Marine environment monitoring method based on low earth orbit satellite

Abstract

The invention discloses a marine environment monitoring method based on a low earth orbit satellite, and belongs to the field of marine environment monitoring, and the method comprises the steps: carrying out the preprocessing of multi-source heterogeneous data, and obtaining a unified representation vector; performing clustering analysis on the unified representation vector to obtain a current marine environment mode; comparing the deviation between the current marine environment mode and the normal state mode, and determining an abnormal state; classifying the abnormal states to obtain abnormal types, and classifying the abnormal types to predefined abnormal types to obtain classified abnormal states; performing abnormal development trend prediction on the classified abnormal states, and evaluating an abnormal diffusion risk based on the predicted abnormal development trend; generating a marine environment anomaly event report of the current sea area based on the anomaly type, the anomaly development trend and the anomaly diffusion risk; and performing dynamic monitoring on marine environment abnormal event reports of a plurality of sea areas based on a distributed computing framework to obtain multi-level marine environment analysis from short-term fluctuation of local sea areas to long-term trend of global scale.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine environment monitoring, and particularly relates to a marine environment monitoring method based on low-earth orbit satellites. Background Art

[0002] Marine environment monitoring is an important field of global ecological protection and resource management, which is directly related to climate change, biodiversity, and the sustainable development of the marine economy. With the increasing impact of human activities on the marine ecosystem, it has become an indispensable requirement to accurately and timely master marine environmental parameters. Due to their wide coverage and high data acquisition efficiency, low-earth orbit satellites have shown irreplaceable value in this field. However, current marine environment monitoring methods still face significant limitations. Traditional monitoring means mostly rely on single data sources, such as buoys or shipborne equipment, with narrow data coverage and insufficient timeliness. Existing satellite monitoring systems often have low accuracy when dealing with multi-source data fusion and complex environmental feature recognition, and it is difficult to meet diverse monitoring needs.

[0003] These limitations point to deeper technical challenges. First, the effective integration of multi-dimensional data is a core bottleneck. The parameter data types such as ocean temperature, salinity, and chlorophyll concentration transmitted back by low-earth orbit satellites are numerous and heterogeneous with multi-source remote sensing information, making it difficult to achieve efficient fusion. Second, it is difficult to automatically identify marine anomalies. Due to the large differences in sea area characteristics, traditional algorithms are difficult to adaptively adjust, resulting in frequent misjudgments or missed judgments. Finally, the dynamic analysis ability of spatio-temporal data is insufficient. Existing systems lack an efficient computing framework when dealing with continuous time series and wide-area spatial information, which limits the real-time and comprehensive nature of monitoring. These unresolved technical factors make it difficult for the monitoring system to achieve high accuracy and intelligence in complex marine environments.

[0004] Therefore, how to construct a multi-dimensional data analysis system that can efficiently process multi-source heterogeneous data, automatically identify marine anomalies, and adapt to different sea area characteristics has become a key issue in enhancing the marine environment monitoring ability of low-earth orbit satellites. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a marine environment monitoring method based on low-earth orbit satellites to solve the problems existing in the above prior art.

[0006] To achieve the above object, the present invention provides a marine environment monitoring method based on low-earth orbit satellites, including:

[0007] Preprocessing multi-source heterogeneous data to obtain a unified representation vector;

[0008] Using a clustering algorithm to perform clustering analysis on the unified representation vector to obtain the current marine environment pattern;

[0009] An anomaly detection algorithm is used to compare the deviation between the current marine environmental pattern and the normal state pattern to determine the abnormal state;

[0010] A deep learning model is used to classify the abnormal state to obtain the abnormal type, and the abnormal type is classified into the predefined abnormal types to obtain the classified abnormal state;

[0011] A time series analysis algorithm is used to predict the abnormal development trend of the classified abnormal state, and the abnormal diffusion risk is evaluated based on the predicted abnormal development trend;

[0012] An ocean environmental anomaly event report for the current sea area is generated based on the abnormal type, abnormal development trend, and abnormal diffusion risk;

[0013] A distributed computing framework is constructed, and based on the distributed computing framework, dynamic monitoring is performed on the ocean environmental anomaly event reports of several sea areas to obtain a multi-level ocean environmental analysis from short-term fluctuations in local sea areas to long-term trends at the global scale.

[0014] Optionally, the process of preprocessing multi-source heterogeneous data to obtain a unified representation vector includes:

[0015] The multi-source heterogeneous data is standardized to obtain standardized data;

[0016] The principal component analysis algorithm is used to extract key features from the standardized data to obtain a feature set;

[0017] The preset weights are used to perform weighted fusion on the feature set to obtain an initial representation vector;

[0018] The distribution law of the initial representation data is judged to obtain the environmental state representation;

[0019] A unified representation vector of the multi-source data is generated based on the environmental state representation.

[0020] Optionally, the process of using a clustering algorithm to perform clustering analysis on the unified representation vector to obtain the current marine environmental pattern includes:

[0021] Based on the unified representation vector, a multi-dimensional feature space is constructed, and the clustering algorithm is used to group the data points in the multi-dimensional feature space to obtain a preliminary grouping result;

[0022] Based on the preliminary grouping result, the inter-class distance is calculated to obtain the distance calculation result;

[0023] Based on the distance calculation result, the intra-class compactness is analyzed to obtain the compactness analysis data;

[0024] Determine the optimal number of clusters based on the compactness analysis data and the distance calculation results to obtain a cluster number representation;

[0025] Adjust the preliminary classification results using the cluster number representation to obtain an optimized grouping set;

[0026] Determine the ocean environment pattern based on the optimized grouping set to obtain an environment pattern representation;

[0027] Generate the distribution law of the multi-dimensional feature space through the environment pattern representation to obtain the current ocean environment pattern.

[0028] Optionally, the process of determining the abnormal state by using an anomaly detection algorithm to compare the deviation between the current ocean environment pattern and the normal state pattern includes:

[0029] Construct a pattern library of the normal state through historical data to obtain a pattern library representation;

[0030] Compare the pattern library representation with the current ocean environment pattern to obtain deviation degree data;

[0031] Process the deviation degree data using an anomaly detection algorithm to obtain an anomaly detection result;

[0032] If the anomaly detection result exceeds the preset threshold, it is determined as an abnormal state.

