A Machine Learning-Based Method and System for Urban Agglomeration Environmental Quality Assessment

Abstract

This invention discloses a machine learning-based method and system for assessing the environmental quality of urban agglomerations. The method includes acquiring and preprocessing environmental data from cities within the urban agglomeration; obtaining an environmental status index for each city using the environmental data; clustering cities with similar environments into similar city clusters using the DBSCAN algorithm based on the environmental data; classifying cities corresponding to all noise points as anomalous city clusters; obtaining the spatiotemporal variation characteristics of the similar city clusters using a concatenated architecture of temporal convolutional networks and graph convolutional networks, and combining this with the environmental status index to obtain a first environmental quality score; constructing a pollution propagation model based on the anomalous city clusters and combining this with the first environmental quality score to obtain a second environmental quality score; inputting the data of the urban agglomeration to be evaluated; constructing an urban agglomeration environmental quality assessment system based on the second environmental quality score; and outputting the evaluation results. This method improves the accuracy of urban agglomeration environmental quality assessment and has strong interpretability.

Description

Technical Field

[0001] This invention relates to the field of quality assessment, and in particular to a method and system for assessing the environmental quality of urban agglomerations based on machine learning. Background Technology

[0002] With the rapid advancement of urbanization, urban agglomerations, as the core carriers of regional economic and social development, have an increasingly urgent need for environmental quality assessment methods. Traditional environmental quality assessment methods for urban agglomerations mainly rely on single indicators such as RSEI or linear weighted models, which are difficult to effectively capture the spatiotemporal coupling relationship and complex nonlinear characteristics of multidimensional environmental data in urban agglomerations. Existing assessment systems usually treat all cities as independent samples or directly treat urban agglomerations as a whole, ignoring individual differences among cities. In particular, they fail to identify and model abnormal cities that may have a significant impact on surrounding areas through pollution transmission, resulting in limited accuracy and reliability of assessment results. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for evaluating the environmental quality of urban agglomerations based on machine learning.

[0004] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0005] This invention includes the following steps:

[0006] Environmental data of each city in the urban agglomeration is obtained, the environmental data is preprocessed, and the environmental status index of each city is obtained using the preprocessed environmental data.

[0007] Based on the environmental data, the DBSCAN algorithm is used to cluster cities with similar environments in the urban agglomeration to obtain similar city clusters, and all cities corresponding to noise points are classified as abnormal city clusters.

[0008] The spatiotemporal variation characteristics of the similar city clusters are obtained based on the concatenated architecture of temporal convolutional network and graph convolutional network, and the first environmental quality score is obtained by using the environmental status index and spatiotemporal variation characteristics.

[0009] A pollution propagation model is constructed based on the abnormal city clusters, and a second environmental quality score is obtained using the pollution propagation model and the first environmental quality score.

[0010] An environmental quality evaluation system for urban agglomerations is constructed based on the second environmental quality score. The data of the urban agglomerations to be evaluated are input into the environmental quality evaluation system for urban agglomerations, and the evaluation results are output.

[0011] Furthermore, the method for obtaining environmental data for each city in the urban agglomeration, preprocessing the environmental data, and using the environmental data to obtain an environmental status index for each city includes:

[0012] The environmental data acquired for each city in the urban agglomeration includes air quality compliance rate, water quality compliance rate, centralized drinking water source quality compliance rate, average regional environmental noise level, average traffic noise level, urban heat island ratio index, chemical oxygen demand emission intensity, ammonia nitrogen emission intensity, sulfur dioxide emission intensity, soot and dust emission intensity, nitrogen oxide emission intensity, solid waste emission intensity, ecological land ratio, green coverage rate, environmental protection investment as a percentage of GDP, centralized urban sewage treatment rate, and urban waste harmless treatment rate.

[0013] The preprocessing method for the environmental data includes filling missing values ​​with linear interpolation and detecting and removing outliers using the interquartile range method;

[0014] An environmental status index is obtained using the preprocessed environmental data.

[0015] Furthermore, based on environmental data, the DBSCAN algorithm is used to cluster cities with similar environments within an urban agglomeration to obtain similar city clusters, and cities corresponding to all noise points are classified as anomalous city clusters. This includes:

[0016] The environmental data is Z-score standardized, and a covariance matrix is ​​constructed based on the standardized urban environmental data. The initial neighborhood radius Eps is the square root of the largest eigenvalue of the covariance matrix, and the initial neighborhood minimum sample point MinPts is the square root of the total number of cities in the urban agglomeration rounded up.

