Urban forest space particulate matter concentration prediction method

Abstract

The application discloses a kind of urban forest space particulate matter concentration prediction method, belong to air particulate matter concentration monitoring field, method includes: S1, the air pollutant concentration of urban forest space target area, urban meteorological data and the hourly observation data of AQI are obtained, and the air quality data set of city is established;S2, based on the air quality data set of adjacent city, the correlation characteristics is filtered from the air quality data set of target city by FCM algorithm, and the key feature data set is established;S3, the key feature data set is input into air particulate matter concentration detection model, and the air particulate matter concentration prediction result is output.Attention module is better to capture the change rule of pollutant concentration by comprehensively extracting the time feature and spatial feature of air pollutant concentration.Bidirectional long short-term memory neural network makes full use of the information of key features forward and backward, can quickly respond to urban environmental changes, improve air particulate matter concentration prediction accuracy and reliability.

Description

Technical Field

[0001] This invention belongs to the field of air particulate matter concentration monitoring, and specifically relates to a method for predicting the concentration of particulate matter in urban forest spaces. Background Technology

[0002] Currently, urban forests, as an important component of urban ecosystems, play a significant role in improving urban environmental quality, mitigating the urban heat island effect, and reducing air pollution. However, existing methods and technologies for assessing fine particulate matter concentration in urban forest spaces still have certain limitations. Traditional assessment methods often rely on fixed-point monitoring, which obtains fine particulate matter concentration data by setting up monitoring stations at specific locations. Although this method can obtain relatively accurate data, it is limited by the number and distribution of monitoring stations and cannot comprehensively reflect the overall situation of particulate matter concentration in urban forest spaces. Summary of the Invention

[0003] To address the aforementioned shortcomings in existing technologies, this invention provides a method for predicting spatial particulate matter concentration in urban forests, which solves the problem of low accuracy in predicting spatial particulate matter concentration using existing methods.

[0004] To achieve the above-mentioned objectives, the technical solution adopted by this invention is: a method for predicting the concentration of particulate matter in urban forest spaces, comprising the following steps: S1. Obtain hourly observation data of air pollutant concentrations, urban meteorological data, and AQI in the target area of ​​urban forest space, and establish an urban air quality dataset; S2. Based on the air quality datasets of neighboring cities, relevant features are selected from the air quality datasets of the target city using the FCM algorithm to establish a key feature dataset; S3. Input the key feature dataset into the air particulate matter concentration detection model and output the air particulate matter concentration prediction results.

[0005] Furthermore: In S1, the urban forest spatial target area is divided into multiple areas, and the division criteria can be tiered from the outside to the inside; air pollutants include ozone, carbon monoxide, carbon dioxide, nitrogen dioxide, and sulfur dioxide; the preprocessing methods for air pollutant concentrations and AQI include data normalization, deletion, and interpolation, and preprocessing is used to ensure the integrity and consistency of the data.

[0006] Furthermore: In S2, the objective function of the FCM algorithm The specific expression is as follows: In the formula, For the first j The location of the area. For the firstk Cluster centers, For the first j Location of each area and the k Cluster centers The differences between them for Belongs to the k Cluster centers membership degree m As a weighted index, C The total number of clusters, N The total number of regions, M The number of neighboring cities, To adjust the parameters.

[0007] Furthermore: In S3, the air particulate matter concentration detection model includes interconnected attention modules and a bidirectional long short-term memory neural network; the specific workflow of the air particulate matter concentration detection model is as follows: S31. Input the key feature dataset into the attention module to capture the spatiotemporal features of air pollutant concentration and meteorological data, and obtain the key features of air particulate matter. S32. Input the key features of air pollutants into a bidirectional long short-term memory neural network and output the predicted concentration of air particulate matter.

[0008] Furthermore: In S31, the attention module extracts features. The expression is as follows: In the formula, The input features are those of the attention module. It is the Sigmoid activation function. For spatial attention submodule, For the time attention submodule, For Kronecker product.