[0033] Optionally, the process of obtaining the classified abnormal state includes:

[0034] Obtain an abnormal feature set through historical data, and train the abnormal feature set using a deep learning model to obtain a type representation;

[0035] Extract feature data from the current abnormal state, and apply a pattern matching algorithm to the feature data to determine the matching degree with the type representation;

[0036] Classify the abnormal state into predefined abnormal types based on the matching degree to obtain the classified abnormal state;

[0037] Adjust the boundary conditions of the type representation for the classified abnormal state to obtain an updated type definition;

[0038] Use the updated type definition to re-compare the abnormal features in the historical data to obtain an optimized feature extraction rule.

[0039] Optionally, the process of evaluating the abnormal diffusion risk includes:

[0040] Obtain time series data through the classified abnormal state;

[0041] Process the time series data using a time series analysis algorithm to obtain a change sequence of the abnormal state;

[0042] Extract the dynamic features of the environmental parameters based on the change sequence of the abnormal state to determine the development trend of the abnormal state;

[0043] Obtain the diffusion features from the environmental parameters in the spatially adjacent regions and judge the distribution range of the diffusion features;

[0044] Based on the development trend of the abnormal state and the diffusion features, obtain the level of the abnormal diffusion risk.

[0045] Optionally, the process of generating a marine environmental anomaly event report for the current sea area based on the abnormal type, abnormal development trend, and abnormal diffusion risk includes:

[0046] Through the combination of the abnormal type and the abnormal diffusion risk, obtain the countermeasures in the preset policy library and determine the solution with the highest matching degree;

[0047] Extract the dynamic features from the abnormal development trend, judge the relevance between the abnormal diffusion risk and the abnormal development trend, and obtain the preliminary distribution of the risk level;

[0048] According to the preliminary distribution of the risk level and the abnormal type, adjust the priority of the countermeasures to obtain an optimized policy solution.

[0049] Optionally, the process of performing a multi-level marine environmental analysis from short-term fluctuations in local sea areas to long-term trends at the global scale by dynamically monitoring marine environmental anomaly event reports based on the distributed computing framework includes:

[0050] By collecting environmental data from different sea areas, obtain the distribution set of sea area characteristics and environmental differences;

[0051] Obtain the distribution set of sea area characteristics and environmental differences through the monitoring requirements and determine the data characteristics;

[0052] According to the data characteristics, use a clustering algorithm to adjust the aggregation granularity to obtain an optimized data aggregation result;

[0053] Based on the optimized data aggregation result, obtain the characteristic distribution of short-term fluctuations and long-term trends and judge the analysis level;

[0054] If the short-term fluctuations exceed the short-wave threshold, then adjust the analysis algorithm through feedback to obtain the change characteristics of the local sea area;

[0055] If the long-term trend deviates from the long-wave range, then extract the global-scale change pattern through a convolutional neural network to determine the adjusted trend distribution;

[0056] Adopt a multi-level analysis method according to the change characteristics of the local sea area and the adjusted trend distribution to obtain a dynamic monitoring summary from the local to the global.

[0057] Compared with the prior art, the present invention has the following advantages and technical effects:

[0058] The present invention discloses an intelligent marine environment anomaly monitoring method based on low-earth orbit satellites. This method constructs a multi-dimensional marine environment feature space, performs standardized processing and feature fusion on multi-source heterogeneous data such as temperature, salinity, and chlorophyll concentration transmitted back by satellites. The present invention uses clustering analysis to determine the current marine environment mode, compares it with the historical normal state mode library, and realizes anomaly detection. For the detected abnormal state, the present invention uses a deep learning model for classification and combines time series analysis to predict the development trend of the anomaly. According to the anomaly type, trend, and diffusion risk, the present invention automatically generates coping strategies and anomaly event reports. Through a distributed computing framework and an adaptive parameter adjustment mechanism, the present invention realizes real-time dynamic monitoring of a large area of sea areas and can optimize the anomaly detection effect according to the characteristics of different sea areas. This method significantly improves the accuracy and efficiency of marine environment monitoring and provides strong support for marine ecological protection and resource management. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0060] Figure 1 It is a flowchart of the marine environment monitoring method based on low-earth orbit satellites according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0062] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0063] Embodiment 1

[0064] As Figure 1 shown, in this embodiment, a marine environment monitoring method based on low-earth orbit satellites is provided, including the following steps:

[0065] S101. Use a data preprocessing module to standardize multi-source heterogeneous data such as ocean temperature, salinity, and chlorophyll concentration transmitted by low-earth orbit satellites, extract key features of various types of data through a feature extraction algorithm, and perform feature fusion according to a preset weight to obtain a unified representation vector.

[0066] Further, the process of preprocessing multi-source heterogeneous data to obtain a unified representation vector includes: standardizing the multi-source heterogeneous data to obtain standardized data; using a principal component analysis algorithm to extract key features from the standardized data to obtain a feature set; using a preset weight to perform weighted fusion on the feature set to obtain an initial representation vector; judging the distribution law of the initial representation data to obtain an environmental state representation; and generating a unified representation vector of the multi-source data based on the environmental state representation.

[0067] Even further, as a specific implementation manner of this embodiment, using a data preprocessing module to perform standardization processing on multi-source heterogeneous data such as ocean temperature, salinity data, and chlorophyll concentration is to eliminate the differences in different data dimensions and ranges.

[0068] The ocean temperature is in degrees Celsius and ranges from -2 to 30, the salinity is expressed in parts per thousand and ranges from 30 to 40, and the chlorophyll concentration is in micrograms per liter and ranges from 0.1 to 10. Standardization processing can map these data to a distribution with a mean of 0 and a variance of 1, which is convenient for subsequent analysis.

[0069] Extract key features from the standardized data using a principal component analysis algorithm. The purpose is to reduce redundant information and focus on the main components that explain the data variability. The standardized dataset of ocean temperature, salinity, and chlorophyll concentration contains 1000 samples, and each sample has 3 variables. Principal component analysis will find that the first principal component explains 70% of the variation and may be highly correlated with temperature and salinity, and the second principal component explains 20% and is related to chlorophyll concentration. In this way, a feature set composed of principal components can be obtained. The advantage of this method is that it can extract key information from multi-dimensional data and reduce the computational complexity. Perform weighted fusion on the feature set according to a preset weight to obtain an initial representation vector, which reflects the importance differences of different features for the final result.