[0017] The DBSCAN algorithm is used to cluster cities with similar environmental quality in urban agglomerations. The silhouette coefficient is used as an indicator to evaluate the quality of the DBSCAN clustering results. When the silhouette coefficient is lower than 0.5, Eps is increased by 0.1 steps and MinPts is increased or decreased by 1 step to re-cluster until the silhouette coefficient is greater than 0.5.

[0018] After adjusting the DBSCAN algorithm parameters, clustering is performed. Each cluster is considered a similar city cluster, and data points that are not clustered are marked as noise points. All cities corresponding to noise points are grouped into a set as an abnormal city cluster.

[0019] Furthermore, the method for obtaining the spatiotemporal variation characteristics of the environment of each similar city cluster using the concatenated architecture of temporal convolutional networks and graph convolutional networks, and obtaining the first environmental quality score using the environmental condition index and spatiotemporal variation characteristics, includes:

[0020] The environmental data was aligned by day according to the time series and then divided into training set, validation set and test set in a 7:2:1 ratio.

[0021] Principal component analysis was used to extract features from the preprocessed environmental data. Principal components with a cumulative variance contribution rate greater than 85% were retained to obtain environmental feature vectors among similar city clusters.

[0022] We construct an inter-cluster superadjacency matrix by treating all similar city clusters as supernodes. The elements in the matrix are determined by the following formula:

[0023] ,

[0024] in It is the first superadjacency matrix of the inter-cluster clusters. Line number Column elements, T It is the number of days that environmental data was collected. It is a moment Time index is Similar city clusters and indexes are The angle between environmental feature vectors of similar city clusters Is the index as Similar city clusters and indexes are Geographical center distance between similar city clusters It is the arithmetic mean of the minimum distances between cities in the cluster;

[0025] The training is performed on the training set using a cascaded architecture. First, the temporal features of the environmental data are extracted through a temporal convolutional network, and then the spatial features of cities within similar city clusters are extracted through a graph convolutional network. The mean squared error is used as the loss function, and the Adam optimizer is used as the optimization algorithm. The training is performed on the test set until the loss on the validation set converges and the training stops.

[0026] The spatiotemporal variation feature vector sequence of similar city clusters is used as the input of a multi-head attention mechanism with 8 heads to calculate the attention weights of the inter-cluster super-adjacency matrix to capture the correlation between similar city clusters;

[0027] The spatiotemporal variation features and attention weights are concatenated into a feature vector and input into a multilayer perceptron. The first environmental quality score is then output by combining the environmental status index of cities in similar city clusters.

[0028] Furthermore, the method for constructing a pollution propagation model based on anomalous city clusters includes:

[0029] Geographic and meteorological information of anomalous city clusters is obtained, including wind direction angle, horizontal diffusion coefficient and vertical diffusion coefficient obtained from meteorological stations in the cities of the anomalous city cluster. Atmospheric stability level is calculated by Pasquill-Gifford method, and wind speed is taken as the daily average value of meteorological monitoring station.

[0030] Based on the geographic and meteorological information of cities in the anomalous city cluster, a Gaussian plume diffusion model is constructed for each city in the anomalous city cluster.

[0031] A pollution propagation model is constructed based on the Gaussian plume diffusion model, and the pollution model formula is as follows:

[0032] ,

[0033] in It is a moment The index in the city cluster is The total pollution intensity of cities within the abnormal city clusters affected by the city. M It represents the number of cities within an abnormal city cluster. N It represents the number of similar city clusters. It is the independent variable of the integral. This refers to the Gaussian plume pollution contribution from the city with index m in the urban cluster at the corresponding time point. It is a pollutant attenuation coefficient obtained based on dynamic fitting of environmental data. It is the Euclidean distance between the city with index m in the abnormal city cluster and the city with index j in the city group. It is the cross-cluster propagation coefficient of the city with index m in the similar city cluster with index n and the abnormal city cluster. It is the time delay for pollution to spread from city m in the anomalous city cluster to the similar city cluster with index n.

[0034] Furthermore, the method for obtaining a second environmental quality score using the pollution propagation model and the first environmental quality score includes:

[0035] The formula for calculating the second environmental quality score is as follows:

[0036] ,

[0037] in It is the second environmental quality score. It is the first environmental quality score. It is the total area of ​​similar city clusters. M represents the area of ​​the anomalous city cluster, and M represents the number of cities in the anomalous city cluster. It represents the area of ​​the city with index m in the abnormal city cluster. It is the environmental condition index of the city with index m in the abnormal city cluster. R represents the weighting of pollution emission intensity in the environmental status index, and R is the number of cities in the urban agglomeration. It represents the area of ​​the city with index j in the city cluster. It represents the total pollution intensity of cities in the anomalous city cluster affecting city j within the city cluster. It represents the current pollution intensity of the city with index j in the city cluster.

