A method for detecting anomaly points in time-series marine ecological environment early warning monitoring data

Abstract

The present invention discloses a method for detecting anomalies in time-series marine ecological environment early warning monitoring data, which relates to the technical field of marine ecological environment analysis, including collecting multiple types of marine monitoring indicators, constructing a multi-source and multi-dimensional time-series data set, and unifying it into a standard input format through preprocessing. Based on the causal relationship and co-evolution characteristics between factors, a graph structure model is constructed to characterize the coupled propagation paths between factors. Subsequently, a sliding window is used to generate a local time series model, and a counterfactual path is constructed to compare the propagation differences of the original path in the graph structure to identify key disturbance moments. Combined with the similarity of historical events and spatial consistency information, the warning credibility of the anomaly is evaluated to achieve efficient and explainable hierarchical warning response. The present invention realizes accurate identification of anomalies and credible warning, has causal modeling, counterfactual reasoning and hierarchical response capabilities, and improves the intelligence and interpretability of marine ecological monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of marine ecological environment analysis, and in particular to a method for detecting abnormal points in time-series marine ecological environment early warning monitoring data. Background Art

[0002] As changes in the marine ecological environment become increasingly complex, monitoring and intelligent early warning based on multi-source data have become important research directions in marine ecological management. In recent years, thanks to the development of sensor networks, remote sensing observations, and marine Internet of Things technologies, marine monitoring systems can achieve high-frequency and continuous collection of physical, chemical, and ecological factors, forming a multidimensional, time-series data structure. In data-driven ecological early warning research, the academic community has gradually explored the use of statistical modeling, machine learning, and deep learning methods to identify key change points, assisting in the early identification and risk warning of events such as red tides, hypoxic areas, and abnormal pollution discharges. However, most current methods are still based on single-factor abnormal fluctuation discrimination or black-box model output, and have problems such as insufficient causal structure modeling and limited system interpretability.

[0003] Although some studies have attempted to introduce a joint modeling mechanism for multi-dimensional indicators, such as using coordinated change trends, feature fusion algorithms or time series clustering methods for outlier detection, existing methods still have obvious limitations in the following key aspects. First, most methods ignore the complex causal relationships and feedback mechanisms between marine ecological factors, and are unable to characterize the dynamic impact path of changes in a certain monitoring factor on the overall ecological structure of the system, resulting in anomaly identification biased towards local fluctuation judgments and a lack of system evolution logic. Secondly, most existing models use one-way time series modeling or fixed threshold judgment, lack a "counterfactual reasoning" mechanism, and cannot effectively measure whether a data point is a key source of disturbance in the evolution of the system, limiting the robustness and interpretability of the identification results. In addition, traditional methods find it difficult to integrate the knowledge structure of historical ecological events with spatially consistent distribution information, resulting in weak model generalization ability and high false alarm rate, making it difficult to support the needs of large-scale monitoring of multiple parameters in the ocean in real scenarios. In contrast, our invention introduces feedback propagation and graph attention mechanisms based on graph structure modeling, integrating causal modeling, counterfactual generation, historical similarity and spatial consistency into a four-dimensional coupling mechanism, effectively solving the bottlenecks of low accuracy in outlier identification, poor interpretability and unclear graded responses in existing technologies, and realizing a marine ecological early warning method with structural intelligence and evolutionary prediction capabilities. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: the existing marine ecological anomaly detection methods lack a causal feedback modeling mechanism, are unable to perform counterfactual disturbance analysis, and the warning output lacks credibility interpretation and hierarchical stratification, as well as how to achieve structured, interpretable, credibility-driven automatic identification and warning response of marine ecological anomalies.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for detecting anomalies in time-series marine ecological environment early warning monitoring data, comprising continuously collecting marine ecological environment background interference factors to form a multi-source and multi-dimensional time-series monitoring data set for marine ecological environment, preprocessing the data set and unifying it into a standard input format; based on the causal relationship and co-evolution characteristics between monitoring factors in the multi-source and multi-dimensional time-series monitoring data set for marine ecological environment, each monitoring factor is used as a node, and the feedback path between monitoring factors is used as an edge. The edge weight is comprehensively determined by combining temporal causality, mutual information strength, feedback propagation and graph attention mechanism, and an EFMG graph is constructed by characterizing the coupled propagation relationship between factors; by comparing and analyzing the propagation path and structural influence of the EFMG graph, a sliding window local time series model is constructed, and a counterfactual evolution path after exclusion is generated for each time point. The feedback propagation difference between the original path and the counterfactual path in the EFMG graph structure is compared, the disturbance moment is identified through the counterfactual evolution path, and the warning credibility is evaluated by combining historical event similarity and spatial consistency information, and a graded warning response is performed.