[0009] Further: Spatial Attention Submodule The specific expression is: In the formula, For max pooling, For average pooling, This is a convolution operation; Time Attention Submodule The specific expression is: In the formula, It is a multilayer perceptron.

[0010] Furthermore, in S32, the bidirectional long short-term memory neural network optimizes its model parameters using an improved PSO algorithm. These parameters include the learning rate and the number of hidden units. The specific optimization method is as follows: S321. Set the maximum number of iterations, initialize the number of iterations, and set the size of the particle population according to the model parameters of the bidirectional long short-term memory neural network. S322. Calculate the fitness of each particle in the particle population at the current iteration number, adjust the inertia weight according to the fitness, and then update the position of the particles in the particle population. S323. Determine whether the number of iterations has reached the maximum number of iterations. If yes, update the position of the particle population as the global optimal position and generate the optimized bidirectional long short-term memory neural network model parameters based on the optimal position. If no, increment the current iteration count by 1 and return to S322.

[0011] Furthermore: In S322, the expression for updating the position of particles in the particle population is specifically as follows: In the formula, For the first During the nth iteration The position of the particle, For the first During the nth iteration The position of the particle, This represents the current iteration number. For the first During the nth iteration The speed of the particles, and As a learning factor, and A random number within the interval [0,1]. For the first During the nth iteration The historical optimal position of the particle For the first The historical best position of the group at the next iteration. For the first Inertia weights in the next iteration.

[0012] Furthermore: Inertia weight in the next iteration The specific expression is: In the formula, The maximum number of iterations. For the fitness of particles, This is the critical fitness value for the particle.

[0013] The beneficial effects of this invention are as follows: (1) This invention provides a method for predicting the concentration of particulate matter in urban forest spaces. It collects data on six major pollutants, air quality index, and meteorological data, and combines an attention module and a bidirectional long short-term memory neural network to construct an air particulate matter concentration detection model to predict future air particulate matter concentrations. Compared to traditional single-point or single-city predictions, the attention module comprehensively extracts the temporal and spatial features of air pollutant concentrations, avoiding the problem of separately extracting data that might disrupt the inherent correlation between data, thus better capturing the changing patterns of pollutant concentrations. The bidirectional long short-term memory neural network fully utilizes the forward and backward information of key features, enabling rapid response to changes in the urban environment, thereby improving the accuracy and reliability of air particulate matter concentration prediction.

[0014] (2) This invention uses the FCM algorithm to adaptively extract highly correlated neighboring city data and target city data as input to achieve air particulate matter concentration prediction at the relevant city scale. Compared with traditional single-point prediction or single-city prediction, regional prediction can more comprehensively grasp the diffusion pattern and spatiotemporal correlation characteristics of air particulate matter concentration.

[0015] (3) The present invention uses an improved PSO algorithm to optimize the model parameters of the bidirectional long short-term memory neural network. By setting dynamic nonlinear inertial weights, the search speed and global optimization capability of the PSO algorithm are improved, thereby further enhancing the accuracy of the air particulate matter concentration detection model prediction. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for predicting particulate matter concentration in urban forest spaces according to the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0018] like Figure 1 As shown, in one embodiment of the present invention, a method for predicting the spatial particulate matter concentration in urban forests includes the following steps: S1. Obtain hourly observation data of air pollutant concentrations, urban meteorological data, and AQI in the target area of ​​urban forest space, and establish an urban air quality dataset; S2. Based on the air quality datasets of neighboring cities, relevant features are selected from the air quality datasets of the target city using the FCM algorithm to establish a key feature dataset; S3. Input the key feature dataset into the air particulate matter concentration detection model and output the air particulate matter concentration prediction results.