[0038] Furthermore, the method for constructing an environmental quality evaluation system for urban agglomerations based on the second environmental quality score includes:

[0039] Based on the distribution characteristics of the environmental status index of cities in urban agglomerations, the second environmental quality score is divided into intervals. When the second environmental quality score... The environmental quality of the urban agglomeration is excellent. The environmental quality of the urban agglomeration was good at that time. The environmental quality of the urban agglomeration was medium at that time. The environmental quality of the urban agglomeration was poor at that time.

[0040] The difference between the maximum and minimum environmental status index values ​​of cities in each similar city cluster is calculated. The arithmetic mean of the difference values ​​of similar city clusters is weighted according to the number of cities in the similar city clusters and used as the difference threshold. When the difference between the second environmental quality score and the first environmental quality score is greater than the difference threshold, it is considered that the environmental quality of the city cluster has declined due to the influence of cities in abnormal city clusters.

[0041] A machine learning-based urban agglomeration environmental quality assessment system, characterized by comprising:

[0042] Data acquisition module: used to acquire environmental data of each city in the urban agglomeration, preprocess the environmental data, and use the environmental data to obtain the environmental status index of each city;

[0043] City classification module: used for city classification within city clusters. It uses the DBSCAN algorithm to cluster cities with similar environmental quality in the city cluster to obtain similar city clusters, and classifies all cities corresponding to noise points as abnormal city clusters.

[0044] Scoring Module: Used to obtain environmental quality scores for urban clusters. Based on a concatenated architecture of temporal convolutional networks and graph convolutional networks, it obtains the spatiotemporal variation characteristics of the environment of each similar urban cluster. It uses the environmental status index and spatiotemporal variation characteristics to obtain a first environmental quality score. It constructs a pollution propagation model based on the geographical information of cities in abnormal urban clusters. It uses the pollution propagation model and the first environmental quality score to obtain a second environmental quality score.

[0045] System construction module: used to construct the urban agglomeration quality evaluation system, and to construct the urban agglomeration environmental quality evaluation system based on the second environmental quality score;

[0046] Output module: Used to input the data of the urban agglomeration to be evaluated into the urban agglomeration environmental quality evaluation system and output the evaluation results.

[0047] The beneficial effects of this invention are:

[0048] This invention relates to a machine learning-based method and system for evaluating the environmental quality of urban agglomerations. Compared with existing technologies, this invention has the following technical advantages:

[0049] (1) This invention comprehensively covers multiple environmental elements such as atmosphere, water, noise, and ecology by collecting 17 core environmental data, including air quality compliance rate, water quality compliance rate, pollutant emission intensity, and ecological land use ratio, thus avoiding the one-sidedness of a single indicator;

[0050] (2) This invention optimizes clustering parameters by using covariance matrix and silhouette coefficient, automatically divides cities with similar environments, effectively identifies cities with abnormal environments, solves the problem of traditional methods ignoring individual differences, treats abnormal cities as a separate set, provides clear objects for subsequent pollution propagation modeling, and avoids interference from outliers on the overall evaluation;

[0051] (3) This invention captures the complex environmental relationships within and between similar city clusters through a sequential and graph convolutional network architecture, effectively capturing the spatiotemporal coupling relationship and complex nonlinear characteristics of multidimensional environmental data of city clusters;

[0052] (4) This invention constructs a pollution propagation model, corrects the first score through the pollution propagation model, incorporates abnormal urban area, environmental status index and pollution diffusion intensity, dynamically assesses the attenuation effect of pollution on overall environmental quality, and makes the evaluation results closer to reality. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the steps of the urban agglomeration environmental quality assessment method and system based on machine learning, as described in this invention. Detailed Implementation

[0054] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0055] The present invention provides a machine learning-based method and system for assessing the environmental quality of urban agglomerations, comprising the following steps:

[0056] like Figure 1 As shown, this embodiment includes the following steps:

[0057] Environmental data of each city in the urban agglomeration is obtained, the environmental data is preprocessed, and the environmental status index of each city is obtained using the preprocessed environmental data.

[0058] Based on the environmental data, the DBSCAN algorithm is used to cluster cities with similar environments in the urban agglomeration to obtain similar city clusters, and all cities corresponding to noise points are classified as abnormal city clusters.

[0059] The spatiotemporal variation characteristics of the similar city clusters are obtained based on the concatenated architecture of temporal convolutional network and graph convolutional network, and the first environmental quality score is obtained by using the environmental status index and spatiotemporal variation characteristics.