[0070] In a possible implementation manner, assume that the weight of the first principal component is set to 0.6 and the weight of the second principal component is 0.4. After weighting, a comprehensive vector is obtained.

[0071] The value of the first principal component of a certain sample is 2, and the value of the second principal component is 1. Then the value of the initial representation vector is 2×0.6 + 1×0.4 = 1.6. This weighted fusion can highlight the features that have a greater impact on the ocean environment and improve the pertinence of the representation. If the dimension of the initial representation vector exceeds a preset threshold, the dimension is adjusted through dimensionality reduction processing to obtain an optimized representation vector.

[0072] As a specific implementation of this embodiment, the initial vector dimension is 10, and the threshold is 5, which is adjusted by retaining the first 5 principal components or using other dimensionality reduction methods such as t-SNE. This dimensionality reduction not only reduces complexity but also retains the main information, improving the efficiency of subsequent processing. For the optimized feature vector, a clustering algorithm is used to judge the distribution law between data and obtain the classification result.

[0073] Using K-means clustering, the data is divided into three categories: high temperature and high salinity, low temperature and low salinity, and intermediate state. Suppose the optimized feature vector value of a certain sample is 1.5, and it is classified into the high temperature and high salinity category after clustering. This classification can reveal the potential patterns of ocean data and provide a basis for environmental state analysis. By matching the classification result with a preset ocean environment model, the environmental state representation is obtained.

[0074] In one embodiment, the preset model includes three states: "eutrophication", "normal", and "oligotrophication". If the high temperature and high salinity category matches eutrophication, then the area is represented as the eutrophication state. This matching can intuitively reflect the characteristics of the ocean environment and facilitate ecological monitoring. According to the environmental state representation, a unified feature vector of multi-source data is generated.

[0075] Preferably, the state label can be combined with the optimized feature vector, such as the vector [1.5, eutrophication]. This unified representation integrates multi-source information and is helpful for downstream tasks such as prediction or visualization.

[0076] For example, when monitoring the outbreak of seaweeds, this representation can provide comprehensive data support and improve the accuracy of early warning.

[0077] S102. Construct a multi-dimensional ocean environmental feature space according to the unified feature vector, use a clustering algorithm to perform clustering analysis on the data points in the feature space, determine the optimal number of clusters by calculating the distance between classes and the compactness within classes, and obtain the current ocean environmental pattern.

[0078] Furthermore, the process of using a clustering algorithm to perform clustering analysis on the unified feature vector to obtain the current ocean environmental pattern includes: constructing a multi-dimensional feature space based on the unified feature vector, using a clustering algorithm to group the data points in the multi-dimensional feature space to obtain a preliminary grouping result; calculating the distance between classes based on the preliminary grouping result to obtain a distance calculation result; analyzing the compactness within classes based on the distance calculation result to obtain compactness analysis data; determining the optimal number of clusters based on the compactness analysis data and the distance calculation result to obtain a cluster number representation; using the cluster number representation to adjust the preliminary classification result to obtain an optimized grouping set; determining the ocean environmental pattern according to the optimized grouping set to obtain an environmental pattern representation; and generating the distribution law of the multi-dimensional feature space through the environmental pattern representation to obtain the current ocean environmental pattern.

[0079] Furthermore, as a specific implementation of this embodiment, when constructing a multi-dimensional feature space through unified representation vectors, it can be understood that multi-source data such as ocean temperature, salinity, and chlorophyll concentration are mapped into a coordinate system of a higher dimension.

[0080] There are three types of data in this embodiment. The temperature ranges from 15 to 25 degrees Celsius, the salinity is between 30 and 35, and the chlorophyll concentration is between 0.1 and 0.5 mg / m³. After standardization, they are mapped into a three-dimensional space, and each data point corresponds to a coordinate.

[0081] In a possible implementation manner, when using a clustering algorithm to group data points, the K-means algorithm can be selected.

[0082] In this embodiment, the initial number of clusters is set to 3, which respectively represent the high-temperature and high-salinity area, the low-temperature and low-salinity area, and the chlorophyll enrichment area. After running the algorithm, the data points are assigned to different groups. The preliminary grouping result shows that the high-temperature and high-salinity area is concentrated in a corner of the space.

[0083] It should be noted that the initial grouping may not be precise enough, so further optimization is required.

[0084] In this embodiment, when calculating the inter-class distance according to the preliminary grouping result, the Euclidean distance is used for measurement.

[0085] Specifically, the center point coordinates of the high-temperature and high-salinity area are (20, 34, 0.2), and the center of the low-temperature and low-salinity area is (16, 31, 0.3). Calculate the distance between the two points to reflect the separation degree between the groups.

[0086] Preferably, the average distance from the data points in each group to the center can be statistically calculated.

[0087] In one embodiment, the average distance from the data points in the high-temperature and high-salinity area to the center is 0.8, while that in the low-temperature and low-salinity area is 1.2. If the preset threshold is 1, the high-temperature and high-salinity area has a better compactness. This step helps to judge whether the grouping is reasonable and avoid overly loose classification. If the compactness analysis data meets the preset threshold, the optimal number of clusters is determined through the distance calculation result and the compactness analysis data.

[0088] In this embodiment, it is initially set to 3, but it is found through analysis that the inter-class distance is small and the compactness is poor. Adjusting it to 2 is more appropriate to obtain the cluster number representation. This adjustment can better reflect the true distribution law of the ocean environment. When using the cluster number representation to adjust the preliminary grouping result, it can be understood as re-running the clustering algorithm.

[0089] After adjusting the number of groups from 3 to 2 in this embodiment, the data points are re-divided into the high-temperature area and the low-temperature area. The optimized grouping set is more in line with the actual environmental characteristics. This optimization can improve the reliability of subsequent analysis. When determining the ocean environment pattern according to the optimized grouping set.

[0090] It can be found that the data points in the high-temperature area are mostly distributed in the upper half of the space, while those in the low-temperature area are in the lower half. The final mode description clearly depicts the characteristics of the data distribution. This distribution law is helpful for subsequent environmental prediction and monitoring.

[0091] S103. Extract the normal state mode library of this sea area from historical data, and use the anomaly detection algorithm to compare the deviation degree between the current marine environment mode and the normal state mode. If the deviation exceeds the preset threshold, it is determined as an abnormal state.

