Marine ecological anomaly detection method and system based on multi-source data

Abstract

The invention relates to the field of ecological monitoring, in particular to a marine ecological anomaly detection method and system based on multi-source data. Comprising the following steps: acquiring multi-source dynamic data of a target sea area; constructing a dynamic quality evaluation system, inputting multi-source dynamic data, and outputting a quality control report; based on the multi-source dynamic data conforming to the preset quality threshold value, establishing a marine ecological dynamic database in combination with a spatio-temporal indexing mechanism; based on the marine ecological dynamic database, space-time anomaly features are extracted, and candidate anomaly regions are determined; and generating a marine ecology comprehensive abnormal index by adopting the dynamic weight calculation model, outputting a prediction result through a preset abnormal trend prediction model based on the marine ecology comprehensive abnormal index, and generating an early warning measure. Through a distributed crawler and sensor gateway collaborative architecture, comprehensive acquisition of macroscopic data and near-shore real-time data is realized, and high-quality basic data is provided for subsequent detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ecological monitoring, and in particular to a marine ecological anomaly detection method and system based on multi-source data. BACKGROUND

[0002] In recent years, ecological disasters have occurred frequently along the coast of China, and research on ecological disasters has become a hot topic in the academic community. The concept of marine ecological disaster was originally derived from the definition of land ecological disaster, and the deterioration of the coastal and offshore ecological environment caused by accidents such as red tide, marine pollution, and oil spills is included in marine ecological disaster. At present, marine ecological disasters in China have become more and more serious, so it is necessary to carry out early warning of marine ecological disasters to minimize the losses caused by marine disasters. Studies have shown that using remote sensing images and numerical simulation methods is an effective means for marine ecological disaster warning and disaster emergency decision-making. Marine ecological disasters are events influenced by multiple factors, and the biological, chemical, and physical mechanisms of their development and extinction processes are complex. Therefore, based on big data and high-intelligent parallel computing, solving complex and variable problems such as marine ecological anomalies is the main development direction for further improving the management of marine resources.

[0003] In the existing marine ecological anomaly detection method, there are significant technical bottlenecks in the integration and quality control of multi-source data. Due to the large differences in spatial and temporal scales, format standards, and precision characteristics of data from different sources, and the lack of unified quality evaluation and calibration mechanisms, multi-source data has problems such as temporal and spatial inconsistency, insufficient completeness, and poor correlation. For example, macro remote sensing data and nearshore real-time monitoring data often have data conflicts due to different collection frequencies and coordinate systems, and there may be missing values, jumps, and other abnormal values in the data, making it difficult to be directly used for marine ecological dynamic analysis. Such data-level defects can further lead to large deviations and low precision in subsequent anomaly detection results, making it difficult to accurately identify marine ecological anomaly regions and trends, and seriously affecting the timeliness and reliability of marine ecological early warning. SUMMARY

[0004] In view of the technical defects in the background art, the present application proposes a marine ecological anomaly detection method and system based on multi-source data, which solves the above technical problems and meets the actual needs. The specific technical solution is as follows: The marine ecological anomaly detection method based on multi-source data comprises the following steps: Obtain multi-source dynamic data of the target sea area through a distributed crawler and a sensor gateway cooperative architecture; Construct a dynamic quality evaluation system based on data attributes, input multi-source dynamic data, output a quality control report, and output multi-source dynamic data meeting the preset quality threshold based on the quality control report; Based on the multi-source dynamic data meeting the preset quality threshold, a marine ecological dynamic database is established in combination with a space-time index mechanism, the marine ecological dynamic database including an original data layer, a preprocessing layer and a feature extraction layer, based on the data stored in the preprocessing layer, a target sea area three-dimensional dynamic flow field and an ecological parameter distribution are generated through a preset physical and biological coupled numerical model, and are stored in the marine ecological dynamic database; Based on the marine ecological dynamic database, a space-time anomaly feature is extracted and a candidate anomaly region is determined; According to the anomaly contribution degree of the space-time anomaly feature corresponding parameter in the candidate anomaly region, a marine ecological comprehensive anomaly index is generated by using a dynamic weight calculation model, and based on the marine ecological comprehensive anomaly index, a prediction result is output and a warning measure is generated by using a preset anomaly trend prediction model.

[0005] Further, the multi-source dynamic data includes remote sensing satellite data, buoy monitoring data, weather station observation data, underwater sensor array data and historical reanalysis data, and the specific steps of obtaining the multi-source dynamic data include: By deploying distributed crawler nodes, using asynchronous IO mechanism and multi-protocol adaptation technology, macro data of the target sea area is regularly captured; A sensor gateway cluster based on a low-power wide-area network communication protocol is built to collect nearshore real-time data; The macro data and the nearshore real-time data are consistently processed by a space-time calibration algorithm to form unified multi-source dynamic data.

[0006] Further, the quality control report includes data quality score, anomaly details and repair scheme, and the specific steps of obtaining the quality control report are as follows: According to the attribute differences of the multi-source dynamic data, a dynamic quality evaluation system including three-level evaluation indexes and dynamic thresholds is constructed, the three-level evaluation indexes including integrity index, time sequence consistency index and spatial correlation index; Based on the integrity index, the multi-source dynamic data is automatically scanned and verified to generate an integrity verification list; Based on the time sequence consistency index and the spatial correlation index, the multi-source dynamic data is subjected to deep verification in time sequence and spatial dimension to generate time sequence and spatial anomaly details; Based on the integrity verification list and the time sequence and spatial anomaly details, the integrity, time sequence consistency and spatial correlation of the multi-source dynamic data are quantified by using the percentage system to generate the data quality score and the anomaly details, the anomaly is traced based on the anomaly details and the repair scheme is generated, and the quality control report is generated in combination with the data quality score, the anomaly details and the repair scheme.

[0007] Further, the original data layer stores multi-source dynamic data that is not processed and meets a preset quality threshold by using a distributed file system, the preprocessing layer performs standardized processing on the multi-source dynamic data that meets the preset quality threshold, and is configured to store the multi-source dynamic data after the standardized processing, the standardized processing includes data format conversion, unit unification, and coordinate calibration, the feature extraction layer performs feature dimension reduction on the data after the standardized processing by using a principal component analysis method, and is configured to store the feature vector after the dimension reduction, and the spatio-temporal index mechanism adopts a hybrid index of space and time.

[0008] Further, the physical and biological coupled numerical model includes a physical model layer and a biological model layer, and the physical model layer and the biological model layer cooperate with each other through a coupling mechanism. The specific steps of generating the three-dimensional dynamic flow field and the ecological parameter distribution of the target sea area are as follows: Based on the multi-source dynamic data in the preprocessing layer and the target sea area, a physical model is established, a biological model is integrated based on the physical model, the physical model and the biological model are coupled through a coupling mechanism, and a physical and biological coupled numerical model is generated; The parameters of the physical model and the biological model are respectively optimized by using a particle swarm algorithm, and the physical and biological coupled numerical model is updated; Based on the updated physical and biological coupled numerical model, the multi-source dynamic data in the preprocessing layer is input, and the three-dimensional dynamic flow field and the ecological parameter distribution of the target sea area are output.