[0060] A pollution propagation model is constructed based on the abnormal city clusters, and a second environmental quality score is obtained using the pollution propagation model and the first environmental quality score.

[0061] An environmental quality evaluation system for urban agglomerations is constructed based on the second environmental quality score. The data of the urban agglomerations to be evaluated are input into the environmental quality evaluation system for urban agglomerations, and the evaluation results are output.

[0062] In this embodiment, the method for acquiring environmental data of each city in the urban agglomeration, preprocessing the environmental data, and using the environmental data to obtain the environmental status index of each city includes:

[0063] The environmental data acquired for each city in the urban agglomeration includes air quality compliance rate, water quality compliance rate, centralized drinking water source quality compliance rate, average regional environmental noise level, average traffic noise level, urban heat island ratio index, chemical oxygen demand emission intensity, ammonia nitrogen emission intensity, sulfur dioxide emission intensity, soot and dust emission intensity, nitrogen oxide emission intensity, solid waste emission intensity, ecological land ratio, green coverage rate, environmental protection investment as a percentage of GDP, centralized urban sewage treatment rate, and urban waste harmless treatment rate.

[0064] The preprocessing method for the environmental data includes filling missing values ​​with linear interpolation and detecting and removing outliers using the interquartile range method;

[0065] An environmental status index is obtained using the preprocessed environmental data.

[0066] In actual assessments, urban environmental data sources within urban agglomerations include environmental monitoring stations, remote sensing data monitoring stations, and environmental statistics from each city. Environmental data is collected daily. When the sources of environmental data do not publish data daily or when data is missing, the missing values ​​are filled using linear interpolation based on environmental data collected seven times before and after the missing value. Outliers are detected and removed using the interquartile range method. The environmental status index is obtained according to the calculation method for the urban ecological environment status index in the Technical Specifications for Ecological Environment Status Assessment. The formula for the environmental status index is as follows: , It is the normalization coefficient of the i-th item of the collected environmental data, through... The corresponding environmental data items can be obtained from the Urban Ecological Environment Quality Evaluation Index Table in the Technical Specifications for Ecological Environment Status Assessment. This refers to the i-th actual data item of the collected environmental data. Taking a certain urban agglomeration as an example, the urban agglomeration includes 16 cities. According to the latest environmental data, the urban condition index of the urban agglomeration is [75.2, 73.8, 77.1, 76.5, 74.9, 78.3, 68.4, 66.7, 69.2, 67.5, 65.8, 58.9, 56.3, 59.1, 57.7, 50.1].

[0067] In this embodiment, a method for clustering cities with similar environments in an urban agglomeration using the DBSCAN algorithm based on environmental data to obtain similar city clusters, and classifying all cities corresponding to noise points as abnormal city clusters, includes:

[0068] The environmental data is Z-score standardized, and a covariance matrix is ​​constructed based on the standardized urban environmental data. The initial neighborhood radius Eps is the square root of the largest eigenvalue of the covariance matrix, and the initial neighborhood minimum sample point MinPts is the square root of the total number of cities in the urban agglomeration rounded up.

[0069] The DBSCAN algorithm is used to cluster cities with similar environmental quality in urban agglomerations. The silhouette coefficient is used as an indicator to evaluate the quality of the DBSCAN clustering results. When the silhouette coefficient is lower than 0.5, Eps is increased by 0.1 steps and MinPts is increased or decreased by 1 step to re-cluster until the silhouette coefficient is greater than 0.5.

[0070] After adjusting the DBSCAN algorithm parameters, clustering is performed. Each cluster is considered a similar city cluster, and data points that are not clustered are marked as noise points. All cities corresponding to noise points are grouped into a set as an abnormal city cluster.

[0071] In the actual evaluation, the largest eigenvalue of the covariance matrix is ​​2.8, so the initial neighborhood radius Eps is 1.67. The largest eigenvalue of the covariance matrix and the initial minimum neighborhood sample point MinPts are 4. After substituting the initial neighborhood radius and the initial minimum neighborhood sample point into the DBSCAN algorithm for clustering, the silhouette coefficient after initial clustering is 0.42. When the neighborhood radius is adjusted to 1.87 and the minimum sample point is adjusted to 3, the silhouette coefficient is 0.65, and the clustering result is 3 clusters. The number of cities in each cluster is [6, 4, 5]. The environmental status indices of the corresponding city clusters are [75.2, 73.8, 77.1, 76.5, 74.9, 78.3], [68.4, 66.7, 69.2, 67.5, 65.8], and [58.9, 56.3, 59.1, 59.1, 59.1, 56.3 ... [57.7], the clusters that have been clustered are considered as similar city clusters. There is one noise point. The cities corresponding to the noise point are considered as an abnormal city cluster. The environmental condition index of the corresponding city is [50.1].

