Environment protection monitoring data analysis method and system based on deep learning

Abstract

The application provides an environmental protection monitoring data analysis method and system based on deep learning, which comprises the following steps: first, obtaining an initial monitoring data set containing multiple monitoring data units in an environment area to be analyzed, each monitoring data unit having a timestamp, a spatial position marker and environmental parameter information; then, performing feature extraction processing on the initial monitoring data set to generate dynamic change features and position correlation features, which respectively represent the dynamic change mode of the environmental parameters in the time dimension and the correlation distribution of the environmental parameters among different monitoring positions; then, inputting the dynamic change features and the position correlation features into a pre-constructed deep network model to output environmental parameter progress features containing time dimension trend feature and spatial dimension distribution correlation feature; finally, identifying an abnormal description label of the environment area to be analyzed according to the environmental parameter progress features and determining the label knowledge data, so as to deeply mine the internal rules of the environmental monitoring data and improve the analysis accuracy and comprehensiveness.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and more specifically, to a method and system for analyzing environmental protection monitoring data based on deep learning. Background Technology

[0002] In the field of environmental protection, accurate and comprehensive monitoring data analysis of environmental areas helps in the timely detection of environmental problems, assessment of environmental quality, and the formulation of scientific and reasonable environmental protection strategies. Currently, traditional environmental monitoring data analysis methods mainly rely on manual experience or simple statistical analysis. While manual experience-based analysis methods can incorporate the knowledge of professionals, they suffer from high subjectivity, low efficiency, and difficulty in handling large-scale data, failing to meet the high demands of modern environmental monitoring for real-time data accuracy. Simple statistical analysis methods, while capable of processing a certain amount of data, often only provide superficial processing and struggle to uncover deeper dynamic patterns and complex correlations between different monitoring locations.

[0003] For example, when analyzing air quality monitoring data for a certain region, traditional methods may only yield simple indicators such as the average or maximum air quality index over a certain period, failing to accurately capture the dynamic changes in air quality parameters over time, such as the daily fluctuations in pollutant concentrations and seasonal trends. Furthermore, it is difficult to reveal the correlation distribution of air quality parameters between different monitoring stations, making it impossible to comprehensively monitor the spatial distribution characteristics of air quality in the region. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for environmental protection monitoring data analysis based on deep learning, the method comprising:

[0005] Acquire an initial monitoring data set for the environmental area to be analyzed. The initial monitoring data set contains multiple monitoring data units that are continuously collected and have time stamps and spatial location markers. Each monitoring data unit carries environmental parameter information collected by environmental monitoring equipment at the corresponding time point and geographical location.

[0006] Feature extraction processing is performed on the initial monitoring data set to generate dynamic change features of the monitoring data unit and location correlation features of the monitoring location. The dynamic change features are used to characterize the dynamic change pattern of environmental parameters in the time dimension, and the location correlation features are used to characterize the correlation distribution of environmental parameters between different monitoring locations.

[0007] The dynamic change features and location correlation features are input into a pre-constructed deep network model, and the environmental parameter progress features of the environmental region to be analyzed are output. The environmental parameter progress features include trend features in the time dimension and distribution correlation features in the spatial dimension.

[0008] Based on the environmental parameter progression characteristics, identify the abnormal description tags of the environmental area to be analyzed, and determine the tag knowledge data of the abnormal description tags.

[0009] In another aspect, embodiments of the present invention also provide an environmental protection monitoring data analysis system based on deep learning, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0010] Based on the above, this embodiment of the invention acquires an initial monitoring data set of the environmental area to be analyzed. This initial monitoring data set contains multiple monitoring data units that are continuously collected and have timestamps and spatial location markers. Each monitoring data unit carries detailed environmental parameter information. Feature extraction processing is performed on the initial monitoring data set to generate dynamic change features of the monitoring data units and location correlation features of the monitoring locations. This allows for in-depth mining of the dynamic change patterns of environmental parameters in the time dimension and the correlation distribution of environmental parameters between different monitoring locations. This overcomes the limitations of traditional methods that can only perform surface data analysis. The dynamic change features and location correlation features are input into a pre-constructed deep network model, which outputs the environmental parameter progress features of the environmental area to be analyzed. These features include trend features in the time dimension and distribution correlation features in the spatial dimension, realizing a comprehensive analysis of environmental parameters in time and space. This can more accurately reflect the evolution trend and spatial distribution characteristics of environmental quality. Based on the environmental parameter progress features, anomaly description tags of the environmental area to be analyzed are identified, and tag knowledge data is determined, greatly improving the accuracy, comprehensiveness, and practicality of environmental protection monitoring data analysis. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of the environmental protection monitoring data analysis method based on deep learning provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of exemplary hardware and software components of the deep learning-based environmental protection monitoring data analysis system provided in this embodiment of the invention. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1This is a flowchart illustrating an embodiment of the deep learning-based environmental protection monitoring data analysis method provided by the present invention. The following is a detailed description of the deep learning-based environmental protection monitoring data analysis method.

[0014] Step S110: Obtain the initial monitoring data set of the environmental area to be analyzed. The initial monitoring data set includes multiple monitoring data units that are continuously collected and have timestamps and spatial location markers. Each monitoring data unit carries environmental parameter information collected by the environmental monitoring equipment at the corresponding time point and geographical location.

[0015] In this embodiment, an industrial park is used as the environmental area to be analyzed. Numerous environmental monitoring devices are distributed within the industrial park, continuously collecting environmental parameters. Each monitoring data unit has a clear timestamp and spatial location marker. The timestamp is accurate to the specific moment, recording the exact time the environmental monitoring device collected the data, such as a specific time period in the morning, afternoon, or evening of a particular day. The spatial location marker is represented by geographic coordinates, accurately locating the position of each monitoring device. The environmental parameter information covers multiple aspects, including air quality parameters such as the concentration of various pollutants and the oxygen content in the air; water quality parameters such as the pH of water and the content of certain chemical substances in water; and meteorological parameters such as temperature, humidity, and wind speed. These environmental monitoring devices continuously collect data at set time intervals, thus forming an initial monitoring data set containing multiple monitoring data units. The data in this initial monitoring data set is continuously collected and can reflect the changes in environmental parameters within the industrial park over time and space.

[0016] Step S120: Perform feature extraction processing on the initial monitoring data set to generate dynamic change features of the monitoring data unit and location correlation features of the monitoring location. The dynamic change features are used to characterize the dynamic change pattern of environmental parameters in the time dimension, and the location correlation features are used to characterize the correlation distribution of environmental parameters between different monitoring locations.

[0017] This embodiment requires feature extraction processing of the initial monitoring dataset. The specific feature extraction process includes generating dynamic change features of the monitoring data units and location correlation features of the monitoring locations. Dynamic change features reflect the changing patterns of environmental parameters over time, such as how environmental parameters rise, fall, or fluctuate over time. Location correlation features reflect the correlation distribution of environmental parameters between different monitoring locations, that is, how environmental parameters at different locations influence and are correlated with each other.

[0018] Step S121: Divide the initial monitoring data set into multiple consecutive data segments according to the timestamp order, and each data segment contains monitoring data units within a preset time period.

[0019] When performing feature extraction, the initial monitoring dataset must first be divided. In this embodiment, the initial monitoring dataset is divided into multiple consecutive data segments according to the timestamp order. The preset duration is a time range determined based on actual needs and analysis objectives. For example, a preset duration is used to divide the entire initial monitoring dataset into several data segments, each containing monitoring data units collected within that preset duration. This division aims to facilitate subsequent analysis of environmental parameter changes within each time period.

[0020] Step S122: For each monitoring data unit within a data segment, calculate the difference in environmental parameters between adjacent monitoring data units, construct a parameter change rate sequence in the time dimension by combining the timestamp interval, perform time-series modeling on the parameter change rate sequence through a recurrent neural network, and generate dynamic change features that characterize the dynamic change pattern in the time dimension. The dynamic change features include the periodic fluctuation features and trend continuity features of parameter changes.

[0021] For each segmented data unit, this embodiment further calculates the environmental parameter differences between adjacent monitoring data units. By comparing the environmental parameters of two adjacent monitoring data units, the differences between them are obtained. Then, combined with the timestamp interval of these two monitoring data units, a time-dimensional parameter change rate sequence can be constructed. The parameter change rate sequence reflects the changes in environmental parameters per unit time.