[0007] As a preferred solution of the method for detecting anomaly points in the time-series marine ecological environment early warning monitoring data described in the present invention, the preprocessing of the data set includes using the K-nearest neighbor algorithm or the Bayesian inference method to fill in missing values ​​of the monitoring factors when the time points are missing, using the Z score or the interquartile range method to screen the initial anomaly points for the time series of each monitoring factor, using a unified timestamp mechanism to complete the time alignment of the monitoring factors, and using the minimum-maximum normalization or standard deviation normalization method to perform unified numerical conversion.

[0008] As a preferred solution for the method for detecting anomaly points in the temporal marine ecological environment early warning monitoring data described in the present invention, the construction of the EFMG graph includes the following steps: nodes represent various monitoring factors in the construction process, edges represent feedback relationships, and the weights of the edges are determined by the fusion of causal analysis, information correlation, time lag and graph attention mechanism. The EFMG graph dynamically changes with the monitoring time window.

[0009] As an optimal solution for the method for detecting anomaly points in the temporal marine ecological environment early warning monitoring data described in the present invention, the EFMG map is used to represent the feedback path between factors, and the indicator evolution process is modeled based on the sliding window during the construction process, and the lagged causal relationship and collaborative information strength are combined to determine whether there is a feedback connection between the monitoring factors.

[0010] As a preferred solution of the method for detecting anomaly points in the time-series marine ecological environment early warning monitoring data described in the present invention, the generation of the excluded counterfactual evolution path for each time point includes using a sliding window local time series model to infer and generate an alternative path in the case where the point did not occur under the previous and next information of the given time series window.

[0011] As a preferred solution of the method for detecting anomaly points in the temporal marine ecological environment early warning monitoring data described in the present invention, the generation of the excluded counterfactual evolution path for each time point includes comparing the propagation results of the original sequence and the counterfactual path in the EFMG spectrum feedback structure, identifying the time point, constructing a candidate anomaly set to evaluate the warning credibility, and performing a graded warning response.

[0012] As a preferred solution of the method for detecting anomaly points in the temporal marine ecological environment early warning monitoring data described in the present invention, the evaluation of the warning credibility and the graded warning response include outputting the warning credibility score of each candidate anomaly point based on four factors: abnormal disturbance score, EFMG spectrum structure change, historical event similarity and spatial consistency, and dividing the warning level into multiple levels according to the scoring results and triggering the corresponding warning response strategy.

[0013] Another object of the present invention is to provide a time-series marine ecological environment early warning monitoring data anomaly detection system, which can construct a local time series model through a sliding window and generate a counterfactual evolution path after exclusion for each time point, solving the problems of the current marine ecological anomaly detection method that is unable to perform counterfactual-level disturbance analysis and the lack of credibility explanation and hierarchical stratification in the early warning output.

[0014] As a preferred solution of the temporal marine ecological environment early warning monitoring data anomaly detection system described in the present invention, it includes: a data acquisition and standardization processing module, an ecological factor-feedback mapping graph construction module, and an anomaly identification module based on counterfactual generation and graph perturbation; the data acquisition and standardization processing module is used to build a unified data input basis; the ecological factor-feedback mapping graph construction module is used to characterize the coupling dynamic relationship between different monitoring indicators in the ecosystem; the anomaly identification based on counterfactual generation and graph perturbation is used to identify disturbance points in the monitoring data, judge the degree of impact on the ecological feedback system, and form a basis for early warning decision-making.

[0015] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a method for detecting abnormal points in time-series marine ecological environment early warning monitoring data.

[0016] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for detecting anomalies in time-series marine ecological environment early warning monitoring data.

