A Big Data Analysis Method for Marine Pollution Source Tracing Based on Spatio-Temporal Graph Neural Network

Abstract

The present invention discloses a method for analyzing ocean pollution source tracing big data based on a spatiotemporal graph neural network. The method is applied to the field of ocean monitoring. The method involves extracting seawater pollution diffusion characteristics based on periodic water quality data and ocean flow field maps of a target sea area, processing these seawater pollution diffusion characteristics based on a graph neural network, obtaining a pollution diffusion network for pollutants in the target sea area, and reconstructing the pollution diffusion network using ocean surface remote sensing data. The method then adjusts the nodes and edges of the pollution diffusion network based on the results of a two-way propagation analysis to obtain a target pollution diffusion network. The method can more accurately simulate the diffusion path of pollutants, thereby improving the accuracy of pollution source location, better adapting to actual marine pollution monitoring conditions, and enhancing the reliability of tracing results.

Description

Technical Field

[0001] The present invention relates to the field of ocean monitoring technology, and in particular to a method for analyzing ocean pollution source tracing big data based on a spatiotemporal graph neural network. Background Art

[0002] Existing methods for tracing the source of marine pollution mainly rely on manual field surveys and analysis of single water quality monitoring data. These methods are time-consuming and labor-intensive and it is difficult to cover vast sea areas, resulting in spatial limitations of the data. Therefore, it is impossible to trace the source of water pollution. Only the marine water quality at the monitoring site can be monitored and managed, and water pollution can be purified when it is detected. This can only solve pollution problems in local areas, making it difficult to detect and effectively control pollution sources in a timely manner. It is impossible to fundamentally eliminate pollution sources, and pollution may occur again, making it difficult to fundamentally solve the problem of marine pollution.

[0003] Therefore, how to optimize the method of tracing the source of marine water pollution so as to accurately trace the source of marine pollution and restore the marine ecological environment has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0004] The present invention provides a marine pollution source tracing big data analysis method based on spatiotemporal graph neural network to accurately trace the source of marine pollution.

[0005] To solve the above technical problems, an embodiment of the present invention provides a method for analyzing marine pollution source tracing big data based on a spatiotemporal graph neural network, comprising:

[0006] In actual marine pollution source tracing analysis, real-time collected seawater data is input into the target pollution diffusion network, and a marine pollution source tracing report for the target sea area is generated based on the output results of the target pollution diffusion network. The construction process of the target pollution diffusion network includes:

[0007] The water quality fingerprint analysis is performed on the acquired water body collection data of the target sea area to obtain the periodic water quality data of the target sea area.

[0008] An ocean current field map of the target sea area is determined, wherein the ocean current field map is obtained by analyzing first ocean dynamics data of the target sea area.

[0009] The seawater pollution diffusion characteristics are extracted based on the periodic water quality data and the ocean flow field map, and the seawater pollution diffusion characteristics are processed based on a graph neural network to obtain a first pollution diffusion network of pollutants in the target sea area.

[0010] The acquired ocean surface remote sensing data is identified and analyzed, and the first pollution diffusion network is reconstructed according to the analysis results to obtain a second pollution diffusion network.

[0011] The periodic water quality data is input into the second pollution diffusion network for bidirectional propagation analysis to obtain a forward propagation path and a reverse propagation path. According to the segmented offset between the forward propagation path and the reverse propagation path, the nodes of the second pollution diffusion network are adjusted to obtain a target pollution diffusion network.

[0012] Furthermore, the water quality fingerprint analysis is performed on the acquired water body data of the target sea area to obtain the periodic water quality data of the target sea area, including:

[0013] The collection interval is preset based on the monitoring requirements of the target sea area and the ocean current speed.

[0014] Seawater samples are collected from each monitoring point in the target sea area at preset collection intervals.

[0015] Key water quality indicators are analyzed for each of the seawater samples to obtain corresponding water quality fingerprint characteristic data.

[0016] Generate random Poisson noise according to the historical pollution statistics of the target sea area, and use the noise replacement value to replace the noise seed of the random Poisson noise. λ The noise replacement value is replaced by a value, wherein the noise replacement value is determined by typical pollution data of the target sea area.

