Finally, calculate the average value at the same time every day to get And record the sample number i belonging to the main type as a new set N;

[0092] Simulation data preprocessing module: The simulation data preprocessing module has the same algorithm as the measured data preprocessing module. j After normalization calculation, we get Y' j , j=0,1,2…23.

[0093] Fitting and parameter optimization module: Assume that the predicted value and the measured value satisfy the linear equation f(x) = ax + b, where the coefficient of the first term For i∈N, the maximum likelihood estimation method is used to calculate the parameters Introducing the inverse distance weight coefficient At this point, the revised equation for the forecast data is obtained, and the new forecast value is calculated

[0094] Second revision algorithm module: According to the rolling forecast principle, after the above steps, M days of complete forecast and one revision have been completed. For i∈M, there is a measured sample set X i-1,j , a revised forecast sample set Y' i,j , use linear regression fitting to find the coefficient μ i =regress(X i,j , Y' i,j ), i = 1, 2, ..., M-1; the obtained μ i Divided into training set, validation set, and test set, the long short-term memory network (LSTM) machine learning model is used to predict the μ of the forecast day. M , and then calculate the secondary correction result Y' M '=μ M ×Y' M .

[0095] The new algorithm introduces real-time data sources and realizes the automatic optimization process of existing forecast results through learning and modeling of historical data sets. It can achieve more timely forecasts and quickly revise the six pollutant concentration indicators of SO2, NO2, O3, CO, PM2.5 and PM10 in air quality forecasts in real time, such as Figure 7-Figure 9 As shown, decision makers and the public can obtain the latest air quality information more quickly, with good real-time and timely performance;

[0096] Compared with traditional methods, the model has low implementation cost. It only requires short-term historical measured data and does not require reconstruction of atmospheric physical and chemical parameters, thus reducing the forecast cost;

[0097] The application of machine learning algorithm models improves forecast accuracy. Through automatic rolling calculations, it can more accurately restore the current characteristics of local pollution and provide more reliable air quality forecast information for urban residents. The entire model process is automatically executed without manual intervention, which improves operational efficiency. This means that forecast results are updated more timely, forecast information is more reliable, and helps the public make timely decisions.

[0098] The low cost and high efficiency of this method make it a potential for sustainable development. By providing reliable air quality management support for cities, it helps promote environmental protection and sustainable development in cities.

[0099] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An intelligent algorithm for adaptively optimizing urban ambient air quality forecasting, characterized in that: The following steps are involved: The hourly value of a pollutant concentration measured 1-90 days before the forecast date is recorded as X i,j , where i = 0, 1, 2 ... 90, j = 0, 1, 2 ... 23, i represents the date ordinal, j represents the hour ordinal; the standardized data column is calculated; X' obtained by pre-processing the historical concentration of a pollutant for 1-90 days i , where the obtained X' i 'Calculate the average value at the same time every day and get And record the sample number i belonging to the main type as a new set N; For the forecast data Y j After normalization calculation, we get Y' j ; Assuming that the predicted value and the measured value satisfy the linear equation f(x)=ax+b, a new forecast value is calculated; According to the rolling forecast principle, after the above steps, the complete forecast for M days and one revision have been completed. The coefficient μ is calculated by linear regression fitting. i , the predicted μ of the day is M , and then calculate the secondary correction result Y' M '=μ M ×Y' M .

2. According to claim 1, the intelligent algorithm for adaptively optimizing urban ambient air quality forecast is characterized in that: For the data column X on day i i,0 , X i,1 , …, X i,23 , calculate its mean Xm i and standard deviation Xs i , get the standardized data column:

3. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 1 is characterized in that: X' i The calculation methods include: First, a clustering algorithm is used to divide the samples in the data set into K (K ≥ 3) different categories: Then, count the K clusters separately; If there is a main cluster (the number of samples accounts for more than 50% of the total samples), extract and save the cluster samples as a data set; If the sample size ratio cannot reach >50%, resampling and clustering are performed.

4. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 3 is characterized in that: The clustering algorithm comprises the following steps: Taking the K-means algorithm as an example, it is regarded as a multidimensional space vector. The specific steps are as follows: V1: Randomly select K initial cluster centers, which will serve as representatives of the final clusters; V2: For each data sample, calculate its distance from each cluster center point and classify it into the cluster represented by the nearest center point; V3: For each cluster, calculate the average value of all samples in the cluster and use it as the new cluster center point; V4: Repeat steps V2 and V3 until the cluster center point no longer changes or the preset number of iterations is reached; V5: Algorithm: Output K clusters, each cluster contains a group of similar data samples.

5. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 1 is characterized in that: The first-order coefficient 6. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 1 is characterized in that: The new forecast value is calculated as follows: For i∈N, the maximum likelihood estimation method is used to calculate the parameters Introducing the inverse distance weight coefficient At this point, the revised equation for the forecast data is obtained, and the new forecast value is calculated 7. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 1 is characterized in that: The μ i The calculation method of μ is: i =regress(X i,j , Y' i,j ), i=1,2,…,M-1.

8. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 1 is characterized in that: The first revision includes: for i∈M, there is a measured sample set X i-1,j , a revised forecast sample set Y' i,j .

9. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 8, characterized in that: The μ i Divided into: training set, validation set, and test set.

10. The intelligent algorithm for adaptively optimizing urban ambient air quality forecast according to claim 9, characterized in that: The predicted μ for the day M , using the Long Short-Term Memory (LSTM) machine learning model.