[0022] Next, a recurrent neural network (RNN) is used to perform time-series modeling of the parameter change rate sequence. RNNs are capable of processing sequential data and capturing the time dependencies within it. In this process, the RNN learns the patterns and regularities in the parameter change rate sequence. By modeling the parameter change rate sequence, dynamic change features characterizing the dynamic changes over time are generated. These dynamic change features include periodic fluctuations and trend continuity characteristics of parameter changes. Periodic fluctuations indicate the regular fluctuations of environmental parameters within a set time period; for example, the concentration of certain pollutants may periodically increase or decrease at specific times of the day. Trend continuity characteristics reflect the persistence of environmental parameter change trends over time, i.e., whether the environmental parameter continues to rise, continuously decrease, or remains stable.

[0023] Step S1221: Extract the environmental parameter values ​​of each monitoring data unit within the data segment, and calculate the absolute difference of environmental parameters between two adjacent monitoring data units in turn to obtain the parameter change sequence.

[0024] Before calculating the parameter change rate sequence, it is necessary to extract the environmental parameter values ​​of each monitoring data unit within each data segment. Each monitoring data unit has corresponding environmental parameter information, such as the air quality, water quality, and meteorological parameters mentioned earlier. This embodiment extracts these parameter values ​​sequentially and calculates the absolute difference in environmental parameters between adjacent monitoring data units. The environmental parameter values ​​of two adjacent monitoring data units are subtracted, and the absolute value is taken to obtain the parameter change rate sequence. This parameter change rate sequence reflects the magnitude of environmental parameter changes between adjacent monitoring data units.

[0025] Step S1222: Calculate the time interval based on the timestamps of adjacent monitoring data units, divide the parameter change sequence by the corresponding time interval, and generate a parameter change rate sequence per unit time.

[0026] After obtaining the parameter change sequence, it is necessary to combine it with time information to calculate the parameter change rate. This embodiment calculates the time interval between adjacent monitoring data units based on their timestamps. By comparing the timestamps of two monitoring data units, the time difference between them is determined. Then, each parameter change in the parameter change sequence is divided by the corresponding time interval to generate a parameter change rate sequence per unit time. This parameter change rate sequence more accurately reflects the rate of change of environmental parameters per unit time.

[0027] Step S1223: Determine the positive or negative value of the rate of change of each parameter to determine the direction of change of the environmental parameter at adjacent time points. A positive value indicates that the parameter is increasing, and a negative value indicates that the parameter is decreasing.

[0028] After obtaining the parameter change rate sequence, it is necessary to further analyze the direction of change of environmental parameters. In this embodiment, the direction of change of environmental parameters at adjacent time points is determined by judging the positive or negative value of the change rate of each parameter. If the parameter change rate is positive, it indicates that the environmental parameter is increasing at adjacent time points; if the parameter change rate is negative, it indicates that the environmental parameter is decreasing at adjacent time points. This allows for a direct monitoring of the changing trend of environmental parameters over time.

[0029] Step S1224: Using the parameter change rate sequence and change direction sequence as input, a recurrent neural network model containing long short-term memory network units is used to model the parameter change pattern in the time dimension and extract the change trend dependence features at continuous time points.

[0030] This embodiment takes the parameter change rate sequence and change direction sequence as input and feeds them into a recurrent neural network model containing long short-term memory (LSM) network units. LSM network units can effectively handle long-distance dependencies in sequence data; they can remember past information in the sequence and predict future changes based on this information. This recurrent neural network model models the time-dimensional pattern of parameter changes. The model learns the patterns and regularities in the parameter change rate and change direction sequences, thereby extracting the trend dependence features of changes at consecutive time points. These features reflect the interdependencies of environmental parameters between consecutive time points, such as how changes in environmental parameters at one time point affect environmental parameters at the next time point.

[0031] Step S1225: Calculate the fluctuation amplitude and frequency of the dependent feature within a preset time window to generate periodic fluctuation features in the dynamic change features.

[0032] After extracting the trend-dependent features, the amplitude and frequency of these features' fluctuations within a preset time window are further calculated. The preset time window is a pre-defined time range within which the fluctuations of the dependent features are analyzed. The amplitude of the fluctuation represents the range of change of the dependent feature within this time window, while the frequency represents how frequently the dependent feature fluctuates. By calculating the amplitude and frequency, periodic fluctuation features within the dynamic change characteristics are generated. These features can reflect the regular fluctuations of environmental parameters within a set time period.

[0033] Step S1226: Extract the trend continuity feature from the dynamic change features through the hidden layer state output of the recurrent neural network model. The trend continuity feature is used to characterize the ability of the environmental parameter change trend to persist over time.

[0034] The hidden layer states of a recurrent neural network (RNN) model contain information learned by the model during the processing of sequential data. This embodiment extracts the output of the hidden layer states of the RNN model to obtain the trend persistence feature within the dynamic change characteristics. This trend persistence feature characterizes the ability of environmental parameter change trends to persist over time, i.e., whether the upward, downward, or stable trend of environmental parameters will continue for a period of time. For example, if the trend persistence feature indicates that the upward trend of environmental parameters has a strong persistence ability, then the environmental parameters may continue to rise in the future.

[0035] Step S123: Extract the geographic coordinates corresponding to each monitoring data unit, construct the spatial coordinate matrix of the monitoring location, calculate the cosine similarity of environmental parameters of any two monitoring locations at the same time, construct a spatial association weight matrix by combining spatial distance, perform feature learning on the spatial association weight matrix through graph neural network, and generate location association features that characterize the degree of correlation between environmental parameters between monitoring locations. The location association features include parameter similarity features and distance dependence features between locations.

[0036] In addition to generating dynamic change features, it is also necessary to generate location association features of the monitoring locations. This embodiment first extracts the geographic coordinates corresponding to each monitoring data unit; these coordinates accurately represent the location of each monitoring device. Then, a spatial coordinate matrix of the monitoring locations is constructed based on these coordinates, reflecting the spatial relationships between the various monitoring locations.

[0037] Next, the cosine similarity of environmental parameters at any two monitoring locations at the same time is calculated. Cosine similarity is an index that measures the degree of similarity between two vectors. By calculating the cosine similarity of the environmental parameter vectors of two monitoring locations at the same time, the parameter similarity index between them can be obtained. The larger the parameter similarity index, the more similar the environmental parameters of the two locations are.

[0038] Simultaneously, the Euclidean distance between any two monitoring locations is calculated to obtain the spatial distance index. The spatial distance index reflects the spatial distance between two monitoring locations; the larger the index, the greater the spatial distance between the two locations.

[0039] A spatial association weight matrix is ​​constructed based on parameter similarity and spatial distance indices. The value of each element in the spatial association weight matrix is ​​obtained by dividing the parameter similarity index by the spatial distance index, which represents the environmental parameter association weight between two locations.

[0040] Finally, the spatial correlation weight matrix is ​​input into a graph neural network model containing graph convolutional layers. The graph neural network model can process graph-structured data and learns features from the spatial correlation weight matrix through graph convolutional layers. During this process, the graph neural network model aggregates features of the environmental parameters at each monitoring location, extracting location correlation features. These location correlation features include parameter similarity features and distance dependency features between locations. Parameter similarity features reflect the degree of similarity between environmental parameters at different locations, while distance dependency features reflect the relationship between the degree of correlation of environmental parameters and spatial distance; that is, the closer the locations are, the higher the degree of correlation of their environmental parameters is likely to be.

[0041] Step S1231: Extract the latitude and longitude coordinates of the monitoring location corresponding to each monitoring data unit to form a spatial coordinate set containing all monitoring locations, and construct a spatial coordinate matrix with the dimension of the number of monitoring locations.

[0042] Before constructing the spatial correlation weight matrix, it is necessary to obtain the spatial coordinate information of the monitoring locations. In this embodiment, the latitude and longitude coordinates of the monitoring location corresponding to each monitoring data unit are extracted, and the latitude and longitude coordinates of all monitoring locations are collected to form a spatial coordinate set containing all monitoring locations. Then, based on this spatial coordinate set, a spatial coordinate matrix with a dimension equal to the number of monitoring locations is constructed. Each row of this spatial coordinate matrix corresponds to the coordinate information of a monitoring location, and the spatial relationship between the various monitoring locations can be accurately represented through this spatial coordinate matrix.

[0043] Step S1232: For any two monitoring locations at the same time, extract their environmental parameter vectors, calculate the cosine similarity between the two environmental parameter vectors, and obtain the parameter similarity index. The larger the parameter similarity index, the more similar the environmental parameters of the two locations are.