[0009] Further, the spatio-temporal anomaly feature is extracted by using a multi-scale sliding window algorithm, and the specific steps are as follows: Based on the dynamic characteristics of different parameters in the multi-source dynamic data, different time windows and space windows are designed; Based on the parameters in the time window and the space window, time window statistics and space window statistics are calculated; Different anomaly determination thresholds are set based on the characteristic differences of different parameters, and the spatio-temporal anomaly feature is screened from the time window statistics and the space window statistics by using the anomaly determination thresholds.

[0010] Further, the specific steps of determining the candidate abnormal area are as follows: Based on the spatial correlation characteristics of the marine ecological anomaly, an improved isolated algorithm model is constructed; The spatio-temporal anomaly feature is input into the isolated algorithm model, and an anomaly score is output and the candidate abnormal area is screened out; The authenticity of the candidate abnormal area is verified by using three types of data, and the candidate abnormal area that passes the verification is divided into different anomaly levels.

[0011] Further, the specific steps of generating the marine ecological comprehensive anomaly index are as follows: Determine the corresponding parameters based on the spatio-temporal anomaly features of the candidate abnormal regions, and quantify the abnormal values of each parameter; Based on the abnormal values of each parameter and the preset influence weight, the abnormal contribution degree of each parameter is calculated; Obtain the dynamically adjusted parameter weight through the preset dynamic weight calculation model, and based on the parameter weight and the abnormal contribution degree of each parameter, obtain the marine ecological comprehensive anomaly index through weighted summation.

[0012] Further, the specific steps of obtaining the prediction result and generating the early warning measures are as follows: Extract historical data from the marine ecological dynamic database as training samples, and construct an abnormal trend prediction model through the training samples; Input the marine ecological comprehensive anomaly index into the abnormal trend prediction model, output the comprehensive anomaly index prediction sequence containing a period of time in the future, and generate the prediction result containing early warning information based on the comprehensive anomaly index prediction sequence and the preset graded early warning threshold; Based on the early warning information in the prediction result, retrieve the corresponding early warning measures from the preset early warning database.

[0013] The marine ecological anomaly detection system based on multi-source data comprises a processor and a memory for storing executable instructions of the processor; The processor is configured to execute the instructions to implement the steps of the marine ecological anomaly detection method based on multi-source data according to any one of claims 1 to 9.

[0014] Compared with the prior art, the marine ecological anomaly detection method and system based on multi-source data provided by the present application has the following beneficial effects: The present application realizes comprehensive collection of macro data and nearshore real-time data through the cooperative architecture of distributed crawler and sensor gateway, combined with asynchronous IO mechanism, multi-protocol adaptation technology and low-power wide-area network communication protocol; then, through the consistency processing of the spatio-temporal calibration algorithm, the spatio-temporal heterogeneity of multi-source data is effectively eliminated, forming a unified dynamic data set, which greatly improves the data coverage range and consistency, and provides high-quality basic data for subsequent detection.

[0015] The present application constructs a quality evaluation system containing three-level evaluation indexes and dynamic threshold, generates a quality control report through automatic scanning and deep verification, and realizes the whole-process management of multi-source data from abnormal identification to trace repair. Compared with the traditional static quality evaluation, this system can accurately quantify the data quality and repair it specifically, significantly improving the integrity, time sequence and spatial consistency of the data, and providing reliable data support for ecological anomaly detection.

[0016] The application adopts a layered storage and space-time hybrid index mechanism to realize efficient management and rapid retrieval of multi-source data; a target sea area three-dimensional dynamic flow field and ecological parameter distribution are generated through a physical and biological coupled numerical model, the physical environment and biological process are deeply related, the defects that the traditional single model is difficult to reflect the ecological dynamics are overcome, and a more realistic ecological parameter basis is provided for abnormal feature extraction.

[0017] The application extracts space-time abnormal features by using a multi-scale sliding window algorithm, determines a candidate abnormal area by combining an improved isolated algorithm model, generates a comprehensive abnormal index through a dynamic weight calculation model, and finally outputs early warning measures based on a trend prediction model. The process realizes the whole-chain precision processing from abnormal feature identification to area positioning and then to trend prediction, and compared with the traditional method, the abnormal detection precision is higher, the early warning timeliness is stronger, and scientific basis can be provided for marine ecological protection and emergency decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 FIG. 1 is a flowchart of a marine ecological anomaly detection method based on multi-source data according to the application. DETAILED DESCRIPTION

[0019] In the description of the application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "middle", "inner" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the devices or elements indicated must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application. In addition, the terms "first", "second" and the like are only for the purpose of description and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features limited by "first", "second" and the like can explicitly or implicitly include one or more of the features. In the description of the application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integrally connected; it can be directly connected, or indirectly connected through an intermediate medium, or the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood through specific circumstances.

[0020] The embodiments of the application will be described below in conjunction with the drawings and related embodiments, and the embodiments of the application are not limited to the following examples, and the application relates to the necessary components in the related technical field, which should be regarded as known technology in the technical field, and is known and mastered by the person skilled in the art.

[0021] Referring to Figure 1 The application provides a marine ecological anomaly detection method based on multi-source data, and the steps are explained as follows: Step S100, multi-source dynamic data of a target sea area is obtained through a distributed crawler and a sensor gateway cooperative architecture.

[0022] The distributed crawler is a parallelized data acquisition technology that achieves efficient crawling of Internet public data sources through multi-node cooperation, such as a Python-based distributed crawler cluster. The sensor gateway is an intermediate device connecting physical sensors and data processing systems, supporting multi-protocol adaptation, and can receive real-time data from near-shore monitoring equipment and transmit it through a low-power wide-area network. The cooperative architecture integrates the two types of data sources through a unified data interface and communication protocol. The technical operations include: the distributed crawler periodically crawls data according to preset rules, the sensor gateway receives sensor data in real time, both of them process data streams in parallel through asynchronous IO mechanism, and use multi-protocol adaptation technology to unify data format. Through synchronous acquisition and protocol standardization, the problem of limited data coverage and inconsistent format in traditional methods is solved, and the comprehensiveness and real-time performance of data acquisition are improved.

[0023] Step S200, a dynamic quality evaluation system is constructed based on data attributes, multi-source dynamic data is input, a quality control report is output, and multi-source dynamic data meeting the preset quality threshold is output based on the quality control report.