[0017] The random Poisson noise is added to the water quality fingerprint feature data to obtain periodic water quality data of the target sea area.

[0018] Furthermore, the key water quality index analysis is performed on each of the seawater samples to obtain corresponding water quality fingerprint feature data, including:

[0019] Key water quality indexes are analyzed for each of the seawater samples to obtain a characteristic index of each key water quality index of the seawater sample.

[0020] A time domain analysis is performed on each characteristic index to obtain a change trend curve of the corresponding key water quality index in a time series, and time domain characteristic data of the seawater sample is obtained by nonlinearly fitting each change trend curve.

[0021] The spectrum data of each characteristic index is obtained by Fourier transform, the power spectrum density of the corresponding characteristic index is calculated according to each spectrum data, the frequency component data of the characteristic index is obtained, and the frequency domain characteristic data of the seawater sample is obtained according to each frequency component data.

[0022] Water quality fingerprint feature data of the seawater sample is generated based on a feature fusion algorithm according to the time domain feature data and the frequency domain feature data.

[0023] Furthermore, the extracting of seawater pollution diffusion characteristics based on the periodic water quality data and the ocean flow field map includes:

[0024] The periodic water quality data is analyzed point by point to obtain statistical characteristic data of each key water quality indicator.

[0025] A normal threshold range is set for each of the key water quality indicators based on the historical pollution statistics and the environmental standards of the target sea area.

[0026] The statistical characteristic data and the normal threshold range are compared and analyzed. If there is a statistical characteristic data that exceeds the corresponding normal threshold range, the corresponding point is marked as a potential pollution point, and the seawater pollution diffusion characteristics of the potential pollution point are extracted.

[0027] Furthermore, determining the ocean current field map of the target sea area includes:

[0028] First ocean dynamics data of the target sea area is acquired, and time alignment is performed on the first ocean dynamics data to obtain second ocean dynamics data.

[0029] Feature extraction is performed on the second ocean dynamics data to obtain ocean motion features.

[0030] The ocean motion grid data of the target sea area is constructed according to the ocean motion characteristics.

[0031] An ocean current field map of the target sea area is constructed based on the ocean motion grid data and the geographic information system data of the target sea area.

[0032] Furthermore, the ocean surface remote sensing data at least includes sea surface temperature data, chlorophyll concentration data, suspended matter concentration data and sea surface oil film distribution data.

[0033] The identification and analysis of the acquired ocean surface remote sensing data includes:

[0034] Acquire ocean surface remote sensing data of the target sea area monitored by a satellite remote sensing system, and preprocess the ocean surface remote sensing data.

[0035] Image recognition analysis is performed on the pre-processed ocean surface remote sensing data to obtain distribution characteristics of ocean surface pollutants.

[0036] Furthermore, the first pollution diffusion network is reconstructed according to the analysis result to obtain a second pollution diffusion network, including:

[0037] The surface diffusion path and surface distribution area of the pollutants in the target sea area are determined based on the distribution characteristics.

[0038] The nodes and connection paths of the first pollution diffusion network are adjusted according to the surface diffusion path and the surface distribution area.

[0039] The first pollution diffusion network is reconstructed according to the adjusted nodes and the connection paths to obtain a second pollution diffusion network.

[0040] Furthermore, the step of inputting the periodic water quality data into the second pollution diffusion network for bidirectional propagation analysis to obtain a forward propagation path and a reverse propagation path includes:

[0041] The periodic water quality data is input into the second pollution diffusion network to simulate the forward diffusion process of pollutants from the pollution source to the affected area, and a forward propagation path is obtained.

[0042] The result of the forward diffusion process is input into the second pollution diffusion network to trace the source of the pollutants and obtain a reverse propagation path.

[0043] Furthermore, the step of adjusting nodes of the second pollution diffusion network according to the segmented offset between the forward propagation path and the reverse propagation path to obtain a target pollution diffusion network includes:

[0044] The spatial offset and the temporal offset between the forward propagation path and the reverse propagation path are calculated.

[0045] The segmented spatial offset and the segmented temporal offset are obtained according to the distribution characteristics of the spatial offset and the temporal offset respectively.