[0017] Beneficial effects of the present invention: The method for detecting anomalies in time-series marine ecological environment early warning monitoring data provided by the present invention achieves high-precision identification and intelligent early warning of anomalies in the marine ecological environment by constructing a complete data processing and anomaly detection process. Specifically, by continuously collecting and preprocessing multiple types of monitoring data such as physical, chemical, ecological and background factors, and uniformly formatting them into high-quality standard inputs, a robust data foundation is provided for subsequent modeling. By constructing an ecological factor-feedback mapping graph (EFMG), dynamic feedback modeling based on lagged causality, mutual information and graph attention mechanisms between monitoring factors is achieved, which can reveal potential systemic evolution paths. Combining sliding windows with local modeling methods, a counterfactual sequence is generated and the impact of the missing point on the propagation of the graph structure is evaluated, thereby quantifying the critical perturbation intensity of a single point. Further, a counterfactual perturbation scoring function and a graph propagation difference function are constructed, and the similarity of historical events and spatial consistency are integrated to design a warning credibility quantification factor with explanatory and discriminative power, thereby achieving risk level determination of anomalies. Ultimately, through the credibility-driven hierarchical early warning mechanism and map-counterfactual-historical comparison report output, the system's early identification capabilities for weak ecological disturbances, the interpretability of early warning results, and the scientific nature of response strategies have been effectively improved. It is suitable for ecological anomaly discovery and risk control scenarios in complex multi-parameter environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is an overall flow chart of a method for detecting anomalies in time-series marine ecological environment early warning monitoring data provided by the first embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0021] Example 1, with reference to Figure 1 , as one embodiment of the present invention, provides a method for detecting anomalies in time-series marine ecological environment early warning monitoring data, comprising:

[0022] S1: Continuously collect background interference factors to form a multi-source and multi-dimensional time series monitoring data set, pre-process the data set, and unify it into a standard input format.

[0023] Furthermore, various monitoring indicators of the marine ecological environment are continuously collected, including but not limited to:

[0024] Physical factors: water temperature (Temp), salinity (Sal), flow rate (Current);

[0025] Chemical factors: dissolved oxygen (DO), ammonia nitrogen (NH3), pH;

[0026] Ecological factors: chlorophyll a (Chla), COD, suspended solids concentration;

[0027] Background interference factors: tidal information, wind speed and direction, ocean current distribution, etc.

[0028] It should be noted that the collected data form a multi-source and multi-dimensional time series dataset, which is standardized into a unified format after preprocessing. The processing steps include missing value interpolation (using KNN or Bayesian inference), initial abnormality screening (based on Z-score or IQR), time alignment (time stamp unification) and normalization (such as Min-Max or Z-score standardization).

[0029] S2: Based on the causal relationship and co-evolution characteristics between monitoring factors, a graph structure model is constructed, with each monitoring factor as a node and the feedback path between factors as an edge. The edge weight is comprehensively determined by combining temporal causality, mutual information strength, feedback propagation and graph attention mechanism to characterize the coupled propagation relationship between factors.

[0030] Furthermore, in order to extract the dynamic coupling relationship between indicators and establish a causal transmission path, the present invention designs an ecological factor-feedback mapping graph (EFMG) and establishes the graph structure in the following way:

[0031] The nodes are indicator variables;

[0032] Edges are feedback relationships between indicators;

[0033] The edge weight is a comprehensive feedback score obtained through time-delay causal analysis, mutual information calculation, time position correction, and graph attention mechanism optimization.

[0034] The complete edge weight scoring function is expressed as:

[0035]

[0036] in, For nodes For Node The ecological feedback strength score, is a local time window, is the maximum lag time interval, For nodes Delayed causality on nodes The exponential weight function, is the Granger coefficient smoothing parameter, Indicates the delay The Granger regression coefficient under is the time position correction function, where is the period length constant, is the joint probability density function, Node and nodes At the moment The marginal probability density of is the graph attention expansion function, is a trainable vector parameter, is the node timing mapping matrix, is the node feature representation;

[0037] represents the Granger-based lagged causal weight function, defined as:

[0038]

[0039] represents the time position correction coefficient, which is defined as:

[0040]

[0041] represents the mutual information enhancement term;

[0042] represents the graph attention expansion factor, defined as

[0043]

[0044] , representing a node right The ecological feedback impact probability score is:

[0045] when : indicates no significant effect;

[0046] when : There is a weak or moderate coupling relationship;

[0047] when : There are significant ecological feedback chain connections, which require special attention.