[0024] The dynamic quality evaluation system includes three evaluation indexes of data integrity, spatio-temporal consistency and parameter accuracy. Data integrity evaluation can be achieved by detecting the proportion of missing values in time series, spatio-temporal consistency evaluation is completed by comparing the differences in spatial coordinates and time stamps of different data sources, and parameter accuracy evaluation uses statistical methods to identify outliers. The dynamic threshold is adjusted adaptively according to the distribution of historical data or real-time environmental conditions, for example, some parameter threshold ranges can be appropriately relaxed during typhoon period. Automatic scanning quickly identifies outliers beyond the physical range through preset rules, and deep checking traces the causes of outliers and attempts to repair through time series interpolation, spatial neighborhood analysis and other algorithms. Through the hierarchical evaluation and repair mechanism, the problem that traditional static quality control cannot cope with dynamic environmental changes is solved, and the data integrity and spatio-temporal consistency are improved.

[0025] Step S300, based on multi-source dynamic data meeting the preset quality threshold, a marine ecological dynamic database is established combining with a spatio-temporal indexing mechanism, the marine ecological dynamic database includes an original data layer, a preprocessing layer and a feature extraction layer. Based on the data of the preprocessing layer, a three-dimensional dynamic flow field and ecological parameter distribution are generated through a preset physical and biological coupled numerical model, and stored in the marine ecological dynamic database.

[0026] The hierarchical storage refers to classifying and managing data according to processing stages: the raw data layer stores raw data without processing, the preprocessing layer stores standardized data after quality control, and the feature extraction layer stores ecological parameters after model processing. The spatio-temporal hybrid index combines spatial index and time index to realize fast retrieval of multi-dimensional data. The physical and biological coupled numerical model integrates physical processes and biological processes, for example, the finite volume method is used to simulate a three-dimensional flow field, and the ecological dynamics equation is used to predict the distribution of biological parameters. Through hierarchical data management and coupled models, the problems of low storage efficiency of multi-source data and weak correlation of ecological parameters are solved, and high spatio-temporal resolution ecological background data are provided for anomaly detection.

[0027] Step S400, based on the marine ecological dynamic database, extracting spatio-temporal anomaly features and determining candidate anomaly regions.

[0028] The spatio-temporal anomaly features reflecting marine ecological anomalies are extracted from the marine ecological dynamic database, and the candidate regions where the ecological anomalies may exist are located according to the features, so as to realize the preliminary screening of the anomaly regions.

[0029] Step S500, according to the anomaly contribution degree of the parameters corresponding to the spatio-temporal anomaly features in the candidate anomaly region, a dynamic weight calculation model is used to generate a marine ecological comprehensive anomaly index, and based on the index, a prediction result is output through a preset anomaly trend prediction model and a warning measure is generated.

[0030] The dynamic weight calculation model allocates weights according to the sensitivity or historical contribution of parameters to ecological anomalies, for example, the weight of chlorophyll concentration in red tide detection may be higher than that of water temperature, and the weight can be determined by machine learning methods such as random forest or expert knowledge. Through the comprehensive contribution of multiple parameters and trend prediction, the problems of relying on a single index or lacking timeliness in traditional methods are solved, and the whole chain processing from anomaly identification to trend prediction is realized.

[0031] The present application improves the data coverage and real-time performance through distributed collection and protocol adaptation technology, guarantees the data integrity and consistency through dynamic quality evaluation system, generates high spatio-temporal resolution ecological background data through physical and biological coupled model, improves the positioning accuracy of abnormal regions through multi-scale analysis and improved algorithm, and realizes comprehensive warning through dynamic weight and trend prediction model. The scheme realizes the whole process solution from data collection to decision support by means of collaborative collection of multi-source data, dynamic quality control, database construction driven by coupled model, multi-scale anomaly detection and collaborative operation of comprehensive warning model, effectively solves the problems of data integration difficulty, quality control deficiency, single model and low warning precision in the prior art, and significantly improves the scientificity, timeliness and reliability of marine ecological anomaly detection.

[0032] In one embodiment of the present application, the multi-source dynamic data includes remote sensing satellite data, buoy monitoring data, weather station observation data, underwater sensor array data, and historical reanalysis data. The specific steps for obtaining the multi-source dynamic data include: deploying distributed crawler nodes, using asynchronous IO mechanism and multi-protocol adaptation technology to regularly capture macro data of the target sea area; building a sensor gateway cluster based on low-power wide-area network communication protocol to collect nearshore real-time data; and performing consistency processing on the macro data and nearshore real-time data through a space-time calibration algorithm to form unified multi-source dynamic data.

[0033] The distributed crawler nodes are distributed data collection units deployed in different geographic locations or computing nodes, and the multi-source data is concurrently acquired through asynchronous IO mechanism and multi-protocol adaptation technology. The asynchronous IO mechanism is a technical solution for improving data crawling efficiency through non-blocking I / O operations. This mechanism allows switching to process other tasks while waiting for network responses, avoiding single-thread blocking. The multi-protocol adaptation technology is a middleware system that supports multiple data source interfaces and format conversion. This technology can include OGC's WMS, WCS protocol analysis module. Through the above technical operations, the macro data of the target sea area can be regularly captured, and the macro information such as remote sensing satellite data and historical reanalysis data can be efficiently collected.

[0034] Building a sensor gateway cluster based on low-power wide-area network communication protocol is a technical solution for building a distributed data relay network. This cluster realizes long-distance low-power data transmission through LoRaWAN or NBIoT protocol, and supports distributed deployment of edge computing nodes. The low-power wide-area network communication protocol can be a wireless communication standard suitable for long-distance and low-bandwidth scenarios. The construction of the sensor gateway cluster can include a protocol analysis module and a data caching mechanism. The module can convert the binary stream of ModbusRTU protocol into MQTT message format and implement error retransmission strategy to ensure data reliability. Through this technical operation, nearshore real-time data can be collected, and stable acquisition of nearshore sensor information such as buoy monitoring data and weather station observation data can be realized, such as real-time monitoring of temperature, salinity or wind speed parameters.

[0035] The spatio-temporal calibration algorithm is a mathematical processing method for eliminating the spatio-temporal differences of multi-source data, and the algorithm includes spatio-temporal alignment, spatial resampling, parameter consistency correction and abnormal value fusion processing. The spatio-temporal alignment is a technology for realizing the continuous coverage of data sequence by time stamp synchronization and interpolation method, and the satellite transit time and the buoy observation time can be aligned to generate time series data. The spatial resampling is a technology for expanding point data into grid data by Kriging interpolation or inverse distance weighting method, and the buoy point data can be converted into 1km*1km grid mean value matched with satellite data. The parameter consistency correction is a technology for eliminating sensor system error based on statistical regression model, and the deviation between satellite inversion chlorophyll concentration and buoy measured value can be corrected by a random forest model. The abnormal value fusion processing is a technology for identifying sensor failure through cross verification of multi-source data, and the source of pollution detection conflict can be judged in combination with adjacent sensor data and historical trend. Through the above technical operation, the macro data and the nearshore real-time data are processed in consistency, and a spatio-tempally unified multi-source dynamic data set can be formed, for example, the system deviation of satellite and buoy in chlorophyll concentration measurement is eliminated.