[0046] The node positions of the second pollution diffusion network are adjusted according to the segmented spatial offset, and the connection path weights of the second pollution diffusion network are adjusted according to the segmented time offset to obtain a target pollution diffusion network.

[0047] Furthermore, the graph neural network-based processing of the seawater pollution diffusion characteristics to obtain a first pollution diffusion network of pollutants in the target sea area includes:

[0048] A graph neural network model is constructed, the seawater pollution diffusion characteristics are input into the graph neural network model, and the dynamic diffusion process of pollutants in the target sea area is simulated through graph convolution operations and time convolution operations.

[0049] Based on the simulation results of the graph neural network model, a first pollution diffusion network of the target sea area is generated.

[0050] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0051] Combining water quality fingerprint analysis, ocean flow field maps and graph neural networks can more accurately simulate the diffusion path of pollutants, thereby improving the accuracy of locating pollution sources; introducing random Poisson noise into the water quality fingerprint feature data, and replacing the characteristic values of the noise seed with typical pollution data, enhances the ocean data adaptability of the data noise and can better adapt to the actual marine pollution monitoring situation; using graph neural networks to process the diffusion characteristics of seawater pollution can quickly simulate the dynamic diffusion process of pollutants and improve analysis efficiency; reconstructing the pollution diffusion network through remote sensing data, and using two-way propagation analysis and segmented offset adjustment to further optimize the pollution diffusion network, reduce errors, and improve the reliability of traceability results. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flowchart of the steps of a method for analyzing big data on the source of marine pollution based on a spatiotemporal graph neural network in one embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0054] In the description of this application, the terms "first," "second," "third," etc. are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first," "second," "third," etc. may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0055] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0056] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meanings as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by those skilled in the art in specific circumstances.

[0057] The sources of marine pollution are extensive and complex, including industrial wastewater, domestic sewage, agricultural non-point source pollution, ship oil pollution, marine garbage, etc. These pollutants enter the ocean through various channels such as rivers, atmospheric deposition, and maritime transportation, making it extremely difficult to locate and trace the source of pollution. The ocean is a complex fluid environment, and the diffusion of pollutants in the ocean is affected by various ocean dynamic factors such as currents, tides, and wind fields. It is difficult to fundamentally solve the problem of seawater pollution by simply monitoring and treating the seawater in the polluted area. Therefore, this embodiment proposes a marine pollution source tracing big data analysis method based on spatiotemporal graph neural network. The specific implementation process is: in the actual marine pollution source tracing analysis, the real-time collected seawater data is input into the target pollution diffusion network, and the marine pollution source tracing report of the target sea area is generated according to the output results of the target pollution diffusion network.

[0058] The construction process of the target pollution diffusion network is shown in Figure 1 , Figure 1 The flowchart of the method for analyzing the source of marine pollution based on a spatiotemporal graph neural network in one embodiment of the present invention includes steps S11 to S15:

[0059] S11. Performing water quality fingerprint analysis on the acquired water body data of the target sea area to obtain periodic water quality data of the target sea area.

[0060] Before collecting water body data, it is necessary to determine the appropriate water body collection interval. In this embodiment, the preset collection interval is determined based on the current velocity of each monitoring point in the target sea area and the monitoring requirements of the target sea area. The faster the seawater flow rate, the shorter the collection interval of the current monitoring point should be, because faster current speeds will cause pollutants to spread to a greater distance in a short period of time. In order to capture the changes in pollutants in a timely manner, samples need to be collected more frequently to more accurately reflect the dynamic changes in water quality.

[0061] The environmental requirements for each sea area are different. The monitoring requirements are obtained based on the standards for marine pollution purification in the target sea area, so the preset collection interval is adjusted, and then seawater samples are collected from each monitoring point in the target sea area at the preset collection interval.

[0062] Water quality fingerprint data refers to a unique data set that can reflect the characteristics of a water body obtained by analyzing a series of key water quality indicators on water samples. The water quality fingerprint is similar to the uniqueness of a human fingerprint and can provide an identifier for the water quality status of a specific water body. The acquisition of water quality fingerprint data requires monitoring of multiple key water quality indicators of the sample, including but not limited to chemical indicators such as chemical oxygen demand, heavy metal content, dissolved oxygen, ammonia nitrogen content and petroleum pollutant concentration, as well as physical indicators such as water temperature and turbidity. It also includes some biological indicators such as plankton species, algae content and bacterial content.