[0048] S3: Build a local time series model through a sliding window, generate an excluded counterfactual evolution path for each time point, and compare the feedback propagation differences between the original path and the counterfactual path in the graph structure to identify the disturbance moment. Combined with the similarity of historical events and spatial consistency information, the credibility of the warning is evaluated to achieve a graded warning response.

[0049] Furthermore, after completing the construction of the ecological factor-feedback mapping graph (EFMG), the present invention further proposes an outlier identification method that integrates local modeling, counterfactual generation and graph perturbation assessment, which is particularly suitable for capturing perturbative temporal anomalies in the ecosystem that are difficult to judge through traditional statistical indicators.

[0050] By comparing the real ecological evolution path with the counterfactual evolution path after "excluding a certain point" in different time windows, and comparing the differences in their feedback propagation paths in combination with the EFMG map structure, it is determined whether each monitoring moment is a key anomaly point.

[0051] Time series segmentation and local modeling, the original time series of each indicator is divided into segments of length Partial fragment , each segment is considered as a micro-evolutionary cycle of an ecosystem.

[0052] Construct a local state space model for each segment, such as linear fitting, autoregression (AR), or encode it into a low-dimensional dynamic state representation by a deep sequence model (such as LSTM) , as the normal evolution representation.

[0053] Counterfactual path generation, for each time segment Every moment in , construct the corresponding counterfactual time series , that is, the evolutionary path of "assuming that this point did not occur".

[0054] The sequence is generated by the Counterfactual Generator (CFG) method, and the generation mechanism is expressed as:

[0055]

[0056] in, is the trained CFG network function, The network adopts a multi-layer Transformer structure and combines the attention information of the nodes in EFMG to reasonably construct the filling points.

[0057] The counterfactual path construction is expressed as:

[0058]

[0059] In order to quantify the disturbance effect of the monitoring point on the ecosystem feedback structure, the following counterfactual difference scoring function is introduced: , which is used to measure the difference between the propagation results of the real path and the counterfactual path in the graph structure, is expressed as:

[0060]

[0061] in, For the window Middle The counterfactual perturbation score of each point, The length is A real sliding window fragment, To remove The counterfactual evolution sequence generated later, represents the ecological feedback edge weight generated by the real or counterfactual sequence, is the path response factor that reflects the combined weight of the state gradient difference and semantic distance between nodes. Representation node The feature change gradient under the current segment, is the Euclidean distance between node vectors, To prevent small constants with denominators of 0, It is a timing-sensitive adaptive anomaly threshold.

[0062] : Graph propagation score (i.e., EFMG edge weight) constructed based on real sequences;

[0063] : The graph edge weight score recalculated under the counterfactual sequence;

[0064] : The response weighting factor of the propagation path, defined as:

[0065]

[0066] Finally, if Above the adaptive threshold , then determine For strong disturbance outliers, record them in the candidate set .

[0067] , represents the disturbance intensity of a missing point on the overall ecological feedback structure; where:

[0068] when : This point does not affect the feedback propagation of the system and is a normal point;

[0069] when : It may be a slightly disturbed point and enter the observation set;

[0070] when : Considered as a strong outlier and entered into the strong candidate set , as the target of subsequent key early warning analysis.

[0071] Design a factor function for calculating the credibility of comprehensive anomalies , this function considers the following four influencing factors and uses a nonlinear fusion function The build includes:

[0072] Counterfactual perturbation scoring ;

[0073] Graph propagation disturbance score ;

[0074] Historical event matching score ;

[0075] Spatial consistency relative score ;

[0076] The formula for quantifying the warning credibility factor is:

[0077]

[0078] The counterfactual perturbation score is expressed as:

[0079]

[0080] The graph propagation perturbation score is expressed as:

[0081]

[0082] in, is the belief propagation distribution under the real graph, Propagate the results under the counterfactual path.

[0083] The historical background event matching score is expressed as:

[0084]

[0085] in, is the event feature vector of the current abnormal point, It is the central feature of historical events.

[0086] The relative score of spatial consistency is expressed as:

[0087]

[0088] in, It is an outlier In the Consistency scores on adjacent sites (probabilistic processing), is the number of adjacent sites.