[0036] The present application realizes efficient heterogeneous collection of macro data by deploying distributed crawler nodes, ensures high reliability of nearshore data transmission by sensor gateway cluster, eliminates spatio-temporal differences of multi-source data and generates unified data set by spatio-temporal calibration algorithm, and can achieve the technical effect of improving the automation level and quality of data integration, so as to provide more accurate input basis for subsequent ecological anomaly detection, for example, effectively identifying misjudgment caused by sensor drift or resolution limitation, and improving the accuracy of abnormal area positioning.

[0037] In an embodiment of the present application, the quality control report includes data quality score, abnormal details and repair scheme, and the specific steps of obtaining the quality control report are as follows: a dynamic quality evaluation system including three-level evaluation indexes and dynamic threshold is constructed according to the attribute differences of multi-source dynamic data, the three-level evaluation indexes include integrity index, time sequence consistency index and spatial correlation index; based on the integrity index, the multi-source dynamic data is executed automatic scanning verification to generate integrity verification list; based on the time sequence consistency index and the spatial correlation index, the multi-source dynamic data is executed deep verification in time sequence and spatial dimension to generate time sequence and spatial abnormal details; based on the integrity verification list and the time sequence and spatial abnormal details, the integrity, time sequence consistency and spatial correlation of the multi-source dynamic data are quantified by percentage system to generate data quality score and abnormal details, the abnormal details are traced to source to generate repair scheme, and the quality control report is generated in combination with the data quality score, the abnormal details and the repair scheme.

[0038] The integrity index is a quantitative standard for evaluating the continuity and spatial coverage of the data time series, obtained by statistically analyzing the proportion of missing sampling points of sensor data or the proportion of invalid pixels of satellite images. The integrity index includes time series completeness and spatial coverage. The time series consistency index is an evaluation parameter for verifying the time dimension synchronization of multi-source data, obtained by comparing the time difference between satellite transit and buoy observation or detecting the trend matching degree of parameter time series. The time series consistency index includes time synchronization deviation threshold and trend correlation coefficient. The spatial correlation index is an evaluation parameter for measuring the rationality of data spatial distribution, obtained by verifying local consistency through spatial interpolation or detecting whether the data spatial gradient conforms to the physical law. The spatial correlation index includes spatial interpolation error threshold and gradient anomaly rate. The dynamic threshold is a dynamic adjustment criterion based on historical data statistical characteristics. The dynamic threshold includes a sudden jump value judgment threshold based on the historical mean value ± 3σ or an offset judgment threshold based on the average value of the spatial neighborhood.

[0039] The automatic scanning verification scans missing values, duplicate records or timestamp abnormalities in the data according to preset rules. This process includes using regular expressions to detect null fields in CSV files or statistically analyzing the proportion of invalid pixels in satellite images to generate an integrity verification list. Deep verification detects parameter mutations through sliding window statistics or identifies isolated abnormal points through spatial neighborhood analysis. For example, the time dimension aligns multi-source time series using dynamic time warping algorithm, and the spatial dimension detects data distribution anomalies using K- nearest neighbor algorithm, thereby generating a time and space anomaly detail. Percent quantitative scoring is achieved by weighted calculation of integrity, time series consistency and spatial correlation. For example, integrity accounts for 40% of the weight, time series consistency accounts for 30%, and spatial correlation accounts for 30%, thereby generating a data quality score. Abnormal source tracing is achieved by correlation analysis to determine the root cause of the anomaly. For example, if satellite and buoy data are inconsistent, the root cause is sensor calibration deviation, and if multi-sensor data is mutated, the root cause is communication network interruption. Repair scheme generation is achieved through automated repair scripts or manual review processes. Missing values are linearly interpolated or Kalman filtered, sudden jumps are filtered using a sliding window median filter, and coordinate offsets are projected and converted using GIS tools.

[0040] The application quantifies data quality characteristics by constructing a dynamic quality evaluation system including integrity indicators, timing consistency indicators and spatial correlation indicators, performs automated scanning verification and deep verification to identify specific abnormal types and positions, generates data quality scores using percentage-based weighted calculation and triggers repair processes, forms a closed-loop management by combining abnormal traceability and repair schemes, and finally integrates key information through structured reports, which can improve the automation level and repair efficiency of multi-source data quality control, solve the problems of low efficiency of traditional quality control relying on manual inspection and unsystematic repair strategies, for example, through dynamic threshold determination and traceability analysis, the temperature abnormality caused by the drift of the buoy sensor can be quickly located, and the calibration process is automatically triggered to avoid ecological abnormal misjudgment.

[0041] In an embodiment of the application, the original data layer stores multi-source dynamic data that has not been processed and meets the preset quality threshold using a distributed file system, the preprocessing layer performs standardized processing on the multi-source dynamic data that meets the preset quality threshold, and is used to store the multi-source dynamic data after standardized processing, the standardized processing includes data format conversion, unit unification and coordinate calibration, the feature extraction layer uses principal component analysis to reduce the dimensionality of the data after standardized processing, and is used to store the feature vectors after dimensionality reduction, and the spatiotemporal indexing mechanism uses a space and time hybrid index.

[0042] The distributed file system is a storage architecture with high fault tolerance and scalability, such as HDFS, Ceph or cloud storage service AWSS3, which realizes data redundancy protection through data sharding storage and replication mechanism. For example, the system can store NetCDF files of satellite images, raw binary data streams of buoys and other unprocessed raw data to ensure data integrity and traceability. The multi-source dynamic data that meets the preset quality threshold is obtained through a data quality verification process, such as abnormal value detection rules defined by statistical thresholds or domain knowledge.

[0043] The preprocessing layer performs standardized processing on the multi-source dynamic data that meets the preset quality threshold, and the standardized processing includes data format conversion, unit unification and coordinate calibration. Data format conversion is to convert heterogeneous data formats into standardized formats, such as using ETL tool ApacheNiFi to realize data parsing and format conversion. Unit unification is to standardize physical quantity units to a preset reference unit, such as converting temperature units to Celsius and depth units to meters, which is realized through predefined conversion rules. Coordinate calibration is to convert data in different coordinate systems to a unified reference coordinate system, such as using GDAL library to perform coordinate system projection conversion. The standardized processing eliminates differences in format, unit and spatial reference, providing a unified data interface for subsequent analysis.