[0044] For any two monitoring locations at the same time, this embodiment extracts their environmental parameter vectors. These vectors contain information on various environmental parameters at that monitoring location at the same time, such as air quality, water quality, and meteorological parameters. Then, the cosine similarity between these two environmental parameter vectors is calculated. Cosine similarity measures the degree of similarity between the two vectors by calculating the cosine of the angle between them. A higher cosine similarity value indicates greater similarity between the two vectors, meaning more similar environmental parameters between the two monitoring locations. The resulting parameter similarity index can intuitively reflect the degree of similarity of environmental parameters between different monitoring locations.

[0045] Step S1233: Calculate the Euclidean distance between any two monitoring locations to obtain the spatial distance index. The larger the spatial distance index, the farther the spatial distance between the two locations.

[0046] In addition to calculating the parameter similarity index, it is also necessary to calculate the spatial distance between any two monitoring locations. This embodiment uses Euclidean distance to measure the spatial distance between two monitoring locations. Euclidean distance refers to the straight-line distance between two points in Euclidean space. By calculating the Euclidean distance between the latitude and longitude coordinates of any two monitoring locations, the spatial distance index is obtained. The larger the spatial distance index, the greater the spatial distance between the two monitoring locations. This spatial distance index can reflect the actual spatial distance relationship between different monitoring locations.

[0047] Step S1234: Based on the parameter similarity index and spatial distance index, construct a spatial association weight matrix. The value of each element in the spatial association weight matrix is ​​obtained by dividing the parameter similarity index by the spatial distance index, which represents the environmental parameter association weight between two locations.

[0048] After obtaining the parameter similarity index and spatial distance index, this embodiment constructs a spatial correlation weight matrix based on these two indices. The value of each element in the spatial correlation weight matrix is ​​obtained by dividing the parameter similarity index by the spatial distance index. This value characterizes the correlation weight of environmental parameters between two monitoring locations; that is, the degree of correlation between the environmental parameters of two locations is influenced by both their parameter similarity and spatial distance. If the parameter similarity of two monitoring locations is high and their spatial distance is short, then the correlation weight between them will be large; conversely, if the parameter similarity is low and the spatial distance is far, then the correlation weight will be small.

[0049] Step S1235: Input the spatial correlation weight matrix into a graph neural network model containing graph convolutional layers, perform feature aggregation on the environmental parameters of each monitoring location, and extract location correlation features.

[0050] In this embodiment, the constructed spatial correlation weight matrix is ​​input into a graph neural network model containing graph convolutional layers. Graph convolutional layers are a crucial component of graph neural networks, enabling convolution operations on graph-structured data. In this graph neural network model, the graph convolutional layers aggregate environmental parameters for each monitoring location. Based on the correlation weights in the spatial correlation weight matrix, they fuse environmental parameter information from adjacent monitoring locations, thereby extracting location-related features. These features reflect the degree of correlation and mutual influence of environmental parameters between different monitoring locations.

[0051] Step S1236: Generate location association features containing multi-level neighbor location information through multi-layer graph convolution operation. The location association features include direct parameter similarity features and indirect distance dependency features between locations.

[0052] To more comprehensively capture the correlations of environmental parameters between different monitoring locations, this embodiment performs multi-layer graph convolution operations. Multi-layer graph convolution operations allow the model to consider neighbor location information at more levels. Through multiple convolution operations, the model can learn the direct and indirect correlations between different monitoring locations. The generated location correlation features include direct parameter similarity features and indirect distance dependency features. Direct parameter similarity features reflect the direct similarity of environmental parameters between adjacent monitoring locations, while indirect distance dependency features reflect the correlations of environmental parameters between different monitoring locations indirectly generated through intermediate locations; these correlations are influenced by spatial distance.

[0053] Step S130: Input the dynamic change features and location correlation features into the pre-constructed deep network model, and output the environmental parameter progress features of the environmental region to be analyzed. The environmental parameter progress features include the trend features in the time dimension and the distribution correlation features in the spatial dimension.

[0054] After generating dynamic change features and location-related features, this embodiment inputs these two features into a pre-built deep network model. The deep network model is a trained model capable of further processing and analyzing the input features. Through processing by the deep network model, the environmental parameter progression features of the analyzed environmental region are output. These environmental parameter progression features include temporal trend features and spatial distribution and correlation features. The trend features reflect the changing trend of environmental parameters over time, such as whether the environmental parameters are increasing, decreasing, or remaining stable. The distribution and correlation features reflect the spatial distribution of environmental parameters and the relationships between different locations.

[0055] Step S131: Organize the dynamic change features into time series features according to timestamp order. The time series features contain parameter fluctuation trajectory information of each monitoring data unit in the time dimension.

[0056] Before inputting dynamic change features into a deep network model, they need to be further organized. In this embodiment, the dynamic change features are organized into time series features according to timestamp order. The timestamp order ensures the temporal continuity of the data, enabling the time series features to accurately reflect the parameter fluctuation trajectory information of each monitoring data unit in the time dimension. For example, time series features can accurately monitor the rise, fall, or fluctuation of environmental parameters at different points in time.

[0057] Step S132: Construct a spatial correlation matrix based on the geographical coordinates of the monitoring locations using the location correlation features. The spatial correlation matrix contains information on the spatial transmission relationship of environmental parameters between each monitoring location.

[0058] For location-related features, this embodiment constructs a spatial correlation matrix based on the geographic coordinates of the monitoring locations. Geographic coordinates accurately represent the geographical location of each monitoring location. By constructing a spatial correlation matrix, the spatial transmission relationship information of environmental parameters between monitoring locations can be accurately reflected. The spatial transmission relationship information includes the propagation and influence of environmental parameters between different monitoring locations, such as how changes in environmental parameters at one location affect the environmental parameters at its neighboring locations.

[0059] Step S133: Input the time series features and spatial correlation matrix into the feature extraction module of the deep network model. The feature extraction module includes a time dimension feature analysis unit and a spatial dimension feature transmission unit.

[0060] In this embodiment, time-series features and spatial correlation matrices are input into the feature extraction module of the deep network model. The feature extraction module is a crucial component of the deep network model, comprising a time-dimensional feature analysis unit and a spatial-dimensional feature transmission unit. The time-dimensional feature analysis unit analyzes the time-series features to extract the characteristic information of environmental parameters in the time dimension. The spatial-dimensional feature transmission unit processes the spatial correlation matrix to analyze the transmission relationships of environmental parameters in the spatial dimension.

[0061] Step S134: The time series features are decomposed into multiple time scales by the time dimension feature analysis unit to separate the periodic features reflecting the periodic changes of environmental parameters and the abnormal features of sudden fluctuations, and a time dimension feature representation is generated. The periodic features are used to characterize the regular changes of environmental parameters under natural time cycles, and the abnormal features are used to characterize the instantaneous fluctuation characteristics of parameters affected by sudden environmental factors.

[0062] The time-scale feature analysis unit performs multi-time-scale periodic decomposition on the input time-series features. Multi-time-scale periodic decomposition is a method that decomposes time-series data into periodic and anomalous components at different time scales. Through this decomposition, periodic features reflecting the periodic changes of environmental parameters and anomalous features indicating sudden fluctuations are separated. Periodic features characterize the regular changes of environmental parameters over natural time cycles. For example, some environmental parameters may exhibit periodic changes within specific time periods each day, week, or month, such as daily morning and evening air quality changes or seasonal water quality changes. Anomalous features characterize the instantaneous fluctuations of parameters affected by sudden environmental factors. For example, sudden industrial accidents or natural disasters may cause drastic fluctuations in environmental parameters within a short period. By separating the time-series features into periodic and anomalous features through multi-time-scale periodic decomposition, a time-dimensional feature representation is generated. This time-dimensional feature representation can more accurately show the different changing characteristics of environmental parameters over time.

[0063] Step S1341: Perform sliding window analysis on the time series features, calculate the mean curve and variance curve of environmental parameters within different time windows, and identify time windows with periodic changes and time windows with abnormal fluctuations.

[0064] When performing multi-timescale periodic decomposition, a sliding window analysis is first applied to the time series characteristics. The sliding window is a fixed-length time interval that moves gradually along the time series characteristics. For the environmental parameter data within each sliding window, its mean curve and variance curve are calculated. The mean curve reflects the average level of the environmental parameter within that time window, while the variance curve reflects the degree of fluctuation of the environmental parameter within that time window. By analyzing the changes in the mean and variance curves, time windows with periodic changes and time windows with abnormal fluctuations can be identified. If the mean curve exhibits periodic fluctuations and the variance curve is relatively stable, then the time window may have periodic changes; if the variance curve suddenly increases, it indicates that there are abnormal fluctuations in the environmental parameter within that time window.