[0092] Furthermore, when using the anomaly detection algorithm to compare the deviation between the current marine environment mode and the normal state mode, the process of determining the abnormal state includes: constructing a mode library of the normal state through historical data to obtain the mode library representation; comparing the mode library representation with the current marine environment mode to obtain the deviation degree data; using the anomaly detection algorithm to process the deviation degree data to obtain the anomaly detection result; if the anomaly detection result exceeds the preset threshold, it is determined as an abnormal state.

[0093] Even further, as a specific implementation manner of this embodiment, when constructing the mode library of the normal state through historical data, it can be understood as extracting typical features from long-term marine environment observations.

[0094] This embodiment collects the temperature and salinity data of the past five years to form a stable reference set.

[0095] In a possible implementation manner, assuming that the temperature is usually between 18 - 22 degrees Celsius and the salinity is between 32 - 34, the mode library representation is generated by statistical mean and range to reflect the distribution law under normal conditions. This method is convenient for subsequent comparison. When comparing the mode library representation with the current mode.

[0096] Specifically, the difference analysis can be performed on the observation data of the current week and the reference in the mode library.

[0097] The current temperature is 23 degrees Celsius and the salinity is 35, exceeding the normal range. When calculating the deviation degree data, it can be simply represented by the difference. For example, the temperature deviation is 1 - 5 degrees Celsius and the salinity deviation is 1 - 3. This deviation data intuitively reflects the deviation of the current state. When applying the anomaly detection algorithm to the deviation degree data.

[0098] Preferably, if the deviation exceeds twice the standard deviation of the mean, it is marked as a potential anomaly.

[0099] For example, the temperature deviation of 5 degrees Celsius exceeds the normal fluctuation range, and the algorithm determines it as an abnormal point. This method is simple and efficient, and can quickly screen out abnormal states. When the anomaly detection result exceeds the preset threshold and it is determined as an abnormal state.

[0100] It should be noted that the threshold can be set according to business requirements.

[0101] In one embodiment, it is set that a temperature deviation greater than 3 degrees Celsius is abnormal. Combining the foregoing example, the current state is determined to be abnormal and a state representation is generated. This representation provides a basis for subsequent adjustments. When adjusting the grouping basis of the current mode according to the state representation, it can be understood as redefining the classification criteria.

[0102] The original grouping basis of this embodiment is the fixed intervals of temperature and salinity. Now, according to the abnormal data, it is adjusted to a dynamic interval. For example, the upper limit of the high-temperature group is increased from 22 degrees Celsius to 24 degrees Celsius. The adjusted mode set is closer to the current environmental changes. When analyzing the distribution law of the environmental mode through the adjusted mode set.

[0103] Specifically, the aggregation of data points under the new grouping can be observed.

[0104] In a possible implementation manner, the data points of the high-temperature group are concentrated in the upper ocean, and the low-temperature group tends to be deeper, generating a distribution representation. This distribution law reveals the spatial characteristics of the environment. When updating the normal state of the mode library using the distribution representation.

[0105] This embodiment incorporates the new high-temperature interval into the mode library and replaces the old reference range.

[0106] S104. Classify the detected abnormal state. Use a deep learning model to learn the characteristic patterns of different types of abnormalities from historical abnormal data, and classify the current abnormality into predefined abnormality types through a pattern matching algorithm.

[0107] Furthermore, the process of obtaining the classified abnormal state includes: obtaining an abnormal feature set through historical data, training the abnormal feature set using a deep learning model to obtain a type representation; extracting feature data from the current abnormal state, applying a pattern matching algorithm to the feature data to determine the matching degree with the type representation; classifying the abnormal state into a predefined abnormal type based on the matching degree to obtain the classified abnormal state; adjusting the boundary conditions of the type representation for the classified abnormal state to obtain an updated type definition; using the updated type definition to re-compare the abnormal features in the historical data to obtain an optimized feature extraction rule.

[0108] Even further, as a specific implementation manner of this embodiment, when obtaining the abnormal feature set through historical data, it can be understood as screening out sample data deviating from the normal state from long-term ocean environment records.

[0109] Suppose that there have been multiple cases of sudden temperature rises or sudden salinity drops in the past ten years, and these abnormal points are summarized into a feature set.

[0110] Specifically, a certain record shows that the temperature reaches 25 degrees Celsius and the salinity drops to 30, significantly deviating from the normal range. These data points form the basis of the abnormal characteristics. When using a deep learning model to train the abnormal characteristics, one possible implementation is to learn the internal pattern of the abnormality through a neural network.

[0111] In this embodiment, the temperature, salinity, and time series are input into the model. After training, a type characterization is generated, which may be "high temperature and low salinity type" or "instantaneous fluctuation type".

[0112] In one embodiment, the model identifies that high-temperature anomalies are often accompanied by changes in surface water flow, forming a unique type characterization. When extracting feature data from the current anomaly.

[0113] Preferably, this embodiment focuses on the key indicators in real-time monitoring. Assuming that the current observed temperature is 24 degrees Celsius and the salinity is 31, these values are extracted as feature data. When applying a pattern matching algorithm to the feature data, it can be understood as comparing the current data with the trained type characterization.

[0114] This embodiment calculates the similarity between the current feature and the "high temperature and low salinity type" and obtains a matching degree of 85%. If the matching degree exceeds a preset threshold, such as 80%, it is determined that the current anomaly belongs to the corresponding type.

[0115] It should be noted that this classification result provides a clear direction for subsequent processing. For example, when the current anomaly is classified as the "high temperature and low salinity type", it helps to quickly locate the source of the problem. When adjusting the boundary conditions of the type characterization according to the classification result.

[0116] In one embodiment, if multiple high-temperature anomalies exceed 24 degrees Celsius, the original boundary of 23 degrees Celsius can be adjusted to 25 degrees Celsius to generate an updated type definition. This adjustment makes the type closer to the actual change. When using the updated type definition to re-compare the abnormal characteristics in historical data.

[0117] This embodiment re-examines the past data and finds that some unrecognized anomalies now meet the new boundary. Therefore, the optimized feature extraction rule is more accurate.

[0118] In one embodiment, the new rule gives priority to the time window of temperature mutation and screens more effective features. When processing the current anomaly through the optimized feature extraction rule, it can be judged whether the algorithm application strategy needs to be adjusted.