[0044] The feature extraction layer adopts principal component analysis method to perform feature dimension reduction on the standardized data, and is used for storing the dimension-reduced feature vectors. The principal component analysis method is to calculate the data covariance matrix and perform eigenvalue decomposition, and select the first k principal components with cumulative variance contribution rate exceeding 95% as the dimension-reduced feature vectors. For example, the method can convert high-dimensional data containing temperature, salinity and other parameters into low-dimensional spatial coordinate representation, for example, reducing 10-dimensional parameters to 3-dimensional principal components. This process reduces the computational complexity by eliminating redundant information while retaining the core features of the data.

[0045] The spatiotemporal index mechanism adopts a hybrid index of space and time. The spatial index is based on a tree structure to organize geographic coordinate data, supporting fast range queries; the time index is based on a B+ tree or other time series database index structure to speed up time series data retrieval. The hybrid index realizes joint query optimization of space and time conditions through composite keys or hierarchical structures, such as the spatial-temporal extension of PostGIS or the composite index of InfluxDB to quickly locate data blocks and avoid full table scanning.

[0046] The present application ensures the complete storage of raw data by using a distributed file system, eliminates the format, unit and spatial reference differences of multi-source data through standardization processing, reduces the data dimension through principal component analysis to improve the computing efficiency, and optimizes the multi-dimensional data retrieval speed through the spatiotemporal hybrid index. The technical effects of systematically solving the core challenges of multi-source data management are achieved, which are specifically manifested as: the raw data layer guarantees data traceability, the preprocessing layer provides a standardized interface, the feature extraction layer improves model training efficiency, and the spatiotemporal index mechanism supports efficient data query, thereby providing full-process optimized data support for ecological anomaly detection, such as quickly retrieving multi-source data in a specific spatiotemporal range and generating a comprehensive feature vector to realize pollution source positioning and diffusion prediction in oil spill event analysis.

[0047] In one embodiment of the present application, the physical and biological coupled numerical model includes a physical model layer and a biological model layer, which cooperate with each other through a coupling mechanism. The specific steps of generating a three-dimensional dynamic flow field and an ecological parameter distribution in the target sea area are as follows: Based on the multi-source dynamic data in the preprocessing layer and the target sea area, a physical model is established, a biological model is integrated based on the physical model, and the physical model and the biological model are coupled through a coupling mechanism to generate a physical and biological coupled numerical model; The physical model layer is a numerical simulation system based on fluid mechanics equations, used to represent the three-dimensional flow field, temperature and salinity distribution characteristics of the target sea area. The core parameters include sea surface wind stress and seabed topography friction coefficient, which can be discretized by finite volume method or finite element method and divided into calculation grid. The biological model layer is an ecological dynamics model, used to simulate biological and geochemical processes. The input parameters include environmental parameters such as light and temperature provided by the physical model, and the output results cover ecological parameters such as chlorophyll concentration and nutrient salt content. The coupling mechanism is a two-way interaction mode realized through a data interface, for example, the flow field of the physical model drives the migration of plankton in the biological model, and the metabolic products of the biological model are fed back to the chemical parameters of the physical model. In a specific embodiment, the coupling mechanism realizes iterative calculation through middleware or joint solver, for example, the physical field data is updated at each time step, then transferred to the biological model to calculate ecological parameters, and the biological feedback parameters are returned to the physical model.

[0048] The parameters of the physical model and the biological model are optimized by the particle swarm algorithm, and the physical and biological coupled numerical model is updated. The particle swarm algorithm is a swarm intelligence algorithm for multi-parameter optimization, and the parameter space definition includes the physical model parameters and the biological model parameters to be optimized. The optimization is realized by initializing the parameter particle swarm and iteratively adjusting the parameter combination. In a specific embodiment, the particle swarm algorithm gradually converges to the global optimal solution through the velocity update rule, and finally the optimized parameters are re-injected into the coupled model to form an updated model version, thereby reducing the subjectivity of parameter setting and improving the fitting degree of the model output and the measured data.

[0049] Based on the updated physical and biological coupled numerical model, the multi-source dynamic data in the preprocessing layer is input, and the three-dimensional dynamic flow field and ecological parameter distribution of the target sea area are output.

[0050] The three-dimensional dynamic flow field is a time and space continuous data set containing different depth flow velocity, flow direction and eddy energy parameters, and its generation depends on the model processing of real-time wind field, initial conditions and other standardized input data. The ecological parameter distribution is the simulation result of the spatial gradient of chlorophyll concentration and the ecological characteristics of nutrient salt depletion area, and its calculation process combines the environmental parameters provided by the physical model and the metabolic process equation of the biological model. In a specific embodiment, the model generates three-dimensional spatiotemporal distribution results through time step advancement calculation, for example, accurately simulates the distribution of red tide in the vertical water layer and the diffusion path with ocean current, and provides a high-resolution reference benchmark for anomaly detection.

[0051] The application establishes a coupled numerical model comprising a physical model layer and a biological model layer, realizes dynamic correlation of physical and biological processes by using a bidirectional data interaction mechanism, optimizes parameters by using a particle swarm algorithm to improve fitting accuracy of the model and measured data, and finally generates three-dimensional flow field and ecological parameter distribution results based on multi-source dynamic data, so that the technical effects of significantly improving simulation accuracy and dynamic adaptability of ecological parameters are achieved. The scheme solves the limitations of traditional one-way driven models by using a hierarchical coupled model, and reduces parameter setting deviation by combining an intelligent optimization algorithm, thereby providing a more reliable model basis for marine ecological anomaly detection.

[0052] In an embodiment of the application, the spatio-temporal anomaly features are extracted by using a multi-scale sliding window algorithm, and the specific process is as follows: Based on the dynamic characteristics of different parameters in multi-source dynamic data, different time windows and space windows are designed. The different time windows are monitoring intervals with different time spans set according to the dynamic characteristics of the parameters. For example, chlorophyll concentration shows hour-level mutation characteristics due to red tide outbreak, and its time window can be a short-time monitoring interval of 1 hour. Water temperature changes slowly, and its time window can be a long-time monitoring interval of 24 hours. The different space windows are monitoring areas with different spatial resolutions set according to the spatial distribution characteristics of the parameters. For example, a fine spatial grid of 500m x 500m can be used for local areas affected by coastal pollution diffusion, and a wide spatial grid of 10km x 10km can be used for nutrient salt distribution affected by ocean currents. This differential design avoids the problems of insufficient sensitivity or computational redundancy caused by traditional fixed windows by matching the time-space variation law of the parameters.