[0063] In addition, since the monitoring object of this embodiment is marine polluted water sources, a considerable portion of the source of marine pollution comes from pollution caused by the production of surrounding chemical plants. Therefore, the key water quality indicators of this embodiment can also include heavy metal content (such as mercury, lead, cadmium, etc.) and organic pollutants (such as polycyclic aromatic hydrocarbons, pesticide residues, etc.). Specifically, the types of pollutants in seawater are shown in Table 1. The detection results of these indicators together constitute the basic water quality fingerprint feature data of the target sea area.

[0064] Table 1 Marine pollutant information

[0065]

[0066] The change of water quality indicators is a dynamic process, which is affected by many factors, such as the emission patterns of pollution sources, ocean dynamics (such as currents and tides), etc. In order to better understand the change patterns of water quality indicators, it is necessary to conduct a time domain analysis of the characteristic index. The specific analysis process is as follows:

[0067] The characteristic indices of seawater samples collected at the same monitoring point at different times are arranged in chronological order to form time series data.

[0068] By drawing the trend curve of time series data, we can intuitively observe the changes of key water quality indicators over time.

[0069] Since changes in water quality indicators are usually nonlinear, a nonlinear fitting method is needed to more accurately describe the trend of change. Nonlinear fitting can use polynomial fitting, exponential fitting, or other suitable mathematical models, which will not be described in detail here.

[0070] Through nonlinear fitting, a smoother and more accurate change trend curve can be obtained, thereby extracting time domain characteristic data that can reflect the change law of water quality indicators in the time dimension.

[0071] In addition to time domain analysis, frequency domain analysis is also an important method for studying changes in water quality indicators. Frequency domain analysis can reflect the periodicity and frequency characteristics of changes in water quality indicators, and help identify pollution sources and understand the diffusion patterns of pollutants. The specific frequency domain analysis process is as follows:

[0072] By performing Fourier transform on the characteristic index of each key water quality indicator, the time series data can be decomposed into a combination of sine waves and cosine waves of different frequencies, thereby obtaining spectrum data.

[0073] The power spectrum density of each characteristic index is calculated based on the spectrum data, which reflects the energy distribution of different frequency components in the signal. By calculating the power spectrum density, the main frequency component data of the changes in water quality indicators can be identified. The frequency domain characteristic data of the seawater sample are obtained based on the frequency component data corresponding to each characteristic index, which is used to provide additional information about the periodicity and regularity of water quality changes.

[0074] The time domain feature data and frequency domain feature data are input into the feature fusion algorithm. Through the processing of the algorithm, these data are fused into a comprehensive feature vector. The feature fusion algorithm can be linear discriminant analysis or other machine learning algorithms. This embodiment uses principal component analysis to fuse the two to obtain water quality fingerprint feature data.

[0075] Because seawater is constantly changing, random fluctuations in water quality data need to be accounted for when performing water quality fingerprint analysis to more realistically simulate the dynamic changes in water quality in the target sea area. This embodiment simulates this random fluctuation by generating random Poisson noise and adding it to the water quality fingerprint feature data, thereby obtaining data that is closer to actual conditions.

[0076] When monitoring water quality in a target sea area, historical pollution statistics are an important reference. These statistics record the pollution situation in the target sea area over a period of time, including information such as the type, concentration, and frequency of occurrence of pollutants. Analyzing this historical data reveals the presence of certain random fluctuations in water quality data. Poisson noise is a typical form of random noise, and its probability distribution conforms to the Poisson distribution, making it a good model for these random fluctuations. Therefore, this embodiment uses historical pollution statistics from the target sea area to generate random Poisson noise to simulate random variations in water quality data.