[0072] In this embodiment, the method for obtaining the spatiotemporal variation characteristics of the environment of each similar city cluster using a concatenated architecture based on temporal convolutional networks and graph convolutional networks, and obtaining a first environmental quality score using the environmental condition index and spatiotemporal variation characteristics, includes:

[0073] The environmental data was aligned by day according to the time series and then divided into training set, validation set and test set in a 7:2:1 ratio.

[0074] Principal component analysis was used to extract features from the preprocessed environmental data. Principal components with a cumulative variance contribution rate greater than 85% were retained to obtain environmental feature vectors among similar city clusters.

[0075] We construct an inter-cluster superadjacency matrix by treating all similar city clusters as supernodes. The elements in the matrix are determined by the following formula:

[0076] ,

[0077] in It is the first superadjacency matrix of the inter-cluster clusters. Line number Column elements, T It is the number of days that environmental data was collected. It is a moment Time index is Similar city clusters and indexes are The angle between environmental feature vectors of similar city clusters Is the index as Similar city clusters and indexes are Geographical center distance between similar city clusters It is the arithmetic mean of the minimum distances between cities in the cluster;

[0078] The training is performed on the training set using a cascaded architecture. First, the temporal features of the environmental data are extracted through a temporal convolutional network, and then the spatial features of cities within similar city clusters are extracted through a graph convolutional network. The mean squared error is used as the loss function, and the Adam optimizer is used as the optimization algorithm. The training is performed on the test set until the loss on the validation set converges and the training stops.

[0079] The spatiotemporal variation feature vector sequence of similar city clusters is used as the input of a multi-head attention mechanism with 8 heads to calculate the attention weights of the inter-cluster super-adjacency matrix to capture the correlation between similar city clusters;

[0080] The spatiotemporal variation features and attention weights are concatenated into a feature vector and input into a multilayer perceptron. The first environmental quality score is output by combining the environmental status index of cities in similar city clusters.

[0081] In the actual assessment, principal component analysis was used to extract features from the preprocessed environmental data, reducing the original 17-dimensional data to 8-dimensional principal components. The environmental feature vector between similar city clusters is the arithmetic mean of the vectors composed of the principal components of all cities within the similar city cluster. The resulting inter-cluster hyperadjacency matrix has a value of 1 if the clusters are completely similar. The inter-cluster hyperadjacency matrix is ​​as follows: [[1,1.06e-1,1.13e-4],[1.06e-1,1,3.72e-3],[1.13e-4,3.72e-3,1]]. The geographical center distance between the similar city clusters is [50, 60, 80], and the arithmetic mean of the minimum distance between cities in the clusters is 20.

[0082] The serial architecture training input layer data dimensions include the number of days of environmental data collection in the training set, the number of cities in similar city clusters, and the number of principal components. The temporal convolutional network adopts 1D convolution with 2 convolutional layers, each with 32 filters and a kernel size of 3. The activation function is ReLU. Temporal features are extracted, and the number of cities in the temporal features is used as the node input graph convolutional network. The features of neighboring nodes are aggregated to output spatiotemporal change features. The multilayer perceptron includes two MLP layers, with 64 neurons in the first layer and 1 neuron in the second layer. The first environmental quality score output of the city cluster is 72.5.

[0083] In this embodiment, the method for constructing a pollution propagation model based on anomalous city clusters includes:

[0084] Geographic and meteorological information of anomalous city clusters is obtained, including wind direction angle, horizontal diffusion coefficient and vertical diffusion coefficient obtained from meteorological stations in the cities of the anomalous city cluster. Atmospheric stability level is calculated by Pasquill-Gifford method, and wind speed is taken as the daily average value of meteorological monitoring station.

[0085] Based on the geographic and meteorological information of cities in the anomalous city cluster, a Gaussian plume diffusion model is constructed for each city in the anomalous city cluster.

[0086] A pollution propagation model is constructed based on the Gaussian plume diffusion model, and the pollution model formula is as follows:

[0087] ,

[0088] in It is a moment The index in the city cluster is The total pollution intensity of cities within the abnormal city clusters affected by the city. M It represents the number of cities within an abnormal city cluster. N It represents the number of similar city clusters. It is the independent variable of the integral. This refers to the Gaussian plume pollution contribution from the city with index m in the urban cluster at the corresponding time point. It is a pollutant attenuation coefficient obtained based on dynamic fitting of environmental data. It is the Euclidean distance between the city with index m in the abnormal city cluster and the city with index j in the city group. It is the cross-cluster propagation coefficient of the city with index m in the similar city cluster with index n and the abnormal city cluster. It is the time delay for pollution to spread from city with index m in the anomalous city cluster to the similar city cluster with index n;

[0089] In actual assessment, Based on the fitting of pollutant degradation over time in environmental data. ,in It is the distance from the city with index m in the abnormal city cluster to the geographical center of the similar city cluster with index n. It represents the maximum distance from city with index m in the abnormal city cluster to cities in the similar city cluster with index n. yes The environmental condition index of the city with index m in the city cluster with time-anomalies. It is the geographically distance-weighted average of the environmental condition index of cities in a similar city cluster with index n.