[0053] Based on the parameters in the time window and the space window, the time window statistics and the space window statistics are calculated. The time window statistics are quantitative indicators representing the time sequence characteristics of the parameters, such as mean, standard deviation, trend slope or mutation detection indicators of the parameter sequence. The time window statistics are obtained by mathematical operation on the parameter values in the window, for example, the mean of chlorophyll concentration in a certain area increases by 3 times within 1 hour and the standard deviation significantly expands. The space window statistics are quantitative indicators representing the spatial distribution characteristics of the parameters, such as spatial gradient, local mean or spatial autocorrelation coefficient. The space window statistics are obtained by spatial analysis on the spatial distribution in the window, for example, the chlorophyll concentration of a certain grid is significantly higher than that of the surrounding area and the spatial gradient exceeds the threshold. These statistics quantify the abnormal degree of the parameters in the time-space dimension, providing calculable numerical basis for subsequent judgment.

[0054] Different anomaly judgment thresholds are set according to the characteristic differences of different parameters, and the spatio-temporal anomaly features are selected from the time window statistics and the space window statistics by using the anomaly judgment thresholds.

[0055] The abnormality determination threshold is a dynamic or fixed numerical boundary set according to the parameter characteristics, for example, a fixed threshold of 3σ deviating from the historical mean for temperature anomaly, and a dynamic threshold based on real-time environmental conditions (such as the product of light intensity and nutrient salt concentration) for chlorophyll concentration. The multi-parameter adaptability threshold setting includes but is not limited to the biological oxygen demand model threshold of dissolved oxygen or the historical distribution percentile threshold of salinity. The screening process is realized by comparing the time window statistics and the spatial window statistics with the corresponding threshold, for example, when the chlorophyll concentration in a certain area suddenly increases in the time window and is significantly higher than the neighborhood in the spatial window, the statistics meet the time mutation and spatial isolation conditions at the same time, and it is marked as a spatio-temporal anomaly feature. This combined determination mechanism improves the accuracy of anomaly positioning by excluding false positives of single time series or spatial anomalies.

[0056] The present application designs the spatio-temporal variation of matching parameters by differentiating time window and spatial window, combines the quantitative analysis of time window statistics and spatial window statistics, and uses a dynamic threshold screening mechanism based on parameter characteristics. The following technical effects can be achieved: multi-parameter adaptability enhancement enables specific monitoring of rapidly changing parameters (such as algal biomass) and slowly changing parameters (such as seabed sediments); the joint determination of spatio-temporal statistics reduces false positives caused by sensor noise or local topography; and the differentiated window design optimizes the allocation of computing resources, using high-resolution windows only for key parameters, thereby improving the accuracy of anomaly positioning while reducing computational redundancy. For example, in red tide detection, the system captures the rapid aggregation of chlorophyll concentration through a 1-hour x 500-meter window, and combines a spatial gradient threshold to exclude naturally aggregated local high values, significantly improving the accuracy of anomaly region positioning.

[0057] In one embodiment of the present application, the specific steps for determining the candidate abnormal region are as follows: Based on the spatial correlation characteristics of marine ecological anomalies, an improved isolation algorithm model is constructed; Input the spatio-temporal anomaly feature into the isolation algorithm model, output the anomaly score and screen out the candidate abnormal region; Verify the authenticity of the candidate abnormal region through three types of data, and divide the candidate abnormal region that passes the verification into abnormal levels.

[0058] The spatial correlation feature is a parameter set describing the distribution rule of marine ecological anomaly, such as spatial coordinates, neighborhood anomaly density and multi-scale sliding window statistics, which is obtained by fusing the time and space data of the marine monitoring grid. The spatial coordinates include latitude and longitude position information, and the neighborhood anomaly density is the weighted average of the anomaly scores of a grid and its adjacent grids. The improved isolated algorithm model is realized by enhancing the spatial features, self-adaptive threshold adjustment and multi-scale integration, such as introducing the spatial coordinate parameter to suppress the misjudgment of isolated noise points, dynamically adjusting the segmentation threshold according to the parameter type, or fusing the weighted average or voting of the statistical results of the hour-level and day-level windows.

[0059] The time and space anomaly feature is a quantitative index reflecting the mutation or abnormal distribution of marine ecological parameters, such as time window mean mutation rate and spatial window gradient value, which is extracted by multi-scale sliding window and combined into a feature vector input model. The anomaly score is the path length quantization value of the data point in the improved isolated algorithm model, and the score of the chlorophyll concentration in a certain area will be significantly higher than the set threshold when the chlorophyll concentration in a certain area suddenly increases in the time window and the spatial neighborhood consistency is high. The process of screening candidate abnormal regions is realized by dynamic threshold, forming a preliminary abnormal region set.

[0060] The verification data includes measured data, historical data and model prediction data. The measured data is the real-time observation value of the sensor, which is used to exclude false anomalies caused by sensor failure; the historical data is the frequency record of past abnormal events in the candidate region, which is used to identify false positives caused by natural fluctuations; the model prediction data is the parameter distribution prediction result of the physical and biological coupled model, which is used to verify the authenticity. The abnormal level division includes anomaly score, influence range and parameter deviation degree, for example, a first-level anomaly can be defined as a high-risk red tide, a second-level as a moderate pollution diffusion, and a third-level as a local parameter anomaly. The dynamic adjustment mechanism combines with the meteorological data (such as wind speed, ocean current direction) to predict the diffusion trend, for example, a second-level region may be upgraded to a first-level due to strong wind.

[0061] The present application can achieve the technical effects of significantly reducing the false positive rate and improving the efficiency of emergency response by optimizing the algorithm and cooperating with multi-dimensional verification. For example, in oil spill detection, the system can accurately identify the abnormal concentration of oil pollution in a continuous area, divide the level after verification with measured data, and predict the diffusion path according to the ocean current model to provide priority basis for oil cleaning action, which realizes more efficient resource allocation compared with traditional methods.

[0062] In an embodiment of the present application, the specific steps of generating the marine ecological comprehensive anomaly index are as follows: determining a corresponding parameter based on the spatio-temporal anomaly feature of the candidate abnormal region, and quantifying an abnormal value of each parameter; calculating an abnormal contribution degree of each parameter based on the abnormal value of each parameter and a preset influence weight; obtaining a dynamically adjusted parameter weight through a preset dynamic weight calculation model, and obtaining a marine ecological comprehensive anomaly index through weighted summation based on the parameter weight and the abnormal contribution degree of each parameter.

[0063] The candidate abnormal region refers to a geographical range that may have an ecological anomaly identified through monitoring or analysis. The spatio-temporal anomaly feature thereof includes statistical or physical characteristics of the parameter deviating from the normal range in a specific time range. The parameter is a physical, chemical or biological indicator reflecting the ecological state, such as chlorophyll concentration, dissolved oxygen, oil film coverage area, etc. These parameters are obtained through environmental sensors, satellite remote sensing or laboratory analysis. Quantifying the abnormal value refers to converting the parameter abnormality degree into a numerical indicator, which can be realized through statistical methods, for example. The parameter selection needs to be combined with the spatio-temporal characteristics of the candidate abnormal region, such as selecting chlorophyll concentration and dissolved oxygen for red tide area, and selecting oil film coverage area and pH value for oil spill area.