[0065] Step S1342: The periodic time window data is processed using a time series decomposition algorithm to extract the trend features of environmental parameters at the first time scale and the periodic features at the second time scale. The trend features are used to characterize the long-term evolution direction of environmental parameters, and the periodic features are used to characterize the regular changes of environmental parameters in the natural time cycle. The first time scale is larger than the second time scale.

[0066] For the identified time-window data exhibiting periodic changes, a time-series decomposition algorithm is employed. This algorithm breaks down time-series data into components at different time scales. In this embodiment, trend characteristics of environmental parameters at a first time scale and periodic characteristics at a second time scale are extracted. The first time scale is relatively long, and trend characteristics characterize the long-term evolution direction of environmental parameters, such as whether the concentration of a pollutant gradually increases or decreases over a longer period. The second time scale is relatively short, and periodic characteristics characterize the regular changes of environmental parameters within natural time cycles, such as daily, weekly, or monthly periodic changes.

[0067] Step S1343: Perform outlier detection on the abnormal fluctuation time window data, screen out abnormal data points that significantly deviate from the normal trend, and extract the parameter fluctuation amplitude and duration of the abnormal data points as abnormal features. The abnormal features are used to characterize the instantaneous fluctuation characteristics of parameters affected by sudden environmental factors.

[0068] For time-window data exhibiting abnormal fluctuations, outlier detection is performed. Outlier detection is a method to identify data points that significantly deviate from the normal pattern. Outlier detection filters out anomalous data points that significantly deviate from the normal trend. These anomalous data points may be caused by sudden environmental events, such as industrial leaks or natural disasters. The amplitude and duration of parameter fluctuations in the anomalous data points are extracted as anomaly features. The amplitude of parameter fluctuations represents the degree of change of the anomalous data point relative to normal data, while the duration represents the length of time the anomalous fluctuation lasts. These anomaly features can accurately characterize the instantaneous fluctuation characteristics of parameters affected by sudden environmental factors.

[0069] Step S1344: Arrange the trend features, cycle features, and anomaly features in chronological order to generate a time-dimensional feature representation that reflects the changing patterns across multiple time scales.

[0070] After extracting trend features, periodic features, and anomaly features, they are arranged in chronological order. This arrangement aims to generate a time-dimensional feature representation that reflects the changing patterns across multiple time scales. By combining features from different time scales in chronological order, the changes in environmental parameters over time can be more comprehensively displayed, including long-term trend changes, periodic regular changes, and sudden abnormal fluctuations.

[0071] Step S135: The spatial correlation matrix is ​​modeled for geographic proximity through the spatial dimension feature transmission unit. The spatial distance influence weight is calculated based on the relative coordinates of the monitoring locations. The spatial transmission weight matrix is ​​constructed by combining the environmental parameter similarity. Neighborhood information is aggregated for the environmental parameter features of each monitoring location to generate a spatial dimension feature representation. The spatial transmission weight matrix represents the degree to which the correlation of environmental parameters between monitoring locations is jointly affected by geographic distance and parameter similarity.

[0072] The spatial dimension feature transmission unit performs geographic proximity modeling on the input spatial association matrix. Geographic proximity modeling considers the impact of geographic distance between monitoring locations on the association of environmental parameters. Based on the relative coordinates of the monitoring locations, the spatial distance influence weight is calculated. Relative coordinates accurately represent the relative positional relationship between each monitoring location, and the spatial distance influence weight calculated using relative coordinates reflects the degree of influence of geographic distance on the transmission of environmental parameters. Simultaneously, a spatial transmission weight matrix is ​​constructed by combining environmental parameter similarity. Environmental parameter similarity reflects the degree of similarity of environmental parameters between different monitoring locations. Combining the spatial distance influence weight and environmental parameter similarity can more accurately characterize the degree to which the association of environmental parameters between monitoring locations is jointly influenced by geographic distance and parameter similarity. Through the spatial transmission weight matrix, neighborhood information aggregation is performed on the environmental parameter features of each monitoring location. Neighborhood information aggregation refers to fusing the environmental parameter information of adjacent monitoring locations to generate a spatial dimension feature representation. This spatial dimension feature representation contains information on the spatial transmission relationship of environmental parameters between monitoring locations, such as how environmental parameters propagate and influence between adjacent locations.

[0073] Step S1351: Establish a relative coordinate system with the geometric center of the monitoring area as the origin, extract the relative coordinates of each monitoring location and construct a coordinate matrix, and calculate the spatial distance between any two monitoring locations to generate a distance matrix.

[0074] When performing geographic proximity modeling, a relative coordinate system is first established with the geometric center of the monitoring area as the origin. The geometric center is a central point of the monitoring area, and establishing a relative coordinate system with it as the origin facilitates the description of the relative positional relationships of each monitoring location. Then, the relative coordinates of each monitoring location are extracted and constructed into a coordinate matrix. Each row of the coordinate matrix corresponds to the relative coordinate information of a monitoring location. Next, the spatial distance between any two monitoring locations is calculated, generating a distance matrix. Each element in the distance matrix represents the spatial distance between two monitoring locations, and this distance matrix allows for accurate monitoring of the spatial distance relationships between each monitoring location.

[0075] Step S1352: Calculate the similarity of environmental parameter vectors at any two monitoring locations at the same time to generate a similarity matrix. The greater the similarity of the environmental parameter vectors, the higher the correlation between the environmental parameters of the two locations.

[0076] In addition to calculating spatial distance, it is also necessary to calculate the similarity of environmental parameter vectors between any two monitoring locations at the same time. Each environmental parameter vector contains information about various environmental parameters at that monitoring location at the same time. By calculating the similarity of the standardized parameter values ​​of the two environmental parameter vectors, a similarity matrix is ​​generated. Each element in the similarity matrix represents the similarity of the environmental parameter vectors of the two monitoring locations at the same time. The higher the similarity, the stronger the correlation of the feature patterns of the environmental parameters between the two locations. Through the similarity matrix, the degree of similarity and correlation of environmental parameters between different monitoring locations can be monitored.

[0077] Step S1353: Transform the distance matrix using a distance decay function to generate a spatial distance influence weight matrix that decreases as the distance increases.

[0078] To more accurately describe the impact of geographical distance on the transmission of environmental parameters, this embodiment transforms the distance matrix using a distance decay function. The distance decay function is a function whose value gradually decreases as distance increases. Using the distance decay function, the distance matrix is ​​converted into a spatial distance influence weight matrix. The elements in the spatial distance influence weight matrix represent the influence weight of the spatial distance between two monitoring locations on the transmission of environmental parameters, and this influence weight decreases as the spatial distance increases. This means that the farther apart the monitoring locations are, the greater the geographical distance influence on the transmission of environmental parameters between them.

[0079] Step S1354: Multiply the corresponding elements of the similarity matrix and the spatial distance influence weight matrix to obtain a spatial transmission weight matrix that combines distance and similarity. The element values ​​of the spatial transmission weight matrix represent the transmission contribution of environmental parameters between two monitoring locations.

[0080] Multiplying the similarity matrix by the corresponding elements of the spatial distance influence weight matrix yields a spatial transmission weight matrix that combines distance and similarity. The value of each element in this spatial transmission weight matrix is ​​obtained by multiplying the elements at corresponding positions in the similarity matrix and the spatial distance influence weight matrix. The element values ​​of the spatial transmission weight matrix characterize the transmission contribution of environmental parameters between two monitoring locations, i.e., the degree of influence of environmental parameters from one monitoring location on environmental parameters from another. This degree of influence is jointly affected by geographical distance and environmental parameter similarity.

[0081] Step S1355: Normalize the spatial transmission weight matrix to generate a standardized weight matrix for neighborhood feature aggregation.

[0082] To ensure comparability and additivity of elements in the spatial transmission weight matrix, it is normalized. Normalization involves scaling the elements of the matrix to meet predetermined conditions. This normalization process generates a standardized weight matrix for neighborhood feature aggregation. The element values ​​in the standardized weight matrix fall within a defined range, facilitating neighborhood feature aggregation operations.