[0090] In this embodiment, the method for obtaining a second environmental quality score using the pollution propagation model and the first environmental quality score includes:

[0091] The formula for calculating the second environmental quality score is as follows:

[0092] ,

[0093] in It is the second environmental quality score. It is the first environmental quality score. It is the total area of ​​similar city clusters. M represents the area of ​​the anomalous city cluster, and M represents the number of cities in the anomalous city cluster. It represents the area of ​​the city with index m in the abnormal city cluster. It is the environmental condition index of the city with index m in the abnormal city cluster. R represents the weighting of pollution emission intensity in the environmental status index, and R is the number of cities in the urban agglomeration. It represents the area of ​​the city with index j in the city cluster. It represents the total pollution intensity of cities in the anomalous city cluster affecting city j within the city cluster. It represents the current pollution intensity of the city with index j in the urban agglomeration;

[0094] In the actual assessment, the number of cities in this urban agglomeration, R, is 16, and the weighting of pollution emission intensity in the environmental status index is... The value is 0.2, corresponding to the city. The sequence is [0.066, 0.129, 0.065, 0.077, 0.074, 0.122, 0.066, 0.056, 0.065, 0.0625, 0.054, 0.023, 0.0218, 0.023, 0.0226, 1.0], the second environmental quality score. It is 69.2.

[0095] In this embodiment, the method for constructing an environmental quality evaluation system for urban agglomerations based on the second environmental quality score includes:

[0096] Based on the distribution characteristics of the environmental status index of cities in urban agglomerations, the second environmental quality score is divided into intervals. When the second environmental quality score... The environmental quality of the urban agglomeration is excellent. The environmental quality of the urban agglomeration was good at that time. The environmental quality of the urban agglomeration was medium at that time. The environmental quality of the urban agglomeration was poor at that time.

[0097] The difference between the maximum and minimum environmental status index values ​​of cities in each similar city cluster is calculated. The arithmetic mean of the difference values ​​of similar city clusters is weighted according to the number of cities in the similar city clusters and used as the difference threshold. When the difference between the second environmental quality score and the first environmental quality score is greater than the difference threshold, it is considered that the environmental quality of the city cluster has declined due to the influence of cities in abnormal city clusters.

[0098] In the actual assessment, based on the second environmental quality score of 69.2 for the urban agglomeration, the environmental quality of the urban agglomeration is medium. The first environmental quality score is 72.5, and the difference between the second and first environmental quality scores is 3.3. The difference threshold in the urban agglomeration is calculated to be 3.68. Therefore, the environmental quality of the urban agglomeration has not been affected by the abnormal urban cluster, which has caused a decline in the environmental quality of the urban agglomeration.

[0099] A machine learning-based urban agglomeration environmental quality assessment system, characterized by comprising:

[0100] Data acquisition module: used to acquire environmental data of each city in the urban agglomeration, preprocess the environmental data, and use the environmental data to obtain the environmental status index of each city;

[0101] City classification module: used for city classification within city clusters. It uses the DBSCAN algorithm to cluster cities with similar environmental quality in the city cluster to obtain similar city clusters, and classifies all cities corresponding to noise points as abnormal city clusters.

[0102] Scoring Module: Used to obtain environmental quality scores for urban clusters. Based on a concatenated architecture of temporal convolutional networks and graph convolutional networks, it obtains the spatiotemporal variation characteristics of the environment of each similar urban cluster. It uses the environmental status index and spatiotemporal variation characteristics to obtain a first environmental quality score. It constructs a pollution propagation model based on the geographical information of cities in abnormal urban clusters. It uses the pollution propagation model and the first environmental quality score to obtain a second environmental quality score.

[0103] System construction module: used to construct the urban agglomeration quality evaluation system, and to construct the urban agglomeration environmental quality evaluation system based on the second environmental quality score;

[0104] Output module: Used to input the data of the urban agglomeration to be evaluated into the urban agglomeration environmental quality evaluation system and output the evaluation results.