[0064] The influence weight is a quantitative indicator of the degree of harm of the parameter to the ecological system, such as a chlorophyll concentration weight of 0.3, a dissolved oxygen weight of 0.25, and an oil pollution coverage area weight of 0.4. The abnormal contribution degree is a comprehensive reflection of the parameter abnormality degree and the weight, and its calculation method is to multiply the abnormal value by the static influence weight. The setting of the static influence weight can be determined through domain expert knowledge or historical data correlation analysis, such as quantifying the correlation between the parameter and the ecological harm through random forest feature importance score.

[0065] The dynamic weight calculation model is an algorithm for adjusting the static weight according to real-time environmental conditions or abnormal features. Its technical operation includes: inputting the current environmental parameter or abnormal feature, generating an adjustment coefficient through a preset adjustment rule, and finally calculating the dynamic weight through a weighted function. For example, if strong wind is detected to accelerate the spread of pollution, the weight of "oil pollution coverage area" is dynamically increased to 0.5; if chlorophyll anomaly is accompanied by a sharp drop in dissolved oxygen, the weights of both are simultaneously increased to reflect the synergistic harm. The product of the dynamic weight and the abnormal contribution degree is summed up to generate a comprehensive anomaly index, which can be normalized to represent the abnormal severity, such as 80-100 points for high-risk anomaly, which needs to be responded immediately.

[0066] The application quantifies parameter outliers to adapt to different types of parameter selection, combines static weight and dynamic weight adjustment mechanism to reflect the influence of environmental conditions and parameter coupling relationship, and finally generates a standardized comprehensive anomaly index through weighted summation, which can improve the spatio-temporal adaptability of ecological anomaly early warning, enhance the representation ability of multi-parameter compound stress, and provide scientific basis for hierarchical response. The scheme solves the problem of rigid early warning of traditional fixed weight method, for example, during the typhoon, the dynamic weight can quantify the promoting effect of wind speed on pollutant diffusion, or when red tide and oil spill coexist, the weight of related parameters is simultaneously increased, thereby triggering joint emergency measures.

[0067] In an embodiment of the application, the specific steps of obtaining the prediction result and generating the early warning measures are as follows: Extract historical data from the marine ecological dynamic database as training samples, and construct an anomaly trend prediction model through the training samples; The historical data is a set of marine ecological comprehensive anomaly index historical sequence and related environmental parameters containing time stamps, which can be obtained through marine monitoring equipment, satellite remote sensing or numerical model output, etc. For example, the historical data can include meteorological data, flow field changes predicted by physical models, etc. The training sample is a preprocessed structured data set, for example, the preprocessing process includes normalization, missing value filling and sliding window segmentation, for example, 72 hours of continuous data are taken as an input window to predict the future 24 hours. The anomaly trend prediction model can be a prediction framework based on time series prediction model or machine learning regression model, for example, it can include one or more of LSTM neural network, ARIMA model or random forest. The construction process can train the model by dividing the training set and the validation set, for example, the LSTM model captures the time dependence through memory cells, and the ARIMA model fits the trend through difference and autoregressive terms. Model hyperparameters (such as the number of neural network layers, time step) are optimized through cross-validation.

[0068] Input the marine ecological comprehensive anomaly index into the anomaly trend prediction model, output a comprehensive anomaly index prediction sequence containing a period of time in the future, and generate a prediction result containing early warning information based on the comprehensive anomaly index prediction sequence and the preset hierarchical early warning threshold; The comprehensive anomaly index is a numerical index quantifying the degree of marine ecological anomaly, which can be generated by a multi-source data fusion algorithm. The graded warning threshold is a preset index threshold interval, which may include, for example, a red warning (index >= 85), an orange warning (70 <= index < 85), a yellow warning (50 <= index < 70), and the like. The warning information can be structured data containing a time range, an affected area, and a level, which may, for example, be labeled with specific latitude and longitude ranges. The generation process can be achieved by threshold comparison and spatio-temporal association, for example, by comparing the index at each time point in the prediction sequence with the threshold, marking the triggered warning level, and combining the spatial position of the candidate anomaly area with the prediction index change trend to generate the warning information.

[0069] Based on the warning information in the prediction result, the corresponding warning measures are retrieved from the preset warning database.

[0070] The warning database is a structured knowledge base that stores preset response measure templates, and its structure may, for example, include hierarchical association, scenario adaptation, and dynamic expansion functions. The hierarchical association can be a mapping relationship between warning levels and basic measures, for example, a red warning triggers the closure of a shipping lane, and an orange warning triggers the launch of a drone patrol. The scenario adaptation can be the addition of special measures according to the type of anomaly, for example, the addition of biological remediation agents for red tide anomalies and the deployment of oil containment booms for oil spill events. The retrieval process can be implemented by a rule engine, for example, by condition matching to retrieve matching measure entries in the database, prioritizing by urgency, or combining similar measures, and finally converting the measure list into a structured report (such as JSON or HTML) containing execution steps, responsible departments, and time requirements.

[0071] The present application also provides a marine ecological anomaly detection system based on multi-source data, comprising a processor and a memory for storing instructions executable by the processor; The processor can be a hardware unit that performs algorithmic operations, which may, for example, use a high-performance CPU or GPU cluster. GPU parallel computing can shorten the prediction delay of the LSTM model. The memory can be a storage carrier for data and instructions, which may, for example, use a distributed storage system such as an object storage and database cluster to manage raw data, preprocessed data, and model parameters, ensuring low latency and high throughput in high-concurrency access.

[0072] The application can achieve the technical effects of significantly improving prediction accuracy, optimizing resource allocation efficiency, shortening response time difference, and supporting mass data processing and multi-model parallelism by using a history data driven prediction model combined with multi-parameter input to reduce hysteresis, using hierarchical early warning thresholds and space-time correlation to generate structured early warning information, using a rule engine to realize automatic retrieval and optimization of early warning measures, and relying on a high-performance processor and a distributed storage system to ensure real-time performance and scalability.

[0073] The above only describes the preferred embodiments of the present application, and it should be noted that those skilled in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements should also be considered within the scope of protection of the present application.