[0119] S105. For the classified abnormal state, extract the change sequence of the recent environmental parameters in this sea area from the time dimension, use the time series analysis algorithm to predict the development trend of the anomaly, and evaluate the anomaly diffusion risk in combination with the environmental status of the spatially adjacent areas.

[0120] Further, the process of evaluating the abnormal diffusion risk includes: obtaining time series data through the classified abnormal states; processing the time series data using time series analysis algorithms to obtain the change sequence of the abnormal states; extracting the dynamic characteristics of environmental parameters based on the change sequence of the abnormal states to determine the development trend of the abnormal states; obtaining the diffusion characteristics from the environmental parameters in the spatially adjacent regions and judging the distribution range of the diffusion characteristics; and obtaining the level of the abnormal diffusion risk based on the development trend and diffusion characteristics of the abnormal states.

[0121] Furthermore, as a specific implementation manner of this embodiment, time series data is obtained through the classified abnormal states and processed using time series analysis algorithms to obtain the change sequence of the abnormal states. The dynamic characteristics of environmental parameters are extracted according to the change sequence to determine the development trend of the abnormal states. The diffusion characteristics are obtained from the environmental parameters in the spatially adjacent regions and the distribution range of the diffusion characteristics is judged. For the development trend and diffusion characteristics, clustering analysis algorithms are used for processing to obtain the diffusion risk level of the abnormal states. If the diffusion risk level exceeds the preset threshold, the boundary conditions of the time series are updated through the regional environmental data to obtain the adjusted change sequence. The sequence analysis is re-run according to the adjusted change sequence to obtain the optimized development trend. The classification rules of the abnormal states are updated through the optimized development trend and diffusion characteristics to obtain the improved state classification result.

[0122] S106. According to the abnormal type, development trend, and diffusion risk, call the pre-set response strategy library, select the response plan with the highest matching degree, and generate an abnormal event report including abnormal description, impact assessment, and recommended measures.

[0123] Further, the process of generating an abnormal event report of the marine environment in the current sea area based on the abnormal type, abnormal development trend, and abnormal diffusion risk includes: obtaining the response strategies in the pre-set strategy library through the combination of the abnormal type and abnormal diffusion risk, and determining the plan with the highest matching degree; extracting the dynamic characteristics from the abnormal development trend, judging the correlation between the abnormal diffusion risk and the abnormal development trend, and obtaining the preliminary distribution of the risk level; and adjusting the priority of the response strategy according to the preliminary distribution of the risk level and the abnormal type to obtain the optimized strategy plan.

[0124] Furthermore, as a specific implementation manner of this embodiment, when obtaining the response strategy through the combination of the abnormal type and diffusion risk, it can be understood as extracting the matching plan from a pre-designed strategy library.

[0125] Suppose the abnormal type is "water pollution", and the diffusion risk is divided into three levels: "low, medium, and high". The strategy library includes options such as "local isolation", "regional monitoring", and "comprehensive treatment".

[0126] If the anomaly type is "water pollution" and the spread risk is "high", the system will first match the "comprehensive treatment" plan because its coverage and intensity match the high risk. The determination of the matching degree can be based on the success rate of historical cases. For example, when treating the pollution in a certain sea area, the efficiency of comprehensive treatment in restoring water quality in a similar scenario reached 80%, so it was selected as the optimal plan. When extracting dynamic features from the development trend.

[0127] As a specific implementation manner of this embodiment, during the monitoring of a certain sea area, the pollution concentration increased from 10 mg / L to 15 mg / L, and the change rate increased by 0.5 mg / L per day. This trend indicates that the anomaly is intensifying. When judging the relevance between the spread risk and the development trend, time window analysis can be used. For example, the pollution range expanded from 1 square kilometer to 3 square kilometers within the past 7 days, indicating a high correlation between the spread risk and the increasing concentration trend. Therefore, the initial distribution of the risk level is set as "high", providing a basis for subsequent strategy adjustment. When adjusting the priority of the response strategy.

[0128] As a specific implementation manner of this embodiment, if the risk level is "high" and the anomaly type involves "heavy metal pollution", the priority is adjusted to "urgent isolation" being superior to "long-term monitoring" because the former can contain the spread faster. After integrating the anomaly description (such as "heavy metal exceeding the standard") and the impact assessment (such as "the impact on fishery production decreased by 30%"), the generated event summary includes the dynamic feature "spread speed 0.2 square kilometers per day", and the applicable scope of the recommended measures is limited to "the core pollution area and the surrounding 5 kilometers". When using the support vector machine algorithm to process the development trend and the spread risk.

[0129] Preferably, the concentration change rate and the spread range can be used as input features to predict the distribution of anomaly events.

[0130] After inputting the concentration data and range data within 7 days in this embodiment, the algorithm predicts that the pollution will cover an additional 2 square kilometers within 3 days. This predicted distribution provides a basis for whether the risk level exceeds the threshold. If it exceeds the preset threshold (such as the range increases to 5 square kilometers), the strategy library will update the response measures, such as upgrading from "local isolation" to "joint treatment", and the adjusted recommended measures include "increasing the number of monitoring points to 10 and starting the purification equipment". When generating a complete event summary based on the predicted distribution and the optimized strategy plan.

[0131] In one embodiment, the anomaly description is "heavy metal pollution in a certain sea area, with a concentration of 20 mg / L", and the recommended measures are "deploying an isolation belt and starting the purification equipment, and monitoring the concentration change daily". This can not only respond to the anomaly quickly but also reduce the long-term impact of the spread through dynamic adjustment.

[0132] After the purification equipment operates, the concentration drops to 12 mg / L within 5 days, and the diffusion range is stable within 3 square kilometers. This method combines prediction and strategy optimization to ensure the pertinence and efficiency of response measures, providing reliable support for anomaly management.

[0133] S107. Use a distributed computing framework to process the continuously input ocean environment monitoring data stream, improve the computing efficiency through data sharding and parallel processing, and achieve real-time dynamic monitoring of a large area of sea areas.

[0134] Obtain the ocean environment monitoring data stream through a distributed computing framework, use data sharding technology to split the continuously input data, and obtain the sharded data set. For the sharded data set, obtain the parallel processing task assignment and determine the intermediate results processed by each node. Integrate the real-time dynamic features through the intermediate results, judge the change trend of the ocean environment, and obtain the preliminary distribution of dynamic monitoring. Use the random forest algorithm to process the preliminary distribution of dynamic monitoring, obtain the abnormal areas within the sea area range, and determine the abnormal distribution range. If the abnormal distribution range exceeds the preset threshold, adjust the parallel processing task through the processing framework to obtain the optimized intermediate results. Update the real-time dynamic features according to the optimized intermediate results, obtain the monitoring adjustment plan for a large area of sea areas, and determine the final distribution. Generate a dynamic monitoring summary of the ocean environment through the final distribution to obtain a complete event processing flow.