[0105] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for evaluating the environmental quality of an urban agglomeration based on machine learning, characterized in that, The method comprises the following steps: obtaining environmental data of each city in the urban agglomeration, preprocessing the environmental data, and obtaining an environmental condition index of each city by using the preprocessed environmental data; clustering cities with similar environments in the urban agglomeration to obtain a similar city cluster by using a DBSCAN algorithm according to the environmental data, and classifying cities corresponding to all noise points into an abnormal city cluster; obtaining a spatiotemporal change feature of the similar city cluster based on a series connection architecture of a time series convolution network and a graph convolution network, and obtaining a first environmental quality score by using the environmental condition index and the spatiotemporal change feature; constructing a pollution propagation model according to the abnormal city cluster, and obtaining a second environmental quality score by using the pollution propagation model and the first environmental quality score; constructing an urban agglomeration environmental quality evaluation system based on the second environmental quality score, inputting data of a to-be-evaluated urban agglomeration into the urban agglomeration environmental quality evaluation system, and outputting an evaluation result; The method for constructing the pollution propagation model according to the abnormal city cluster comprises: obtaining geographical and meteorological information of the abnormal city cluster, including a wind direction angle, a horizontal diffusion coefficient, and a vertical diffusion coefficient monitored by a meteorological station in the city in the abnormal city cluster, and an atmospheric stability degree calculated by using a Pasquill-Gifford method, and a daily average wind speed of the meteorological monitoring station; constructing a Gaussian plume diffusion model for the abnormal city cluster in each city according to the geographical and meteorological information of the city in the abnormal city cluster; constructing the pollution propagation model according to the Gaussian plume diffusion model, and the pollution model formula is as follows: ； ； in It is a moment The index in the city cluster is The total pollution intensity of cities within the abnormal city clusters affected by the city. It represents the number of cities within an abnormal city cluster. It represents the number of similar city clusters. It is the independent variable of the integral. It is an index in the city cluster The city was affected by the abnormal city cluster index at the corresponding time. The contribution of Gaussian plume pollution to the city, It is a pollutant attenuation coefficient obtained based on dynamic fitting of environmental data. The index in the abnormal city cluster is The index in cities and city clusters is The geographical European distance of the city Is the index as The indexes in similar city clusters and anomalous city clusters are The cross-cluster propagation coefficient of the city, Pollution is indexed from anomalous city clusters. The city spread to the index is The time delay of similar city clusters. 2.The method of claim 1, wherein, The method for obtaining the environmental data of each city in the urban agglomeration, preprocessing the environmental data, and obtaining the environmental condition index of each city by using the environmental data comprises: The environmental data of each city in the urban agglomeration comprises an air quality compliance rate, a water quality compliance rate, a centralized drinking water source water quality compliance rate, a regional environmental noise average value, a traffic trunk line noise average value, a city heat island proportion index, a chemical oxygen demand emission intensity, an ammonia nitrogen emission intensity, a sulfur dioxide emission intensity, a smoke dust emission intensity, a nitrogen oxide emission intensity, a solid waste emission intensity, an ecological land proportion, a green land coverage, an environmental protection investment proportion of GDP, a municipal sewage centralized treatment rate, and a city garbage harmless treatment rate; The preprocessing method of the environmental data comprises using a linear interpolation method to fill in missing values and using a quartile range method to detect and remove abnormal values; The environmental condition index is obtained by using the preprocessed environmental data. 3.The method of claim 1, wherein, The method for clustering cities with similar environments in the urban agglomeration to obtain a similar city cluster by using a DBSCAN algorithm according to the environmental data, and classifying cities corresponding to all noise points into an abnormal city cluster comprises: performing Z-Score standardization on the environmental data, constructing a covariance matrix according to the standardized city environmental data, taking a square root of a maximum eigenvalue of the covariance matrix as an initial neighborhood radius Eps, and taking an upward integer square root of a total number of cities in the urban agglomeration as an initial neighborhood minimum sample point MinPts. The cities with similar environmental quality conditions in the urban agglomeration are clustered by the DBSCAN algorithm, and the silhouette coefficient is used as an index for evaluating the quality of the clustering results of DBSCAN. When the silhouette coefficient is less than 0.5, the Eps is increased by 0.1 step, and the MinPts is increased or decreased by 1 step according to the silhouette coefficient, until the silhouette coefficient is greater than 0.5; After adjusting the parameters of the DBSCAN algorithm for clustering, each completed cluster of each cluster is regarded as a similar city cluster, and the data points that are not clustered are marked as noise points. The cities corresponding to all noise points are grouped into a set as an abnormal city cluster. 