Claims

1. A method for detecting marine ecological anomalies based on multi-source data, characterized in that, Includes the following steps: Multi-source dynamic data of the target sea area is obtained through a collaborative architecture of distributed crawlers and sensor gateways. A dynamic quality assessment system is built based on data attributes. It takes multi-source dynamic data as input, outputs quality control reports, and outputs multi-source dynamic data that meets preset quality thresholds based on the quality control reports. Based on multi-source dynamic data that meets preset quality thresholds, a marine ecological dynamic database is established in conjunction with a spatiotemporal indexing mechanism. The marine ecological dynamic database includes a raw data layer, a preprocessing layer, and a feature extraction layer. Based on the data stored in the preprocessing layer, a three-dimensional dynamic flow field and ecological parameter distribution of the target sea area are generated through a preset physical and biological coupling numerical model and stored in the marine ecological dynamic database. Based on the marine ecological dynamics database, spatiotemporal anomaly features were extracted and candidate anomaly regions were identified. Based on the anomaly contribution of parameters corresponding to spatiotemporal anomalies within the candidate anomaly region, a dynamic weight calculation model is used to generate a comprehensive marine ecological anomaly index. Based on the comprehensive marine ecological anomaly index, a pre-set anomaly trend prediction model is used to output prediction results and generate early warning measures.

2. The marine ecological anomaly detection method based on multi-source data according to claim 1, characterized in that, The multi-source dynamic data includes remote sensing satellite data, buoy monitoring data, meteorological station observation data, underwater sensor array data, and historical reanalysis data. The specific steps for acquiring the multi-source dynamic data include: By deploying distributed crawler nodes and employing asynchronous I / O mechanisms and multi-protocol adaptation technology, macroscopic data of the target sea area is captured periodically. Build a sensor gateway cluster based on a low-power wide-area network communication protocol to collect real-time near-shore data; The spatiotemporal calibration algorithm is used to process macroscopic data and nearshore real-time data for consistency, forming unified multi-source dynamic data.

3. The marine ecological anomaly detection method based on multi-source data according to claim 1, characterized in that, The quality control report includes a data quality score, anomaly details, and remediation plans. The specific steps to obtain the quality control report are as follows: A dynamic quality assessment system is constructed based on the attribute differences of multi-source dynamic data, which includes three-level evaluation indicators and dynamic thresholds. The three-level evaluation indicators include integrity indicators, temporal consistency indicators, and spatial correlation indicators. Based on integrity indicators, automated scanning and verification are performed on multi-source dynamic data to generate an integrity verification list. Based on temporal consistency and spatial correlation indicators, deep verification of temporal and spatial dimensions is performed on multi-source dynamic data to generate detailed information on temporal and spatial anomalies. Based on the integrity checklist and the details of temporal and spatial anomalies, the integrity, temporal consistency and spatial correlation of multi-source dynamic data are quantified using a percentage system. Data quality scores and anomaly details are generated. Anomaly tracing is performed based on the anomaly details and a repair plan is generated. A quality control report is generated by combining the data quality scores, anomaly details and repair plans.

4. The marine ecological anomaly detection method based on multi-source data according to claim 1, characterized in that, The raw data layer uses a distributed file system to store unprocessed multi-source dynamic data that meets a preset quality threshold. The preprocessing layer performs standardization processing on the multi-source dynamic data that meets the preset quality threshold, and stores the standardized multi-source dynamic data. The standardization processing includes data format conversion, unit unification, and coordinate calibration. The feature extraction layer uses principal component analysis to perform feature dimensionality reduction on the standardized data, and stores the dimensionality-reduced feature vectors. The spatiotemporal indexing mechanism uses a hybrid spatial and temporal index.

5. The marine ecological anomaly detection method based on multi-source data according to claim 4, characterized in that, The physical-biological coupled numerical model includes a physical model layer and a biological model layer, which cooperate with each other through a coupling mechanism. The specific steps for generating the three-dimensional dynamic flow field and ecological parameter distribution of the target sea area are as follows: Based on the multi-source dynamic data and target sea area in the preprocessing layer, a physical model is established, and a biological model is integrated on the basis of the physical model. The physical model and the biological model are coupled through a coupling mechanism to generate a physical-biological coupled numerical model. The parameters of the physical and biological models are optimized separately using the particle swarm optimization algorithm, and the physical-biological coupled numerical model is updated. Based on the updated physical and biological coupled numerical model, the multi-source dynamic data in the preprocessing layer are input, and the three-dimensional dynamic flow field and ecological parameter distribution of the target sea area are output.

6. The marine ecological anomaly detection method based on multi-source data according to claim 1, characterized in that, The spatiotemporal anomaly features were extracted using a multi-scale sliding window algorithm, as detailed below: Based on the dynamic characteristics of different parameters in multi-source dynamic data, differentiated time windows and spatial windows are designed. Calculate time window statistics and spatial window statistics based on parameters within the time window and spatial window; Different anomaly detection thresholds are set based on the characteristic differences of different parameters, and spatiotemporal anomaly features are screened out from time window statistics and spatial window statistics using the anomaly detection thresholds.

7. The marine ecological anomaly detection method based on multi-source data according to claim 1, characterized in that, The specific steps for determining the candidate anomaly region are as follows: Based on the spatial correlation characteristics of marine ecological anomalies, an improved isolation algorithm model is constructed; The spatiotemporal anomaly features are input into the isolated algorithm model, and the anomaly score is output and candidate anomaly regions are selected. The authenticity of candidate anomaly regions is verified using three types of data, and the anomaly levels of the verified candidate anomaly regions are classified.

8. The marine ecological anomaly detection method based on multi-source data according to claim 1, characterized in that, The specific steps for generating the comprehensive marine ecological anomaly index are as follows: Based on the spatiotemporal anomaly characteristics of candidate anomaly regions, the corresponding parameters are determined, and the anomaly value of each parameter is quantified. Based on the outlier values ​​of each parameter and the preset impact weight, calculate the outlier contribution of each parameter; The dynamically adjusted parameter weights are obtained through a preset dynamic weight calculation model. Based on the parameter weights and the abnormal contribution of each parameter, the comprehensive marine ecological anomaly index is obtained by weighted summation.

9. The marine ecological anomaly detection method based on multi-source data according to claim 1, characterized in that, The specific steps for obtaining the prediction results and generating early warning measures are as follows: Historical data is extracted from the marine ecological dynamics database as training samples, and an anomaly trend prediction model is constructed using the training samples. The marine ecological comprehensive anomaly index is input into the anomaly trend prediction model, and the output is a comprehensive anomaly index prediction sequence containing a future period. Based on the comprehensive anomaly index prediction sequence and the preset graded early warning threshold, a prediction result containing early warning information is generated. Based on the early warning information in the prediction results, the corresponding early warning measures are retrieved from the preset early warning database.

10. A marine ecological anomaly detection system based on multi-source data, characterized in that, Includes a processor and a memory for storing processor-executable instructions; The processor is configured to execute the instructions to implement the steps of the marine ecological anomaly detection method based on multi-source data as described in any one of claims 1 to 9.

Images