[0077] The generation of Poisson noise depends on the λ value of the noise seed, which determines the intensity and distribution characteristics of the noise. However, the pollution characteristics of different sea areas are different, and the Poisson noise generated directly using the default λ value may not accurately reflect the actual water quality fluctuations in the target sea area. In order to make the generated Poisson noise more consistent with the pollution characteristics of the target sea area, this embodiment introduces a noise replacement value. The noise replacement value is determined based on the typical pollution data of the target sea area, which reflects the main characteristics and common patterns of pollution in the target sea area. By replacing the λ value of the noise seed with the noise replacement value, the distribution characteristics of the Poisson noise can be adjusted to make it closer to the random fluctuations in the actual water quality data of the target sea area.

[0078] Specifically, the typical pollution data used in this embodiment is obtained based on historical pollution statistics. The historical pollution statistics record the pollution situation of the target sea area over a period of time in the past, including information such as the type, concentration, frequency of occurrence, time and spatial distribution of pollution events, etc. By analyzing these historical data, representative pollution events and pollutant characteristics can be screened out, thereby determining typical pollution data.

[0079] For example, if high concentrations of heavy metal mercury have been found in a certain sea area in multiple past tests, and its source is mainly concentrated in a specific area, then the concentration of mercury and its distribution characteristics can be regarded as one of the typical pollution data of the sea area.

[0080] Typical pollution data are also related to the main pollution sources of the target sea area. If the main pollution source of a sea area is chemical plant emissions, then the typical pollution data may include the concentration and distribution of specific organic pollutants (such as polycyclic aromatic hydrocarbons) or heavy metals (such as mercury, lead, cadmium, etc.).

[0081] By adding the random Poisson noise to the water quality fingerprint characteristic data, the periodic water quality data of the target sea area obtained not only includes basic information such as the water quality fingerprint characteristic data, but also takes into account the random fluctuations of the water body, which can more realistically reflect the changes in the water quality of the target sea area in different time periods.

[0082] S12. Determine an ocean current field map of the target sea area, wherein the ocean current field map is obtained by analyzing first ocean dynamics data of the target sea area.

[0083] The first ocean dynamics data of the target sea area is obtained. In order to ensure that the ocean dynamics data can be effectively integrated with the water quality data and other relevant data, it is necessary to time-align the first ocean dynamics data, adjust the data from different sources and different time resolutions to a unified time base, and obtain the second ocean dynamics data.

[0084] Feature extraction is performed on the second ocean dynamics data to obtain ocean motion features, and then the sea area is divided into multiple grid units according to the geographical environment characteristics of the target sea area. The size of the grid can be adjusted according to the resources and needs in the specific actual operation to obtain the ocean motion grid data of the target sea area.

[0085] Geographic Information Systems (GIS) can obtain geographic information of the target sea area, including information on the coastline, sea surface conditions, and other factors that affect the distribution and direction of ocean currents. Ocean motion grid data and GIS data can be combined with visualization software to construct ocean current field maps.

[0086] S13. Extract seawater pollution diffusion characteristics based on the periodic water quality data and the ocean flow field map, process the seawater pollution diffusion characteristics based on a graph neural network, and obtain a first pollution diffusion network of pollutants in the target sea area.

[0087] Since the water quality data at each monitoring point may include multiple key water quality indicators, the statistical characteristic data of each indicator at that point can be extracted through point-by-point analysis.

[0088] To determine whether water quality is abnormal, it's necessary to establish reasonable normal thresholds based on historical pollution statistics and the environmental standards for the target sea area. Historical pollution statistics document pollution conditions in the target sea area over a period of time, including information such as pollutant types, concentrations, and frequency of occurrence. By analyzing this data, it's possible to determine the normal fluctuation range for each key water quality indicator. Environmental standards, on the other hand, are a crucial basis for determining whether water quality meets these standards. These standards, typically set by environmental protection authorities, reflect minimum requirements for protecting the marine ecosystem and human health.

[0089] The statistical characteristic data of each monitoring point is compared with the set normal threshold range. If the statistical characteristic data of a key water quality indicator at a monitoring point exceeds the normal threshold range, the point is considered abnormal and marked as a potential pollution point. Potential pollution points may be direct discharge points of pollution sources or key nodes on the pollution diffusion path.

[0090] The seawater pollution diffusion characteristics of potential pollution points are extracted. By analyzing these characteristics, we can gain a preliminary understanding of the diffusion patterns of pollutants.