[0019] In S1, the urban forest spatial target area is divided into multiple areas, and the division criteria can be used to divide the area in a tiered manner from the outside to the inside; air pollutants include ozone, carbon monoxide, carbon dioxide, nitrogen dioxide, and sulfur dioxide; Preprocessing methods for air pollutant concentrations and AQI include data normalization, deletion, and interpolation. Preprocessing is used to ensure the integrity and consistency of the data.

[0020] In S2, the objective function of the FCM algorithm is... The specific expression is as follows: In the formula, For the first j The location of the area. For the first k Cluster centers, For the first j Location of each area and the k Cluster centers The differences between them for Belongs to the k Cluster centers membership degree m As a weighted index, C The total number of clusters, N The total number of regions, M The number of neighboring cities, To adjust the parameters.

[0021] In this embodiment, the algorithm minimizes the objective function value by iterating through the membership matrix and cluster center vector, thereby completing the clustering of the regional data most relevant to the current target city and establishing a key feature dataset. This ensures that the input data dimensions of each city region are consistent while preserving the correlation features between them as much as possible.

[0022] In S3, the air particulate matter concentration detection model includes interconnected attention modules and a bidirectional long short-term memory neural network (Bi-LSTM). In this embodiment, the air particulate matter concentration detection model constructed by the present invention adopts a two-layer architecture, which can efficiently extract the spatiotemporal features of the data and generate high-precision prediction results.

[0023] The workflow of the air particulate matter concentration detection model is as follows: S31. Input the key feature dataset into the attention module to capture the spatiotemporal features of air pollutant concentration and meteorological data, and obtain the key features of air particulate matter. S32. Input the key features of air pollutants into a bidirectional long short-term memory neural network (Bi-LSTM) and output the predicted air particulate matter concentration. In S31, the attention module combines temporal and spatial attention sub-modules to capture the spatiotemporal characteristics of air pollutant concentrations and meteorological data. Spatial attention focuses on the importance of data at different spatial locations, while temporal attention is used to mine the dynamic changes of data over time. This combination of attention mechanisms helps to rationally allocate the weights of spatiotemporal features, enhances the model's understanding of key spatiotemporal information, and thus improves the accuracy of pollutant concentration prediction. The attention module extracts features. The expression is as follows: In the formula, The input features are those of the attention module. It is the Sigmoid activation function. For spatial attention submodule, For the time attention submodule, For Kronecker product; In the formula, For max pooling, For average pooling, This is a convolution operation; In the spatial attention submodule, max pooling and average pooling operations are first performed on the input features. Max pooling extracts significant spatial information from the input features, while average pooling focuses on the mean information of the global features. By concatenating the outputs of these two different characteristics, a comprehensive feature map is generated, which can more effectively represent the spatial features of the input data. Subsequently, a convolution operation is used to reduce the dimensionality of the feature map to a single channel, thereby simplifying information representation and reducing computational complexity. After the convolution operation, the feature map is processed using the Sigmoid activation function to generate the final spatial attention weight distribution. The generated spatial attention weight distribution is combined with the original input features through element-wise multiplication, thereby highlighting important spatial features and weakening irrelevant or weakly correlated features. In the formula, It is a multilayer perceptron.

[0024] In the temporal attention submodule, the input data is first subjected to average pooling and max pooling operations to extract features from different dimensions, generating two independent feature vectors. The purpose of these two vectors is to highlight key temporal features in the input data while preserving the diversity of overall features. These vectors are then fed into a shared multilayer perceptron (MLP) for non-linear transformation. The MLP uses a shared weight structure to perform deep processing on the input features to generate more expressive features. Next, the transformed feature vectors are merged using element-wise summation, thus combining the different information from the two types of features. Finally, a sigmoid activation function is used to perform a non-linear transformation on the merged features to generate the weight distribution for temporal attention.