[0083] Step S1356: Perform matrix operations on the standardized weight matrix and the environmental parameter feature vectors of each monitoring location to aggregate the surrounding information of each monitoring location feature and generate a spatial dimension feature representation that includes the influence of geographical proximity. The spatial dimension feature representation includes the diffusion and transmission characteristics of environmental parameters between monitoring locations and the geographical barrier attenuation characteristics.

[0084] A matrix operation is performed between the standardized weight matrix and the environmental parameter feature vectors of each monitoring location. This matrix operation involves weighted summation of the elements in the standardized weight matrix with their corresponding environmental parameter feature vectors, thereby aggregating the surrounding information for each monitoring location. Surrounding information aggregation refers to fusing the environmental parameter information of adjacent monitoring locations according to the weights in the standardized weight matrix. This process generates a spatial dimension feature representation that includes the influence of geographical proximity. This spatial dimension feature representation includes the diffusion and transmission characteristics of environmental parameters between monitoring locations and the geographical barrier attenuation characteristics. The diffusion and transmission characteristics reflect the propagation and diffusion of environmental parameters between adjacent monitoring locations, while the geographical barrier attenuation characteristics reflect the hindering effect of geographical distance on the transmission of environmental parameters; that is, the greater the distance, the weaker the transmission effect of environmental parameters.

[0085] Step S136: Calculate the feature association weights between different time points and spatial locations of the time dimension feature representation and the spatial dimension feature representation through a spatiotemporal interaction mechanism. The feature association weights reflect the intensity of the influence of environmental parameter changes on the transmission of surrounding location parameters at the reference time point.

[0086] After obtaining the temporal and spatial feature representations, a spatiotemporal interaction mechanism is used to calculate the feature association weights between different time points and spatial locations. This mechanism considers the mutual influence of both time and space dimensions. The feature association weights reflect the strength of the influence of environmental parameter changes on the transmission of parameters from surrounding locations at a reference time point. For example, at a specific time point, the environmental parameter changes at one monitoring location may be influenced by the transmission of environmental parameters from multiple surrounding monitoring locations at or before that time point. The spatiotemporal interaction mechanism allows for the calculation of the strength of this influence, i.e., the feature association weights.

[0087] Step S137: Based on the feature association weights, the time dimension feature representation and the spatial dimension feature representation are weighted and fused to generate a comprehensive feature vector containing spatiotemporal coupling relationships.

[0088] Based on the calculated feature association weights, the temporal and spatial feature representations are weighted and fused. Weighted fusion refers to combining elements from both the temporal and spatial feature representations according to their feature association weights. Through weighted fusion, a comprehensive feature vector incorporating spatiotemporal coupling relationships is generated. This comprehensive feature vector considers the interaction between the temporal and spatial dimensions, and can more comprehensively reflect the changes in environmental parameters across these dimensions.

[0089] Step S138: Input the comprehensive feature vector into the output layer of the deep network model, and generate environmental parameter progression features through nonlinear transformation.

[0090] The generated composite feature vector is input into the output layer of the deep network model. The output layer is the last layer of the deep network model, which performs a nonlinear transformation on the input composite feature vector. The nonlinear transformation maps the composite feature vector to a new feature space, thereby generating environmental parameter progression features. Environmental parameter progression features include trend characteristics in the time dimension and distribution correlation characteristics in the spatial dimension, which can more accurately describe the changes and development trends of environmental parameters in the spatiotemporal dimensions of the area under analysis.

[0091] Step S140: Identify the abnormal description tags of the environmental area to be analyzed based on the environmental parameter progression characteristics, and determine the tag knowledge data of the abnormal description tags.

[0092] After obtaining the characteristics of environmental parameter progression, it is necessary to identify anomaly description tags for the environmental area to be analyzed based on these characteristics, and determine the tag knowledge data for these anomaly description tags. Anomaly description tags are tags that provide a detailed description of anomalies occurring in the environmental area, accurately reflecting information such as the type, characteristics, and scope of impact of the anomaly. Tag knowledge data consists of detailed information related to the anomaly description tags, including information related to both time and space dimensions, such as the time range and spatial location of the anomaly.

[0093] Step S141: Perform time-series segmentation on the trend characteristics of the time dimension, dividing them into continuous anomalous phases. Each anomalous phase includes a time start marker, a time end marker, and the trend change pattern within the phase.

[0094] First, the trend characteristics over time are segmented into time series. This segmentation divides the trend characteristics into continuous fluctuation phases based on their changes. Each fluctuation phase has a clear start and end time marker, as well as a pattern of trend change within that phase. The start and end time markers define the time range of the fluctuation phase, while the trend change pattern describes the changing trend of environmental parameters within that phase, such as increase, decrease, or fluctuation.

[0095] Step S1411: Extract time series data of the trend characteristics of the time dimension, wherein the time series data includes trend change values ​​arranged in order of timestamps.

[0096] When performing time-series segmentation, the first step is to extract time-series data that demonstrates trend characteristics over time. This time-series data is arranged in chronological order by timestamps and includes trend change values. These trend change values ​​reflect the changing trends of environmental parameters at different points in time. By utilizing time-series data, the trajectory of environmental parameters' trend changes over time can be accurately monitored.

[0097] Step S1412: Traverse the time series data and identify the turning point of the trend change value. The turning point is the time point when the trend change value changes from rising to falling or from falling to rising.

[0098] The extracted time-series data is traversed to identify inflection points in the direction of trend changes. Inflection points are the points in time when the trend changes from rising to falling or vice versa. These inflection points are key time points in the trend changes of environmental parameters, marking a change in the direction of change. For example, when the trend change in air quality changes from rising to falling, it may mean that air quality has begun to improve.

[0099] Step S1413: Identify the rate abrupt change point of the trend change value, wherein the rate abrupt change point is the time point at which the change in the trend change value per unit time exceeds the baseline change.

[0100] In addition to identifying directional inflection points, it is also necessary to identify abrupt rate changes in the trend change value. A rate abrupt change point is the point in time where the change in the trend change value per unit time exceeds a baseline change. The baseline change is a pre-set threshold used to judge the drastic nature of the trend change. When the change in the trend change value per unit time exceeds the baseline change, it indicates that the environmental parameter has changed rapidly at that point in time, possibly due to a sudden environmental event.

[0101] Step S1414: Merge the direction inflection points and rate change points into key time markers and arrange them in chronological order.

[0102] The identified directional inflection points and rate abrupt change points are merged into key time markers and arranged in chronological order. These key time markers are important time nodes in the trend changes of environmental parameters, dividing the time series data into different stages. Analyzing these key time markers allows for more accurate monitoring of changes in environmental parameters over time.

[0103] Step S1415: Using adjacent key time markers as boundaries, divide the time series data into multiple continuous subsequence intervals, with each subsequence interval corresponding to an abnormal phase.

[0104] The time series data is divided into multiple consecutive sub-series intervals using adjacent key time markers as boundaries. Each sub-series interval corresponds to a period of change, and this division aims to analyze the changing trends of environmental parameters in more detail over different time periods. Each period of change has its unique trend pattern and can be analyzed and studied independently.

[0105] Step S1416: Perform morphological analysis on the trend change values ​​within each subsequence interval, and extract the persistence and volatility characteristics of the trend change within the stage as the trend change pattern.

[0106] A morphological analysis is performed on the trend changes within each anomaly phase. This analysis primarily extracts the persistence and volatility characteristics of the trend changes within each phase. Persistence characteristics reflect the duration and stability of the trend change within that phase, while volatility characteristics reflect the degree of fluctuation in the trend change. By analyzing persistence and volatility characteristics, the changing trend patterns of environmental parameters within each anomaly phase can be described more accurately.

[0107] Step S1417: Record the start and end time markers and trend change patterns for each abnormal phase.

[0108] Record the start and end times of each anomaly phase, as well as the trend changes within that phase. The start and end times define the time frame of the anomaly phase, while the trend changes within that phase describe the changing trends of environmental parameters during that phase. This information is crucial for subsequent anomaly analysis and tag identification.

[0109] Step S142: Perform spatial path tracking processing on the spatial dimension distribution association features to divide them into continuous anomaly propagation path segments. Each anomaly propagation path segment includes spatial start coordinates, spatial end coordinates, and distribution change pattern within the path segment.