[0091] The graph neural network model can learn the spatiotemporal dependencies in the data. In the simulation of marine pollution diffusion, the nodes of the model can represent monitoring points, and the edges can represent the diffusion relationship between monitoring points. This embodiment constructs a graph neural network model and inputs the seawater pollution diffusion characteristics into the graph neural network model, including the concentration, diffusion direction, and diffusion speed of pollutants.

[0092] Through graph convolution operations and time convolution operations, a graph neural network model is used to simulate the dynamic diffusion process of pollutants in the target sea area, including how pollutants spread from one monitoring point to another and how they change over time.

[0093] The simulation results of the graph neural network model include information such as the diffusion path, diffusion range and diffusion speed of pollutants in the target sea area. Based on the above simulation results, the first pollution diffusion network of the target sea area is generated, providing an initial diffusion analysis model for marine pollution source tracing analysis.

[0094] S14. Identify and analyze the acquired ocean surface remote sensing data, and reconstruct the first pollution diffusion network based on the analysis results to obtain a second pollution diffusion network.

[0095] In the analysis of marine pollution sources, ocean surface data is one of the important data sources, mainly because the surface of the ocean is the location that most intuitively and clearly shows the status of seawater pollution. Therefore, it is necessary to reconstruct the first pollution diffusion network of the target sea area based on the ocean surface data.

[0096] Satellite remote sensing systems can provide large-scale, high-resolution ocean surface information. Therefore, this embodiment uses ocean surface remote sensing data obtained by the satellite remote sensing system to analyze the pollution status of the ocean surface. Specifically, the ocean surface remote sensing data includes at least sea surface temperature data, chlorophyll concentration data, suspended matter concentration data, and sea surface oil film distribution data.

[0097] The acquired ocean surface remote sensing data is preprocessed, and then image recognition analysis is used to obtain the distribution characteristics of ocean surface pollutants. The surface diffusion path and surface distribution area of pollutants in the target sea area are determined based on the distribution characteristics. The nodes and edges of the first pollution diffusion network are adjusted based on the surface diffusion path and surface distribution area. The nodes represent pollution sources or severely affected areas, and the edges are connection paths of the first pollution diffusion network, representing the propagation paths of pollutants. The first pollution diffusion network is reconstructed based on the adjusted nodes and connection paths to obtain a second pollution diffusion network combined with satellite remote sensing data, thereby further improving the accuracy and efficiency of ocean pollution tracing.

[0098] S15. Input the periodic water quality data into the second pollution diffusion network for bidirectional propagation analysis to obtain a forward propagation path and a reverse propagation path. According to the segmented offset between the forward propagation path and the reverse propagation path, adjust the nodes of the second pollution diffusion network to obtain a target pollution diffusion network.

[0099] Once the second pollution diffusion network is constructed, it can be used to trace the propagation paths and sources of pollutants. However, the actual marine environment is complex and changeable, and a network generated by a one-way simulation of collected data may not fully and accurately reflect the actual spread of pollutants. Therefore, to further improve the accuracy and reliability of source tracing analysis, more detailed optimization and adjustment of the model are required.

[0100] Therefore, this embodiment adopts a two-way propagation analysis method. First, the periodic water quality data is input into the second pollution diffusion network in chronological order and at certain time intervals to simulate the forward diffusion process of pollutants from the pollution source to the affected area and obtain the forward propagation path.

[0101] Next, the results of the forward diffusion process are fed into a second diffusion network to trace the source of the pollutants and obtain a reverse propagation path. The reverse propagation path is obtained by reversely analyzing the results of the forward diffusion process to determine the source of the pollutants.

[0102] Through bidirectional propagation analysis, the consistency of forward and reverse propagation can be verified based on the diffusion characteristics of pollutants in the ocean. If the forward and reverse propagation paths can match well, it means that the simulation results of the second diffusion network are relatively accurate. If there is a deviation between the two, it is necessary to analyze the reasons for these deviations and adjust the model. Specifically, the adjustment process is as follows:

[0103] The spatial offset reflects the difference in space between the forward and reverse propagation paths, while the temporal offset reflects the difference in time between the two. Based on the specific data points on the path, the path offsets of the two are obtained by analyzing the position and timeline information of the corresponding points on the forward and reverse propagation paths, and the spatial offset and temporal offset between the forward and reverse propagation paths are calculated.