[0025] In S32, the bidirectional long short-term memory neural network optimizes its model parameters using an improved PSO algorithm. These parameters include the learning rate and the number of hidden units. The specific optimization method is as follows: S321. Set the maximum number of iterations, initialize the number of iterations, and set the size of the particle population according to the model parameters of the bidirectional long short-term memory neural network. S322. Calculate the fitness of each particle in the particle population at the current iteration number, adjust the inertia weight according to the fitness, and then update the position of the particles in the particle population. S323. Determine whether the number of iterations has reached the maximum number of iterations. If yes, then the position of the updated particle population is taken as the global optimal position, and the optimized bidirectional long short-term memory neural network model parameters are generated based on the optimal position. If no, then the current iteration count is incremented by 1, and the process returns to S322. In S322, the expression for updating the position of particles in the particle population is as follows: In the formula, For the first During the nth iteration The position of the particle, For the first During the nth iteration The position of the particle, This represents the current iteration number. For the first During the nth iteration The speed of the particles, and As a learning factor, and A random number within the interval [0,1]. For the first During the nth iteration The historical optimal position of the particle For the first The historical best position of the group at the next iteration. For the first Inertia weights in the next iteration; In the formula, The maximum number of iterations. For the fitness of particles, This is the critical fitness value for the particle; In this embodiment, the selection of inertia weights has a significant impact on the algorithm's performance. To improve the global optimization capability and convergence speed of the PSO algorithm, the inertia weights of PSO are improved, establishing an improved particle swarm optimization algorithm. In the initial iteration, a larger weight is used to expand the algorithm's global search range. As iterations progress, the weight is gradually reduced to improve convergence accuracy. While a smaller inertia weight makes the algorithm more likely to converge, it is also prone to getting trapped in local optima. Therefore, this invention proposes dynamically adjusting the inertia weights based on the iteration results, aiming to optimize the performance of the particle swarm optimization algorithm. During the fitness function optimization process, if the change in particle fitness is less than a critical value, the iteration is considered to have fallen into a local optimum. By adjusting the weights in a timely manner during iteration, the search range is continuously corrected, allowing for flexible adjustment of the algorithm's optimization capability, thereby enhancing the efficiency of searching for the global optimum.

[0026] The beneficial effects of this invention are as follows: This invention provides a method for predicting particulate matter concentration in urban forests. It collects data on six major pollutants, air quality index, and meteorological data, and combines an attention module and a bidirectional long short-term memory neural network to construct an air particulate matter concentration detection model. This model predicts future air particulate matter concentrations. Compared to traditional single-point or single-city predictions, the attention module comprehensively extracts the temporal and spatial features of air pollutant concentrations, avoiding the problem of separately extracting data that might disrupt the inherent correlation between data, thus better capturing the changing patterns of pollutant concentrations. The bidirectional long short-term memory neural network fully utilizes forward and backward information from key features, enabling rapid response to changes in the urban environment, thereby improving the accuracy and reliability of air particulate matter concentration prediction.

[0027] This invention uses the FCM algorithm to adaptively extract highly correlated neighboring city data and target city data as input to achieve air particulate matter concentration prediction at the relevant city scale. Compared with traditional single-point prediction or single-city prediction, regional prediction can more comprehensively grasp the diffusion patterns and spatiotemporal correlation characteristics of air particulate matter concentration.

[0028] This invention employs an improved PSO algorithm to optimize the model parameters of a bidirectional long short-term memory neural network. By setting dynamic nonlinear inertial weights, the search speed and global optimization capability of the PSO algorithm are improved, further enhancing the accuracy of air particulate matter concentration detection model prediction.

[0029] In the description of this invention, it should be understood that the terms center, thickness, upper, lower, horizontal, top, bottom, inner, outer, radial, etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms first, second, and third are used for descriptive purposes only and should not be construed as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the feature defined by first, second, and third may explicitly or implicitly include one or more of that feature.