[0110] Spatial path tracing is performed on the spatial distribution correlation features. This process aims to identify the propagation paths of environmental parameter anomalies across the spatial dimension. The spatial distribution correlation features are segmented into continuous anomaly propagation path segments. Each segment includes the spatial starting coordinates, spatial ending coordinates, and the distribution variation pattern within the segment. The starting and ending coordinates define the spatial extent of the anomaly propagation path segment, while the distribution variation pattern within the segment describes how the environmental parameters are distributed and changed within that segment, such as how the parameters propagate from the starting position to the ending position and the trend of parameter changes during propagation.

[0111] Step S1421: Extract the spatial location data of the distribution correlation features of the spatial dimension, wherein the spatial location data includes the geographic coordinates of each monitoring location and the corresponding distribution correlation value.

[0112] When performing spatial path tracking, the spatial location data of the distribution correlation features in the spatial dimension is extracted first. This spatial location data includes the geographic coordinates of each monitoring location and its corresponding distribution correlation value. The geographic coordinates determine the spatial location of the monitoring location, while the distribution correlation value reflects the degree of correlation between the environmental parameters of that monitoring location and other monitoring locations. By analyzing the spatial location data, the distribution of environmental parameters in the spatial dimension can be monitored.

[0113] Step S1422: Construct a spatial location connection graph, where nodes in the spatial location connection graph are monitoring locations and edges are the distribution correlation value differences between adjacent monitoring locations.

[0114] Based on the extracted spatial location data, a spatial location connectivity map is constructed. This map is a graph structure where nodes represent monitoring locations and edges represent the differences in distribution correlation values ​​between adjacent monitoring locations. These differences reflect the varying degrees of correlation between environmental parameters between adjacent monitoring locations. By constructing this spatial location connectivity map, the spatial relationships between monitoring locations and the correlation of environmental parameters can be displayed more intuitively.

[0115] Step S1423: Select monitoring locations with distribution correlation values ​​higher than the baseline correlation value as candidate starting points in the spatial location connection map.

[0116] In the spatial location connectivity map, monitoring locations with distribution correlation values ​​higher than the baseline correlation value are selected as candidate starting points. The baseline correlation value is a pre-set threshold used to filter monitoring locations that may be the starting points of anomaly propagation. Monitoring locations with distribution correlation values ​​higher than the baseline correlation value have a higher degree of correlation between their environmental parameters and other locations, making them more likely to be the starting points of anomaly propagation. By selecting candidate starting points, the scope of subsequent path tracing can be narrowed, improving analysis efficiency.

[0117] Step S1424: Starting from each candidate starting point, traverse adjacent monitoring locations along the direction of decreasing distribution correlation value difference, record the sequence of continuously visited monitoring locations, and form the initial propagation path segment.

[0118] Starting from each candidate initiating point, the spatial location connectivity map is traversed along the direction of decreasing distribution correlation value difference. This decreasing direction indicates a gradual change in the correlation degree of environmental parameters. Traversing adjacent monitoring locations along this direction allows for the tracking of possible propagation paths for environmental parameter anomalies. During the traversal, a sequence of consecutively visited monitoring locations is recorded; these sequences form the initial propagation path segments. These initial propagation path segments reflect the preliminary path of possible propagation of environmental parameter anomalies starting from the candidate initiating point.

[0119] Step S1425: Verify the continuity of the initial propagation path segment, remove discontinuous location nodes with abrupt changes in distribution correlation value difference, and retain continuous location nodes to form an effective propagation path segment.

[0120] After obtaining the initial propagation path segment, its continuity needs to be verified. Path continuity verification checks whether the differences in the distribution correlation values ​​between monitoring locations within the initial propagation path segment change continuously. If the difference in the distribution correlation value of a certain location node shows a sudden change, it indicates that the degree of change in the correlation between that node and its neighboring nodes is too drastic, which may not conform to the continuous pattern of anomaly propagation; such a location node should be discarded as a discontinuous location node. Continuous location nodes are retained, forming an effective propagation path segment. An effective propagation path segment more accurately reflects the spatial propagation path of environmental parameter anomalies.

[0121] Step S1426: Perform morphological analysis on the distribution correlation values ​​within each effective propagation path segment, and extract the diffusion and aggregation characteristics of the distribution changes within the effective propagation path segment as the distribution change morphology.

[0122] Morphological analysis was performed on the distribution correlation values ​​within each effective propagation path segment. This analysis primarily extracted the diffusion and clustering characteristics of the distribution changes within the effective propagation path segment. Diffusion characteristics describe how environmental parameter anomalies spread to their surroundings during propagation, such as whether the anomalies spread rapidly or slowly. Clustering characteristics indicate whether environmental parameter anomalies cluster in certain areas, such as whether the anomaly level is higher in a specific spatial region. By analyzing diffusion and clustering characteristics, the spatial distribution patterns of environmental parameter anomalies can be monitored in more detail.

[0123] Step S1427: Record the spatial starting coordinates, spatial ending coordinates, and distribution change pattern within each abnormal propagation path segment.

[0124] Record the spatial starting and ending coordinates of each anomaly propagation path segment, as well as the distribution and variation patterns within that segment. The spatial starting and ending coordinates determine the spatial extent of the anomaly propagation path segment, while the distribution and variation patterns within the segment describe the specific changes in environmental parameters within that segment. This information is crucial for subsequent identification of anomaly description labels and determination of label knowledge data.

[0125] Step S143: Perform spatiotemporal overlap verification on each abnormality stage and the corresponding abnormal propagation path segment, and filter out spatiotemporal correlation feature groups that overlap in time range and are adjacent in spatial location.

[0126] After obtaining the anomaly stages and propagation path segments, spatiotemporal overlap verification is required. This verification checks whether each anomaly stage and its corresponding propagation path segment have temporal and spatial overlap and adjacency relationships. If the time range of an anomaly stage overlaps with the time range of an anomaly propagation path segment, and their spatial locations are adjacent, then they are combined into a spatiotemporal correlation feature group. This group more accurately reflects the spatiotemporal correlation of environmental parameter anomalies.

[0127] Step S144: Input the spatiotemporal correlation feature group into a preset environmental anomaly pattern knowledge base, and determine the anomaly description label through feature morphology matching operation.

[0128] The selected spatiotemporal correlation feature groups are input into a pre-defined environmental anomaly pattern knowledge base. This knowledge base is a database containing various environmental anomaly patterns and their corresponding descriptive tags. Through feature morphology matching, the feature morphologies in the spatiotemporal correlation feature groups are compared with the anomaly patterns in the knowledge base. If a matching anomaly pattern is found, the corresponding anomaly descriptive tag can be determined. Anomaly descriptive tags are tags that provide a detailed description of the anomalies occurring in the environmental region, accurately reflecting information such as the type, characteristics, and scope of impact of the anomaly.

[0129] Step S145: Integrate the time start markers, time end markers, and trend change patterns of all anomaly phases to generate time-dimensional tagged knowledge data.

[0130] By integrating the start and end time markers and trend change patterns of all anomaly phases, time-dimensional tagged knowledge data is generated. This time-dimensional tagged knowledge data records in detail the occurrence of anomalies over time, including the start and end times of the anomalies and their trends within different time periods. This information is of significant reference value for further analysis of the causes and impacts of anomalies.

[0131] Step S146: Integrate the spatial starting coordinates, spatial ending coordinates, and distribution change patterns of all abnormal propagation path segments to generate spatial dimension labeled knowledge data.

[0132] By integrating the spatial starting coordinates, spatial ending coordinates, and distribution patterns of all anomaly propagation path segments, spatially tagged knowledge data is generated. This spatially tagged knowledge data describes the propagation of anomalies in space, including the spatial location of the anomaly's origin, its termination location, and its distribution patterns during propagation.

[0133] Regarding the construction and training process of pre-built deep network models, the necessary modules, layers, and connection relationships of the deep network model are first determined.

[0134] The deep network model mainly consists of a feature extraction module and an output layer. The feature extraction module further includes a temporal feature analysis unit and a spatial feature propagation unit. The temporal feature analysis unit processes time-series features, separating periodic and anomalous features through multi-timescale periodic decomposition. The spatial feature propagation unit processes spatial correlation matrices, generating spatial feature representations through geographic proximity modeling and neighborhood information aggregation. The output of the feature extraction module is input to the output layer, which generates environmental parameter progression features through nonlinear transformations.

[0135] The connections between the layers are as follows: time series features and spatial correlation matrix are input into the time dimension feature analysis unit and spatial dimension feature transmission unit of the feature extraction module, respectively. The outputs of the time dimension feature analysis unit and spatial dimension feature transmission unit are weighted and fused through a spatiotemporal interaction mechanism to generate a comprehensive feature vector, which is then input into the output layer.