[0104] By observing the variation characteristics of the offset within different sections or time periods, the network is divided into different segments and the segmented spatial offsets of the spatial offsets in each segment are calculated. If the variation range of the segmented spatial offsets between adjacent segments remains within a constant range, it indicates that there is no spatial offset between the forward and reverse propagation paths, but rather a temporal offset. In this case, the structure of the second pollution diffusion network does not need to be adjusted. Otherwise, the positions of the nodes and connection paths of the second pollution diffusion network need to be adjusted based on the segmented spatial offsets.

[0105] After processing the second pollution diffusion network according to the spatial offset, its segmented time offset in different segments is obtained according to the time offset. The weights of each connection path in the second pollution diffusion network are adjusted according to the segmented time offset to indirectly adjust the diffusion speed of pollutants simulated by the second pollution diffusion network to obtain the target pollution diffusion network.

[0106] The marine pollution source tracing big data analysis method based on spatiotemporal graph neural network of the present invention combines water quality fingerprint analysis, ocean flow field map and graph neural network, which can more accurately simulate the diffusion path of pollutants, thereby improving the accuracy of pollution source positioning; introducing random Poisson noise into the water quality fingerprint feature data, and replacing the characteristic value of the noise seed with typical pollution data, thereby enhancing the marine data adaptability of data noise and being able to better adapt to the actual marine pollution monitoring situation; using graph neural network to process the seawater pollution diffusion characteristics, it can quickly simulate the dynamic diffusion process of pollutants and improve analysis efficiency; reconstructing the pollution diffusion network through remote sensing data, and using two-way propagation analysis and segmented offset adjustment to further optimize the pollution diffusion network, reduce errors, and improve the reliability of tracing results.

[0107] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network, characterized by: include: In actual marine pollution source tracing analysis, real-time collected seawater data is input into the target pollution diffusion network, and a marine pollution source tracing report for the target sea area is generated based on the output results of the target pollution diffusion network. The construction process of the target pollution diffusion network includes: Performing water quality fingerprint analysis on the acquired water body data of the target sea area to obtain periodic water quality data of the target sea area; Determining an ocean current field map of the target sea area, wherein the ocean current field map is obtained by analyzing first ocean dynamics data of the target sea area; Extracting seawater pollution diffusion characteristics based on the periodic water quality data and the ocean flow field map, processing the seawater pollution diffusion characteristics based on a graph neural network to obtain a first pollution diffusion network of pollutants in the target sea area; Identify and analyze the acquired ocean surface remote sensing data, and reconstruct the first pollution diffusion network based on the analysis results to obtain a second pollution diffusion network; The periodic water quality data is input into the second pollution diffusion network for bidirectional propagation analysis to obtain a forward propagation path and a reverse propagation path. According to the segmented offset between the forward propagation path and the reverse propagation path, the nodes of the second pollution diffusion network are adjusted to obtain a target pollution diffusion network.

2. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 1, characterized in that: The water quality fingerprint analysis is performed on the acquired water body data of the target sea area to obtain the periodic water quality data of the target sea area, including: Presetting the collection interval based on the monitoring requirements of the target sea area and the ocean current speed; Collecting seawater samples at each monitoring point in the target sea area at a preset collection interval; Analyzing key water quality indicators of each of the seawater samples to obtain corresponding water quality fingerprint characteristic data; Generate random Poisson noise according to the historical pollution statistics of the target sea area, and use the noise replacement value to replace the noise seed of the random Poisson noise. λ The noise replacement value is replaced by a value, wherein the noise replacement value is determined by typical pollution data of the target sea area; The random Poisson noise is added to the water quality fingerprint feature data to obtain periodic water quality data of the target sea area.

3. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 2, characterized in that: The key water quality index analysis of each seawater sample is performed to obtain corresponding water quality fingerprint characteristic data, including: performing key water quality index analysis on each of the seawater samples to obtain a characteristic index of each key water quality index of the seawater sample; Performing time domain analysis on each characteristic index to obtain a change trend curve of the corresponding key water quality index in a time series, and obtaining time domain characteristic data of the seawater sample by nonlinearly fitting each change trend curve; Obtaining frequency spectrum data of each characteristic index through Fourier transform, calculating the power spectrum density of the corresponding characteristic index based on each frequency spectrum data to obtain frequency component data of the characteristic index, and obtaining frequency domain characteristic data of the seawater sample based on each frequency component data; Water quality fingerprint feature data of the seawater sample is generated based on a feature fusion algorithm according to the time domain feature data and the frequency domain feature data.

4. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 3 is characterized in that: The extracting of seawater pollution diffusion characteristics based on the periodic water quality data and the ocean flow field map includes: Performing point-by-point analysis on the periodic water quality data to obtain statistical characteristic data of each of the key water quality indicators; Setting a normal threshold range for each of the key water quality indicators based on the historical pollution statistics and the environmental standards of the target sea area; The statistical characteristic data and the normal threshold range are compared and analyzed. If there is a statistical characteristic data that exceeds the corresponding normal threshold range, the corresponding point is marked as a potential pollution point, and the seawater pollution diffusion characteristics of the potential pollution point are extracted.

5. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 1, characterized in that: Determining the ocean current field map of the target sea area includes: Acquiring first ocean dynamics data of the target sea area, and performing time alignment on the first ocean dynamics data to obtain second ocean dynamics data; performing feature extraction on the second ocean dynamics data to obtain ocean motion features; Constructing ocean motion grid data of the target sea area according to the ocean motion characteristics; An ocean current field map of the target sea area is constructed based on the ocean motion grid data and the geographic information system data of the target sea area.

6. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 1, characterized in that: The ocean surface remote sensing data at least includes sea surface temperature data, chlorophyll concentration data, suspended matter concentration data and sea surface oil film distribution data; The identification and analysis of the acquired ocean surface remote sensing data includes: Acquiring ocean surface remote sensing data of the target sea area monitored by a satellite remote sensing system, and preprocessing the ocean surface remote sensing data; Image recognition analysis is performed on the pre-processed ocean surface remote sensing data to obtain distribution characteristics of ocean surface pollutants.

7. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 6, characterized in that: The reconstructing the first pollution diffusion network according to the analysis result to obtain a second pollution diffusion network includes: determining the surface diffusion path and surface distribution area of the pollutants in the target sea area based on the distribution characteristics; adjusting nodes and connection paths of the first pollution diffusion network according to the surface diffusion path and the surface distribution area; The first pollution diffusion network is reconstructed according to the adjusted nodes and the connection paths to obtain a second pollution diffusion network.

8. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 1, characterized in that: The step of inputting the periodic water quality data into the second pollution diffusion network for bidirectional propagation analysis to obtain a forward propagation path and a reverse propagation path includes: Inputting the periodic water quality data into the second pollution diffusion network to simulate the forward diffusion process of pollutants from the pollution source to the affected area, and obtaining a forward propagation path; The result of the forward diffusion process is input into the second pollution diffusion network to trace the source of the pollutants and obtain a reverse propagation path.

9. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 8, characterized in that: The step of adjusting nodes of the second pollution diffusion network according to the segmented offset between the forward propagation path and the reverse propagation path to obtain a target pollution diffusion network includes: Calculating a spatial offset and a temporal offset between the forward propagation path and the reverse propagation path; Obtaining a segmented spatial offset and a segmented temporal offset according to distribution characteristics of the spatial offset and the temporal offset respectively; The node positions of the second pollution diffusion network are adjusted according to the segmented spatial offset, and the connection path weights of the second pollution diffusion network are adjusted according to the segmented time offset to obtain a target pollution diffusion network.

10. The method for analyzing marine pollution source tracing big data based on spatiotemporal graph neural network according to claim 1, characterized in that: The processing of the seawater pollution diffusion characteristics based on the graph neural network to obtain a first pollution diffusion network of pollutants in the target sea area includes: Constructing a graph neural network model, inputting the seawater pollution diffusion characteristics into the graph neural network model, and simulating the dynamic diffusion process of pollutants in the target sea area through graph convolution operations and time convolution operations; Based on the simulation results of the graph neural network model, a first pollution diffusion network of the target sea area is generated.

Images