[0135] As a specific implementation manner of this embodiment, the distributed computing framework plays a key role in ocean environment monitoring, which can be understood as processing a large amount of data streams through multi-node collaboration.

[0136] As a specific implementation manner of this embodiment, the monitoring system continuously collects data such as temperature, salinity, and ocean current speed from sea area sensors, and the data volume per second reaches 10 GB. When using data sharding technology, the data can be split according to time periods, for example, each 5 minutes is a shard, generating multiple data sets of about 1 GB. This method is convenient for subsequent parallel processing and avoids overloading a single node.

[0137] Preferably, each shard can be assigned to different computing nodes, such as one node processes temperature data and another processes ocean current data to ensure balanced tasks. When obtaining the parallel processing task assignment for the sharded data set.

[0138] Specifically, it can be dynamically adjusted according to the node performance.

[0139] For example, high-performance nodes process slices with large changes in ocean current speed, while low-performance nodes process stable temperature data. The intermediate result is the local trend calculated by each node, such as the temperature in a certain sea area rising by 0.5 degrees Celsius per hour. After integrating these intermediate results, real-time dynamic features can be extracted, such as determining whether the acceleration of ocean currents in a certain area is related to temperature anomalies. The preliminary distribution shows that abnormal trends exist in 10% of the sea areas. This distributed method significantly improves the processing efficiency.

[0140] In one embodiment, when the random forest algorithm processes the preliminary distribution, it can train a model based on historical data to identify abnormal areas.

[0141] As a specific implementation of this embodiment, when the temperature in a certain sea area exceeds 28 degrees Celsius and the ocean current speed increases by 20%, the model predicts it as an abnormal point. The abnormal distribution range covers 50 square kilometers. If it exceeds a preset threshold, such as 30 square kilometers, the processing framework will increase node resources, such as expanding from 5 nodes to 8 nodes, to optimize the intermediate results. The updated dynamic features show that the abnormal area has expanded to 60 square kilometers, and the monitoring and adjustment plan will then cover a larger sea area.

[0142] It should be noted that when generating the dynamic monitoring summary from the final distribution, the spatial and temporal dimensions can be combined.

[0143] As a specific implementation of this embodiment, when the abnormal area in a certain sea area spreads from east to west at a speed of 2 kilometers per hour, the summary will record this trend and recommend dispatching additional monitoring ships to the western sea area. The complete event processing flow can also reflect the cause of the anomaly, such as undersea hydrothermal activity. This method not only improves the monitoring accuracy but also provides a basis for emergency decision-making.

[0144] As a specific implementation of this embodiment, if a sudden 5% drop in salinity is detected in the abnormal area during a certain monitoring, through multi-faceted analysis, it can be speculated that fresh water has flowed in. Distributed computing confirms that the area with the drop is 20 square kilometers, and the random forest predicts that it will spread to 30 square kilometers. The adjustment plan recommends deploying buoys to track the source of the fresh water. Examples from multiple directions support each other. For example, temperature, salinity, and ocean current data all point to the same abnormal event, ensuring the consistency of the analysis. This process can quickly respond to changes in the marine environment and ensure the comprehensiveness of monitoring.

[0145] S108. In view of the differences in environmental characteristics of different sea areas, construct an adaptive parameter adjustment mechanism, and continuously optimize the anomaly detection threshold, feature extraction weight, and prediction model parameters through feedback learning to improve the adaptability of the system to complex marine environments.

[0146] By collecting environmental data from different sea areas, a distribution set of sea area characteristics and environmental differences is obtained. According to the distribution set, a feedback learning method is used to adjust the weights of feature extraction, and an optimized feature set is obtained. For the optimized feature set, intermediate parameters of the prediction model are obtained to determine the preliminary environmental change trend. If the preliminary trend shows that the anomaly detection exceeds the preset detection threshold, the model parameters are updated through feedback learning to obtain an adjusted prediction result. According to the adjusted prediction result, the anomaly distribution range in a complex environment is obtained to determine whether further optimization is required. Through the anomaly distribution range, an adaptive parameter adjustment mechanism is used to update the detection threshold to obtain the final environmental adaptation plan. According to the final plan, a dynamic monitoring summary corresponding to the sea area characteristics is obtained to determine the complete event processing process.

[0147] As a specific implementation manner of this embodiment, by collecting environmental data from different sea areas, a distribution set of sea area characteristics and environmental differences is obtained.

[0148] For example, data such as temperature, salinity, and ocean current velocity are collected from multiple sea areas such as the East China Sea and the South China Sea to initially form a distribution set containing various environmental variables. These data show that the surface water temperature in the East China Sea fluctuates around 25 degrees, while the water temperature in the South China Sea is generally higher, reaching above 28 degrees, and the salinity varies due to differences in rainfall.

[0149] As a specific implementation manner of this embodiment, such a distribution set can intuitively reflect the characteristic differences of different sea areas and provide a basis for subsequent analysis. According to the distribution set, a feedback learning method is used to adjust the weights of feature extraction, and an optimized feature set is obtained.

[0150] Specifically, feedback learning can compare historical data with current data to dynamically adjust the weight of each variable. For example, assuming that the initial model believes that the influence of temperature on environmental change accounts for 40%, and through multiple feedbacks, it is found that salinity changes can better reflect the anomaly trend, then the salinity weight can be increased to 50%, and the temperature can be reduced to 30%.

[0151] In a possible implementation manner, a certain area in the East China Sea is given a higher weight due to a sudden change in salinity, and the optimized feature set can better highlight the environmental differences driven by salinity. For the optimized feature set, intermediate parameters of the prediction model are obtained to determine the preliminary environmental change trend.

[0152] Preferably, the prediction model is based on time series analysis and combines the data of the previous 3 months to infer the temperature decline trend in a certain sea area within the next 15 days.

[0153] As a specific implementation manner of this embodiment, due to the influence of cold air in a certain area of the East China Sea, the temperature drops from 25 degrees to 22 degrees, and the intermediate parameters will record this change rate.