[0136] The training data consists of a large amount of historical monitoring data, which also includes monitoring data units with timestamps and spatial location markers. The historical monitoring data undergoes the same processing as the previous steps, namely, extracting dynamic change features and location correlation features, and then organizing them into time series features and spatial correlation matrices, which serve as the input data for the model. Simultaneously, environmental parameter progression features are labeled to represent the training data as tag data, which also includes time-dimensional trend features and spatial distribution correlation features.

[0137] Training parameters include learning rate, number of iterations, and batch size. The learning rate controls the step size of parameter updates during model training, the number of iterations represents the number of rounds the model trains on the training data, and the batch size is the amount of data input into the model for each training iteration.

[0138] The input and label data of the training data are fed into the deep network model. The model begins forward propagation, generating predicted environmental parameter progression features through the feature extraction module and the output layer. Then, the loss value between the predicted environmental parameter progression features and the label data is calculated; the loss value represents the degree of difference between the predicted and actual results. Next, the backpropagation algorithm is used to adjust the model parameters based on the loss value, gradually reducing the loss. This process is repeated until a preset number of iterations is reached or the loss value converges to a small value, completing the model training.

[0139] In the data acquisition process of this embodiment, encryption technology is used to protect potentially privacy-sensitive data in environmental monitoring data, such as production emission data of certain specific enterprises. During the data acquisition phase, a symmetric encryption algorithm is used to encrypt the privacy-sensitive data, converting it into ciphertext for transmission and storage. During the data processing and analysis phase, only authorized modules and personnel can use the decryption key to decrypt the ciphertext data. Simultaneously, a strict access control mechanism is established to manage data access permissions, ensuring that only personnel with the appropriate permissions can access and process privacy-sensitive data. Furthermore, the data storage system is regularly audited to check for data leakage risks, promptly identify and address potential security issues, and ensure the security of privacy-sensitive data.

[0140] Figure 2 The illustration shows exemplary hardware and software components of a deep learning-based environmental protection monitoring data analysis system 100, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the deep learning-based environmental protection monitoring data analysis system 100 and to perform the functions in this application.

[0141] The deep learning-based environmental protection monitoring data analysis system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the deep learning-based environmental protection monitoring data analysis method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0142] For example, the deep learning-based environmental protection monitoring data analysis system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the deep learning-based environmental protection monitoring data analysis system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The deep learning-based environmental protection monitoring data analysis system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0143] For ease of explanation, only one processor is described in the deep learning-based environmental protection monitoring data analysis system 100. However, it should be noted that the deep learning-based environmental protection monitoring data analysis system 100 of this application may also include multiple processors. Therefore, the steps executed by one processor as described in this application may also be executed jointly or individually by multiple processors. For example, if the processor of the deep learning-based environmental protection monitoring data analysis system 100 executes steps A and B, it should be understood that steps A and B may also be executed jointly by two different processors or individually by one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0144] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned environmental protection monitoring data analysis method based on deep learning is implemented.