Claims

1. A method for predicting spatial particulate matter concentration in urban forests, characterized in that, Includes the following steps: S1. Obtain hourly observation data of air pollutant concentrations, urban meteorological data, and AQI in the target area of ​​urban forest space, and establish an urban air quality dataset; S2. Based on the air quality datasets of neighboring cities, relevant features are selected from the air quality datasets of the target city using the FCM algorithm to establish a key feature dataset; S3. Input the key feature dataset into the air particulate matter concentration detection model and output the air particulate matter concentration prediction results.

2. The method for predicting spatial particulate matter concentration in urban forests according to claim 1, characterized in that, In S1, the urban forest spatial target area is divided into multiple regions, and the division criteria can be tiered from the outside to the inside; air pollutants include ozone, carbon monoxide, carbon dioxide, nitrogen dioxide, and sulfur dioxide; the preprocessing methods for air pollutant concentrations and AQI include data normalization, deletion, and interpolation, and preprocessing is used to ensure the integrity and consistency of the data.

3. The method for predicting spatial particulate matter concentration in urban forests according to claim 1, characterized in that, In S2, the objective function of the FCM algorithm is... The specific expression is as follows: In the formula, For the first j The location of the area. For the first k Cluster centers, For the first j Location of each area and the k Cluster centers The differences between them For belonging to the first k Cluster centers membership degree m As a weighted index, C The total number of clusters, N The total number of regions, M The number of neighboring cities, To adjust the parameters.

4. The method for predicting spatial particulate matter concentration in urban forests according to claim 1, characterized in that, In S3, the air particulate matter concentration detection model includes interconnected attention modules and a bidirectional long short-term memory neural network; the specific workflow of the air particulate matter concentration detection model is as follows: S31. Input the key feature dataset into the attention module to capture the spatiotemporal features of air pollutant concentration and meteorological data, and obtain the key features of air particulate matter. S32. Input the key features of air pollutants into a bidirectional long short-term memory neural network and output the predicted concentration of air particulate matter.

5. The method for predicting spatial particulate matter concentration in urban forests according to claim 4, characterized in that, In S31, the attention module extracts features. The expression is as follows: In the formula, The input features are those of the attention module. It is the Sigmoid activation function. For spatial attention submodule, For the time attention submodule, For Kronecker product.

6. The method for predicting spatial particulate matter concentration in urban forests according to claim 5, characterized in that, Spatial Attention Submodule The specific expression is: In the formula, For max pooling, For average pooling, This is a convolution operation; Time Attention Submodule The specific expression is: In the formula, It is a multilayer perceptron.

7. The method for predicting spatial particulate matter concentration in urban forests according to claim 4, characterized in that, In S32, the bidirectional long short-term memory neural network optimizes its model parameters using an improved PSO algorithm. These parameters include the learning rate and the number of hidden units. The specific optimization method is as follows: S321. Set the maximum number of iterations, initialize the number of iterations, and set the size of the particle population according to the model parameters of the bidirectional long short-term memory neural network. S322. Calculate the fitness of each particle in the particle population at the current iteration number, adjust the inertia weight according to the fitness, and then update the position of the particles in the particle population. S323. Determine whether the number of iterations has reached the maximum number of iterations. If yes, update the position of the particle population as the global optimal position and generate the optimized bidirectional long short-term memory neural network model parameters based on the optimal position. If no, increment the current iteration count by 1 and return to S322.

8. The method for predicting spatial particulate matter concentration in urban forests according to claim 7, characterized in that, In S322, the expression for updating the position of particles in the particle population is as follows: In the formula, For the first During the nth iteration The position of the particle, For the first During the nth iteration The position of the particle, This represents the current iteration number. For the first During the nth iteration The speed of the particles, and As a learning factor, and A random number within the interval [0,1]. For the first During the nth iteration The historical optimal position of the particle For the first The historical best position of the group at the next iteration. For the first Inertia weights in the next iteration.

9. The method for predicting spatial particulate matter concentration in urban forests according to claim 8, characterized in that, No. Inertia weight in the next iteration The specific expression is: In the formula, The maximum number of iterations. For the fitness of particles, This is the critical fitness value for the particle.

Images