[0154] It is understandable that these parameters provide a basis for subsequent anomaly detection. If the preliminary trend shows that the anomaly detection exceeds the preset detection threshold, the model parameters are updated through feedback learning to obtain an adjusted prediction result.

[0155] It should be noted that assuming the threshold is set at a temperature change exceeding 3 degrees per week, if the temperature in a certain area of the East China Sea drops by 4 degrees, an update will be triggered.

[0156] In one embodiment, the model will re-analyze the ocean current and wind speed data, find that the weakening of the ocean current is the main cause, and thus adjust the parameters to make the prediction closer to the actual situation. According to the adjusted prediction result, the anomaly distribution range in a complex environment is obtained.

[0157] For example, in a certain area of the South China Sea, due to a typhoon, the temperature and salinity are both abnormal at the same time, and the distribution range covers 200 square kilometers. Through multi-faceted analysis, such as combining data of wind speed of 20 meters per second and rainfall of 50 millimeters per day, it is judged that the anomaly is driven by short-term weather rather than long-term trends. This multi-dimensional support ensures the accuracy of the distribution range. Through the anomaly distribution range, an adaptive parameter adjustment mechanism is used to update the detection threshold to obtain the final environment adaptation plan.

[0158] In one embodiment, if the anomaly range continues to expand, the threshold can be adjusted from 3 degrees per week to 4 degrees per week to avoid frequent false alarms.

[0159] As a specific implementation manner of this embodiment, the adjusted solution can better adapt to the high-fluctuation environment in the typhoon season and improve the robustness of monitoring. According to the final solution, the dynamic monitoring summary corresponding to the sea area characteristics is obtained, and the complete event processing process is determined.

[0160] As a specific implementation manner of this embodiment, the monitoring summary of a certain area in the East China Sea shows that "the temperature decline trend slows down, and the salinity anomaly needs to be continuously monitored", and the process includes links such as data collection, feature optimization, and anomaly confirmation.

[0161] Preferably, this method can timely capture environmental changes and provide support for countermeasures.

[0162] S109. Design a multi-scale data analysis module, automatically select appropriate data aggregation granularity and analysis algorithms for monitoring requirements with different spatial ranges and time spans, and realize multi-level ocean environment analysis from short-term fluctuations in local sea areas to long-term trends at the global scale.

[0163] By monitoring requirements, obtain the distribution set of spatial range and time span, and determine the data characteristics. According to the data characteristics, use a clustering algorithm to adjust the aggregation granularity to obtain an optimized data aggregation result. For the optimized data aggregation result, obtain the characteristic distributions of short-term fluctuations and long-term trends, and judge the analysis level. If the short-term fluctuations exceed the preset threshold, adjust the analysis algorithm through feedback to obtain the change characteristics of the local sea area. If the long-term trend deviates from the preset range, extract the global-scale change pattern through a convolutional neural network to determine the adjusted trend distribution. According to the change characteristics and trend distribution, adopt a multi-level analysis method to obtain a dynamic monitoring summary from local to global. Update the aggregation granularity and analysis algorithm through the dynamic monitoring summary to determine the final multi-scale analysis result.

[0164] As a specific implementation manner of this embodiment, by monitoring requirements, obtain the distribution set of spatial range and time span, and determine the data characteristics.

[0165] It can be understood that the spatial range covers dozens of kilometers in the local sea area to thousands of kilometers at the global scale, and the time span ranges from short-term monitoring at the hourly level to long-term analysis at the annual level.

[0166] As a specific implementation manner of this embodiment, in a certain area of the East China Sea, the monitoring requirements lock in the ocean current changes within a range of 50 kilometers, and the time span is 72 hours. By collecting temperature, salinity, and flow velocity data, it is determined that the data characteristics are coexistence of high-frequency fluctuations and low-frequency trends. According to the data characteristics, use a clustering algorithm to adjust the aggregation granularity to obtain an optimized data aggregation result.

[0167] As a specific implementation manner of this embodiment, K-means clustering can be used to group similar data points. In the above-mentioned East China Sea scenario, if the temperature data is collected every 10 minutes, the 72-hour data can be clustered by hour to reduce noise interference, and the aggregation granularity is adjusted from the minute level to the hour level. This can more clearly capture the law of ocean current changes. For the optimized data aggregation result, obtain the characteristic distributions of short-term fluctuations and long-term trends, and judge the analysis level.

[0168] Specifically, the short-term fluctuations can be obtained by statistically calculating the standard deviation of the data within each hour. For example, if the temperature fluctuation exceeds 0.5 degrees Celsius, it is regarded as abnormal; the long-term trend is extracted by the moving average method. For example, the salinity has gradually decreased by 0.2 units in the past 30 days. Therefore, the analysis level is divided into two dimensions: local anomaly detection and global trend assessment. If the short-term fluctuations exceed the preset threshold, adjust the analysis algorithm through feedback to obtain the change characteristics of the local sea area.

[0169] In a possible implementation, if the standard deviation of the temperature in a certain hour reaches 0.8 degrees Celsius, exceeding the threshold of 0.5, an adaptive filter can be introduced. After removing accidental noise and recalculating, the characteristics of the local ocean current acceleration can be obtained. This adjustment can improve the accuracy of anomaly recognition. If the long-term trend deviates from the preset range, the convolutional neural network is used to extract the global-scale change pattern to determine the adjusted trend distribution.

[0170] As a specific implementation of this embodiment, when the decrease in the salinity of the East China Sea exceeds the expected range by 0.1 unit, the convolutional neural network can be used to analyze the global sea temperature and salinity image data, extract the change pattern under the influence of El Niño, and adjust the trend prediction to be closer to the actual situation. According to the change characteristics and trend distribution, a multi-level analysis method is adopted to obtain a dynamic monitoring summary from local to global.

[0171] Preferably, the local characteristics show an abnormal acceleration of the ocean current, and the global trend reflects the influence of the climate pattern. The combination of the two generates a dynamic summary, such as "the local ocean current anomaly in the East China Sea is driven by global warming". This provides a comprehensive perspective for multi-scale monitoring. By updating the aggregation granularity and analysis algorithm through the dynamic monitoring summary, the final multi-scale analysis result is determined.