[0145] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1.A deep learning-based environmental protection monitoring data analysis method, characterized in that, The method includes: Acquire an initial monitoring data set for the environmental area to be analyzed. The initial monitoring data set contains multiple monitoring data units that are continuously collected and have time stamps and spatial location markers. Each monitoring data unit carries environmental parameter information collected by environmental monitoring equipment at the corresponding time point and geographical location. Feature extraction processing is performed on the initial monitoring data set to generate dynamic change features of the monitoring data unit and location correlation features of the monitoring location. The dynamic change features are used to characterize the dynamic change pattern of environmental parameters in the time dimension, and the location correlation features are used to characterize the correlation distribution of environmental parameters between different monitoring locations. The dynamic change features and location correlation features are input into a pre-constructed deep network model, and the environmental parameter progress features of the environmental region to be analyzed are output. The environmental parameter progress features include trend features in the time dimension and distribution correlation features in the spatial dimension. Based on the environmental parameter progression characteristics, identify the anomaly description tags of the environmental area to be analyzed, and determine the tag knowledge data of the anomaly description tags; The step of identifying anomaly description tags for the environmental region to be analyzed based on the environmental parameter progression characteristics, and determining the tag knowledge data for the anomaly description tags, includes: The trend characteristics of the time dimension are processed by time-series segmentation and divided into continuous abnormal phases. Each abnormal phase includes a time start mark, a time end mark, and the trend change pattern within the phase. The spatial dimension distribution correlation features are processed by spatial path tracking and divided into continuous anomaly propagation path segments. Each anomaly propagation path segment includes spatial start coordinates, spatial end coordinates, and distribution change pattern within the path segment. For each abnormal stage and the corresponding abnormal propagation path segment, spatiotemporal overlap verification is performed to screen out spatiotemporal correlation feature groups that overlap in time range and are adjacent in spatial location. The spatiotemporal correlation feature group is input into a preset environmental anomaly pattern knowledge base, and anomaly description tags are determined through feature morphology matching. Integrate the start and end time markers and trend change patterns of all anomaly phases to generate time-dimensional tagged knowledge data; By integrating the spatial starting coordinates, spatial ending coordinates, and distribution change patterns of all abnormal propagation path segments, spatial dimension labeled knowledge data is generated. 2.The deep learning-based environmental protection monitoring data analysis method of claim 1, wherein, The step of performing feature extraction processing on the initial monitoring data set to generate dynamic change features of the monitoring data units and location association features of the monitoring locations includes: The initial monitoring data set is divided into multiple consecutive data segments according to the timestamp order, and each data segment contains monitoring data units within a preset duration. For each monitoring data unit within a data segment, the environmental parameter difference between adjacent monitoring data units is calculated. A parameter change rate sequence in the time dimension is constructed by combining the timestamp interval. The parameter change rate sequence is then modeled in time series using a recurrent neural network to generate dynamic change features that characterize the dynamic change pattern in the time dimension. These dynamic change features include periodic fluctuation features and trend continuity features of parameter changes. Extract the geographic coordinates corresponding to each monitoring data unit, construct a spatial coordinate matrix of the monitoring location, calculate the cosine similarity of environmental parameters between any two monitoring locations at the same time, construct a spatial association weight matrix by combining spatial distance, and perform feature learning on the spatial association weight matrix through a graph neural network to generate location association features that characterize the degree of correlation between environmental parameters between monitoring locations. The location association features include parameter similarity features and distance dependence features between locations. 3.The deep learning-based environmental protection monitoring data analysis method of claim 2, wherein, For each monitoring data unit within a data segment, the environmental parameter differences between adjacent monitoring data units are calculated. A time-dimensional parameter change rate sequence is constructed by combining this sequence with timestamp intervals. A recurrent neural network is then used to perform time-series modeling on this parameter change rate sequence, generating dynamic change features that characterize the dynamic changes over time, including: Extract the environmental parameter values ​​of each monitoring data unit within the data segment, and calculate the absolute difference of environmental parameters between two adjacent monitoring data units in turn to obtain the parameter change sequence. The time interval is calculated based on the timestamps of adjacent monitoring data units, and the parameter change sequence is divided by the corresponding time interval to generate a parameter change rate sequence per unit time. Determine the sign of the rate of change of each parameter to identify the direction of change of the environmental parameter at adjacent time points; a positive value indicates that the parameter is increasing, and a negative value indicates that the parameter is decreasing. Using the parameter change rate sequence and change direction sequence as input, a recurrent neural network model containing long short-term memory network units is used to model the parameter change pattern in the time dimension and extract the change trend dependence features at continuous time points. Calculate the fluctuation amplitude and frequency of the dependent feature within a preset time window to generate periodic fluctuation features in the dynamic change features; By extracting the trend continuity feature from the hidden layer state output of the recurrent neural network model, the trend continuity feature is used to characterize the ability of the environmental parameter change trend to persist over time. 4.The deep learning-based environmental protection monitoring data analysis method of claim 2, wherein, The process involves extracting the geographic coordinates corresponding to each monitoring data unit, constructing a spatial coordinate matrix for the monitoring location, calculating the cosine similarity of environmental parameters between any two monitoring locations at the same time, constructing a spatial association weight matrix based on spatial distance, and performing feature learning on the spatial association weight matrix using a graph neural network to generate location association features characterizing the degree of correlation between environmental parameters between monitoring locations, including: For each monitoring data unit, the latitude and longitude coordinates of the monitoring location are extracted to form a spatial coordinate set containing all monitoring locations, and a spatial coordinate matrix with the dimension of the number of monitoring locations is constructed. For any two monitoring locations at the same time, extract their environmental parameter vectors, calculate the cosine similarity between the two environmental parameter vectors, and obtain the parameter similarity index. The larger the parameter similarity index, the more similar the environmental parameters of the two locations are. Calculate the Euclidean distance between any two monitoring locations to obtain a spatial distance index. The larger the spatial distance index, the farther the spatial distance between the two locations. Based on the parameter similarity index and spatial distance index, a spatial association weight matrix is ​​constructed. The value of each element in the spatial association weight matrix is ​​obtained by dividing the parameter similarity index by the spatial distance index, which represents the environmental parameter association weight between two locations. The spatial correlation weight matrix is ​​input into a graph neural network model containing graph convolutional layers to perform feature aggregation on the environmental parameters of each monitoring location and extract location correlation features. By performing multi-layer graph convolution operations, location association features containing multi-level neighbor location information are generated. These location association features include direct parameter similarity features and indirect distance dependency features between locations. 5.The deep learning-based environmental protection monitoring data analysis method of claim 1, wherein, The process of inputting the dynamic change features and location-related features into a pre-constructed deep network model and outputting the environmental parameter progression features of the area to be analyzed includes: The dynamic change features are organized into time series features in timestamp order, and the time series features contain parameter fluctuation trajectory information of each monitoring data unit in the time dimension; The location association features are used to construct a spatial association matrix based on the geographic coordinates of the monitoring locations. The spatial association matrix contains information on the spatial transmission relationship of environmental parameters between each monitoring location. The time series features and spatial correlation matrix are input into the feature extraction module of the deep network model. The feature extraction module includes a time dimension feature analysis unit and a spatial dimension feature transmission unit. The time-dimensional feature analysis unit performs multi-time-scale periodic decomposition on the time series features, separating the periodic features reflecting the periodic changes of environmental parameters and the abnormal features of sudden fluctuations, generating a time-dimensional feature representation. The periodic features are used to characterize the regular changes of environmental parameters under natural time cycles, and the abnormal features are used to characterize the instantaneous fluctuation characteristics of parameters affected by sudden environmental factors. The spatial correlation matrix is ​​modeled using the spatial dimension feature transmission unit. The spatial distance influence weight is calculated based on the relative coordinates of the monitoring locations. A spatial transmission weight matrix is ​​constructed by combining environmental parameter similarity. Neighborhood information is aggregated for the environmental parameter features of each monitoring location to generate a spatial dimension feature representation. The spatial transmission weight matrix characterizes the degree to which the correlation of environmental parameters between monitoring locations is jointly influenced by geographical distance and parameter similarity. The feature association weights between different time points and spatial locations are calculated through a spatiotemporal interaction mechanism. The feature association weights reflect the intensity of the influence of environmental parameter changes on the transmission of surrounding location parameters at the reference time point. Based on the aforementioned feature association weights, the time-dimensional feature representations and spatial-dimensional feature representations are weighted and fused to generate a comprehensive feature vector containing spatiotemporal coupling relationships; The comprehensive feature vector is input into the output layer of the deep network model, and environmental parameter progression features are generated through nonlinear transformation. 6.The deep learning-based environmental protection monitoring data analysis method of claim 5, wherein, The step of performing multi-timescale periodic decomposition on the time series features through the time dimension feature analysis unit to separate periodic features reflecting periodic changes in environmental parameters and anomalous features reflecting sudden fluctuations includes: Sliding window analysis is performed on the time series features to calculate the mean curve and variance curve of environmental parameters within different time windows, and to identify time windows with periodic changes and time windows with abnormal fluctuations. A time series decomposition algorithm is used to process periodic time window data to extract trend features of environmental parameters at a first time scale and periodic features at a second time scale. The trend features are used to characterize the long-term evolution direction of environmental parameters, and the periodic features are used to characterize the seasonal fluctuation pattern of environmental parameters. The first time scale is larger than the second time scale. Outlier detection is performed on abnormal fluctuation time window data to screen out abnormal data points that significantly deviate from the normal trend. The parameter fluctuation amplitude and duration of abnormal data points are extracted as abnormal features, which are used to characterize parameter mutations caused by sudden environmental events. By arranging trend features, cycle features, and anomaly features in chronological order, a time-dimensional feature representation reflecting the changing patterns across multiple time scales is generated. And, the step of performing geographic proximity modeling on the spatial association matrix through the spatial dimension feature transmission unit, calculating the spatial distance influence weight based on the relative coordinates of the monitoring location, and constructing a spatial transmission weight matrix by combining environmental parameter similarity includes: A relative coordinate system is established with the geometric center of the monitoring area as the origin. The relative coordinates of each monitoring location are extracted and a coordinate matrix is ​​constructed. The spatial distance between any two monitoring locations is calculated to generate a distance matrix. A similarity matrix is ​​generated by calculating the similarity of environmental parameter vectors between any two monitoring locations at the same time. The greater the similarity of the environmental parameter vectors, the higher the correlation between the environmental parameters of the two locations. The distance matrix is ​​transformed by a distance decay function to generate a spatial distance influence weight matrix that decreases as the distance increases. Multiply the similarity matrix by the corresponding elements of the spatial distance influence weight matrix to obtain a spatial transmission weight matrix that combines distance and similarity. The element values ​​of the spatial transmission weight matrix represent the transmission contribution of environmental parameters between two monitoring locations. The spatial transmission weight matrix is ​​normalized to generate a standardized weight matrix for neighborhood feature aggregation. The standardized weight matrix is ​​used to perform matrix operations with the environmental parameter feature vectors of each monitoring location to aggregate the surrounding information of each monitoring location feature and generate a spatial dimension feature representation that includes the influence of geographical proximity. The spatial dimension feature representation includes the diffusion and transmission characteristics of environmental parameters between monitoring locations and the geographical barrier attenuation characteristics. 7.The deep learning-based environmental protection monitoring data analysis method of claim 1, wherein, The step of performing time-series segmentation on the trend characteristics of the time dimension, dividing them into continuous anomalous phases, includes: Extract time series data that shows the trend characteristics of the time dimension, wherein the time series data contains trend change values ​​arranged in order of timestamps; Traverse the time series data to identify the turning points in the direction of trend change values. The turning points in the direction are the time points when the trend change values ​​change from rising to falling or from falling to rising. Identify the rate abrupt change point of the trend change value, wherein the rate abrupt change point is the time point at which the change in the trend change value per unit time exceeds the baseline change. The directional inflection points and rate abrupt change points are combined into key time markers and arranged in chronological order. Using adjacent key time markers as boundaries, the time series data is divided into multiple consecutive subsequence intervals, each subsequence interval corresponding to an abnormal phase; A morphological analysis is performed on the trend change values ​​within each subsequence interval to extract the persistence and volatility characteristics of the trend change within the stage as the trend change pattern. Record the start and end times of each anomaly phase. 8.The deep learning-based environmental protection monitoring data analysis method of claim 1, wherein, The step of performing spatial path tracing processing on the distribution correlation features of the spatial dimension and segmenting them into continuous anomaly propagation path segments includes: Extract spatial location data of the distribution correlation features of the spatial dimension, wherein the spatial location data includes the geographic coordinates of each monitoring location and the corresponding distribution correlation value; Construct a spatial location connection graph, where nodes in the spatial location connection graph are monitoring locations and edges are the differences in distribution correlation values ​​between adjacent monitoring locations; In the spatial location connection map, monitoring locations with distribution correlation values ​​higher than the baseline correlation value are selected as candidate starting points; Starting from each candidate starting point, traverse adjacent monitoring locations along the direction of decreasing distribution correlation value difference, record the sequence of continuously visited monitoring locations, and form the initial propagation path segment; The initial propagation path segment is verified for continuity. Non-continuous location nodes with abrupt changes in distribution correlation value difference are removed, and continuous location nodes are retained to form an effective propagation path segment. Morphological analysis was performed on the distribution correlation values ​​within each effective propagation path segment to extract the diffusion and aggregation characteristics of the distribution changes within the effective propagation path segment as the distribution change morphology. Record the spatial start and end coordinates of each anomaly propagation path segment. 9.A deep learning-based environmental protection monitoring data analysis system, characterized by, The method includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the deep learning-based environmental protection monitoring data analysis method according to any one of claims 1-8.

Images