4.The method of claim 1, wherein, The time-space change feature of each similar city cluster is obtained based on the series connection architecture of the time series convolution network and the graph convolution network. The method for obtaining the first environmental quality score by using the environmental condition index and the time-space change feature includes: After aligning the environmental data by time series per day, the data is divided into a training set, a validation set and a test set in a ratio of 7:2:1; The principal component analysis method is used to extract features from the preprocessed environmental data, and the principal components with a cumulative variance contribution rate greater than 85% are retained to obtain the environmental feature vector between similar city clusters; All similar city clusters are regarded as super nodes to construct an inter-cluster super adjacency matrix, and the elements in the matrix are determined by the following formula: ； in It is the first cluster superadjacency matrix. Line number Column elements, It is the number of days that environmental data was collected. It is a moment Time index is Similar city clusters and indexes are The angle between environmental feature vectors of similar city clusters Is the index as Similar city clusters and indexes are Geographical center distance between similar city clusters It is the arithmetic mean of the minimum distances between cities in the cluster; The series connection architecture is trained on the training set. The time series features of the environmental data are extracted by the time series convolution network, and the spatial features of the cities in the similar city cluster are extracted by the graph convolution network. The mean square error is used as the loss function, and the Adam optimizer is used as the optimization algorithm. The training is stopped until the loss on the validation set converges; The time-space change feature vector sequence of the similar city cluster is input into the multi-head attention mechanism with a head number of 8 to calculate the attention weight of the inter-cluster super adjacency matrix to capture the correlation between similar city clusters; The time-space change feature and the attention weight are concatenated into a feature vector to input a multilayer perceptron, and the first environmental quality score is output combined with the environmental condition index of the city in the similar city cluster. 5.The method of claim 1, wherein, The method for obtaining the second environmental quality score by using the pollution propagation model and the first environmental quality score includes: The calculation formula of the second environmental quality score is as follows: ； in It is the second environmental quality score. It is the first environmental quality score. It is the total area of ​​similar city clusters. It is the area of ​​an abnormal city cluster. It represents the number of cities in the abnormal city cluster. It represents the area of ​​the city with index m in the abnormal city cluster. It is the environmental condition index of the city with index m in the abnormal city cluster. It is the weighting ratio of pollution emission intensity to the environmental status index. It refers to the number of cities within an urban agglomeration. It is an index in the city cluster The city's area, It is an index in the city cluster The total pollution intensity of cities within the abnormal city clusters affected by the city. It is an index in the city cluster The current pollution level of the city. 6.The method of claim 1, wherein, The method for constructing the urban agglomeration environmental quality evaluation system based on the second environmental quality score includes: According to the distribution characteristics of the environmental condition indexes of the cities in the urban agglomeration, the second environmental quality score is divided into intervals, and when the second environmental quality score is the environmental quality of the urban agglomeration is excellent, the environmental quality of the urban agglomeration is good, the environmental quality of the urban agglomeration is medium, the environmental quality of the urban agglomeration is poor. The difference value between the maximum value and the minimum value of the environmental condition index of each city in each similar city cluster is calculated. The arithmetic mean of the difference values of the similar city clusters weighted according to the number of cities in the similar city clusters is regarded as a difference threshold. When the difference between the second environmental quality score and the first environmental quality score is greater than the difference threshold, it is considered that the urban agglomeration environmental quality is affected by the cities in the abnormal city cluster, and the urban agglomeration environmental quality is decreased.

7. The urban agglomeration environment quality evaluation system based on machine learning, for performing the urban agglomeration environment quality evaluation method based on machine learning according to any one of claims 1 to 6, characterized in that, The system includes: A data acquisition module is configured to acquire environmental data of each city in an urban agglomeration, preprocess the environmental data, and obtain an environmental condition index of each city by using the preprocessed environmental data. The city classification module is configured to classify cities in the urban agglomeration, to cluster cities with similar environmental quality conditions in the urban agglomeration by using a DBSCAN algorithm to obtain similar city clusters, and to classify cities corresponding to all noise points as an abnormal city cluster; The scoring module is configured to obtain an environmental quality score of the urban agglomeration, to obtain spatiotemporal variation characteristics of each similar city cluster based on a cascaded architecture of a time series convolution network and a graph convolution network, to obtain a first environmental quality score by using the environmental condition index and the spatiotemporal variation characteristics, to construct a pollution propagation model according to the abnormal city cluster, and to obtain a second environmental quality score by using the pollution propagation model and the first environmental quality score; The system construction module is configured to construct an urban agglomeration quality evaluation system based on the second environmental quality score; The output module is configured to input data of an urban agglomeration to be evaluated into the urban agglomeration environmental quality evaluation system and output an evaluation result.

Images