[0172] In one embodiment, if the summary shows frequent short-term anomalies, the aggregation granularity can be shortened from the hourly level to the half-hourly level, and at the same time, the algorithm parameters are optimized to improve the response ability to rapid changes. This adaptive update can significantly improve the real-time performance and reliability of the monitoring.

[0173] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for monitoring the marine environment based on a low-orbit satellite, characterized in that: The following steps are involved: Preprocess multi-source heterogeneous data to obtain a unified representation vector; Using a clustering algorithm to perform cluster analysis on the unified characterization vector to obtain a current ocean environment model; Using an anomaly detection algorithm to compare the deviation between the current ocean environment mode and the normal state mode to determine the abnormal state; Using a deep learning model to classify the abnormal state to obtain an abnormal type, and classifying the abnormal type into a predefined abnormal type to obtain a classified abnormal state; Using a time series analysis algorithm to predict the abnormal development trend of the classified abnormal state, and evaluating the abnormal diffusion risk based on the predicted abnormal development trend; Generate a marine environment abnormal event report for the current sea area based on the abnormal type, abnormal development trend and abnormal diffusion risk; A distributed computing framework is constructed, and based on the distributed computing framework, reports of abnormal marine environmental events in several sea areas are dynamically monitored to obtain a multi-level marine environmental analysis ranging from short-term fluctuations in local sea areas to long-term trends at the global scale.

2. The method for monitoring the marine environment based on a low-orbit satellite according to claim 1, characterized in that: The process of preprocessing multi-source heterogeneous data to obtain a unified representation vector includes: Performing standardization processing on the multi-source heterogeneous data to obtain standardized data; Using a principal component analysis algorithm to extract key features from the standardized data to obtain a feature set; Using preset weights to perform weighted fusion on the feature set to obtain an initial representation vector; Determining the distribution law of the initial characterization data to obtain an environmental state characterization; A unified representation vector of multi-source data is generated based on the environmental state representation.

3. The method for monitoring the marine environment based on a low-orbit satellite according to claim 2, characterized in that: The process of using a clustering algorithm to perform cluster analysis on the unified characterization vector to obtain the current ocean environment model includes: Constructing a multidimensional feature space based on the unified representation vector, and grouping the data points in the multidimensional feature space using a clustering algorithm to obtain a preliminary grouping result; Calculate the distance between classes based on the preliminary grouping result to obtain a distance calculation result; Analyze the intra-class compactness based on the distance calculation result to obtain compactness analysis data; Determining the optimal number of clusters based on the compactness analysis data and the distance calculation result to obtain a cluster number representation; The cluster number representation is used to adjust the preliminary classification result to obtain an optimized grouping set; Determine the ocean environment mode according to the optimized grouping set and obtain the environmental mode representation; The distribution law of multi-dimensional feature space is generated through environmental pattern characterization to obtain the current ocean environment pattern.

4. The method for monitoring the ocean environment based on a low-orbit satellite according to claim 3, characterized in that: The process of using an anomaly detection algorithm to compare the deviation between the current ocean environment mode and the normal state mode and determining the abnormal state includes: Build a normal state pattern library through historical data and obtain the pattern library representation; The model library representation is compared with the current ocean environment model to obtain the deviation degree data; Anomaly detection algorithms are used to process the deviation degree data to obtain anomaly detection results; If the abnormal detection result exceeds the preset threshold, it is determined to be an abnormal state.

5. The method for monitoring the ocean environment based on a low-orbit satellite according to claim 3, characterized in that: The process of obtaining the classified abnormal state includes: Obtain anomaly feature sets through historical data, train the anomaly feature sets using a deep learning model, and obtain type representation; Extracting feature data from the current abnormal state, applying a pattern matching algorithm to the feature data, and determining the degree of match with the type representation; Classifying the abnormal state into a predefined abnormal type based on the matching degree to obtain a classified abnormal state; The abnormal state after classification adjusts the boundary conditions of the type representation to obtain the updated type definition; The updated type definition is used to re-match the abnormal features in the historical data to obtain the optimized feature extraction rules.

6. The method for monitoring the ocean environment based on a low-orbit satellite according to claim 5, characterized in that: The process of assessing the risk of abnormal proliferation includes: Obtain time series data through classified abnormal states; Using a time series analysis algorithm to process the time series data to obtain a change sequence of abnormal states; Extracting dynamic characteristics of environmental parameters based on the change sequence of the abnormal state to determine the development trend of the abnormal state; Obtain diffusion characteristics from environmental parameters of adjacent areas in space and determine the distribution range of diffusion characteristics; Based on the development trend and diffusion characteristics of the abnormal state, the level of abnormal diffusion risk is obtained.

7. The method for monitoring the ocean environment based on a low-orbit satellite according to claim 6, characterized in that: The process of generating a marine environment abnormal event report of the current sea area based on the abnormal type, abnormal development trend and abnormal diffusion risk includes: By combining the anomaly type and the risk of anomaly diffusion, we can obtain the response strategies in the pre-set strategy library and determine the solution with the highest matching degree. Extract dynamic features from the abnormal development trend, determine the correlation between abnormal diffusion risk and abnormal development trend, and obtain the preliminary distribution of risk levels; According to the preliminary distribution of risk levels and abnormal types, the priority of response strategies is adjusted to obtain optimized strategic solutions.

8. The method for monitoring the ocean environment based on a low-orbit satellite according to claim 7, characterized in that: The process of dynamically monitoring reports of marine environmental abnormal events in several sea areas based on the distributed computing framework to obtain multi-level marine environmental analysis from short-term fluctuations in local sea areas to long-term trends on a global scale includes: By collecting environmental data from different sea areas, we can obtain a distribution set of sea area characteristics and environmental differences; Obtain the distribution set of sea area characteristics and environmental differences through monitoring needs and determine data characteristics; According to the data characteristics, clustering algorithm is used to adjust the aggregation granularity to obtain optimized data aggregation results; Based on the optimized data aggregation results, characteristic distribution of short-term fluctuations and long-term trends is obtained to determine the analysis level; If the short-term fluctuation exceeds the short-wave threshold, the analysis algorithm is adjusted through feedback to obtain the change characteristics of the local sea area; If the long-term trend deviates from the long-wave range, the global-scale variation pattern is extracted through a convolutional neural network to determine the adjusted trend distribution; A multi-level analysis method is used based on the changing characteristics of local sea areas and the adjusted trend distribution to obtain a dynamic monitoring overview from local to global.

Images