[0089] The confidence margin enhancement factor is expressed as:

[0090]

[0091] in, is the cumulative value of the deviation rate under the historical abnormal confidence window (taken from the sliding window history), For the The warning credibility score of each candidate anomaly point, is the counterfactual perturbation score, is the confidence difference under graph propagation, is the cosine similarity score between the current event and the historical typical abnormal events, is the spatial consistency information entropy value, is the historical confidence trend shift enhancement factor, and The following are true and counterfactual The spectrum feedback value of is the feedback disturbance weighting factor, and are the belief propagation results under real and counterfactual conditions, respectively. and are the vector representations of current abnormal events and historical events respectively, Anomaly In the The consistency probability of adjacent points, is the historical confidence cumulative drift strength, is the adjustment constant of the similarity nonlinear enhancement factor.

[0092]

[0093] like : It is judged as a low risk warning level (Level I), indicating that there is potential local fluctuation and operation and maintenance personnel are advised to pay attention; if : Determined to be a high-risk warning level (Level II), the system automatically triggers an alarm and performs regional expansion analysis; if : Determined to be an extremely high risk level (Level III), it is a core alarm event that requires immediate response and comparison and tracing with the historical database.

[0094] This early warning layering strategy combines spectral perturbations, spatial consistency, and historical event similarity, and has good resolution and decision-making stability.

[0095] It should be noted that the basic information of outliers includes:

[0096] Abnormal indicator name (such as dissolved oxygen, chlorophyll a, etc.);

[0097] Exception occurrence timestamp ;

[0098] Monitoring station number and geographical location coordinates ;

[0099] Abnormality intensity score , credibility score .

[0100] Call the current time graph structure adjacency matrix in the Ecological Feedback Mapping Graph (EFMG) module , and extract the first-order and second-order paths around the abnormal nodes high-impact links to form a feedback chain path diagram to identify its systemic spread possibility.

[0101] Calling up the counterfactual generation path , and make a graphical comparison with the original sequence before and after the outlier point (such as error heat map, trend difference map), to show the system evolution deviation that could be avoided if the outlier point had not occurred.

[0102] The current event feature vector and the central vectors of all known ecological events in the historical label database Calculate cosine similarity , and extract the top similarity rankings The system can identify historical events (such as red tides, hypoxic areas, and sewage impacts) and output their reference impact duration, area, and response measures, thereby building a warning interpretation chain with a two-way correlation between the current and the past.

[0103] Example 2, an embodiment of the present invention, provides a time-series marine ecological environment early warning monitoring data anomaly detection system, including a data acquisition and standardization processing module, an ecological factor-feedback mapping map construction module, and an anomaly identification module based on counterfactual generation and map disturbance. The data acquisition and standardization processing module is used to build a unified data input basis, the ecological factor-feedback mapping map construction module is used to characterize the coupling dynamic relationship between different monitoring indicators in the ecosystem, and the anomaly identification based on counterfactual generation and map disturbance is used to identify disturbance points in the monitoring data, judge the degree of impact on the ecological feedback system, and form a basis for early warning decision-making.

[0104] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0105] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0106] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0107] It should be understood that various aspects of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following technologies known in the art may be used: a discrete logic circuit having logic gates for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gates, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc. It should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will understand that modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and such modifications are intended to be encompassed by the claims of the present invention.

[0108] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting abnormal points in time-series marine ecological environment early warning monitoring data, characterized in that: include: Continuously collect background interference factors of the marine ecological environment to form a multi-source and multi-dimensional time-series monitoring data set of the marine ecological environment, pre-process the data set, and unify it into a standard input format; Based on the causal relationship and co-evolution characteristics between monitoring factors in the multi-source and multi-dimensional time series monitoring dataset of the marine ecological environment, each monitoring factor is regarded as a node, and the feedback path between monitoring factors is regarded as an edge. The edge weight is determined by combining temporal causality, mutual information strength, feedback propagation and graph attention mechanism. The EFMG graph is constructed by characterizing the coupling propagation relationship between monitoring factors. Constructing EFMG atlas structure: The nodes are indicator variables; Edges are feedback relationships between indicators; The edge weight is a comprehensive feedback score obtained through time-delay causal analysis, mutual information calculation, time position correction, and graph attention mechanism optimization. The edge weight scoring function is expressed as: in, For nodes For Node The ecological feedback strength score, is a local time window, is the maximum lag time interval, For nodes Delayed causality on nodes The exponential weight function, is the Granger coefficient smoothing parameter, Indicates the delay The Granger regression coefficient under is the time position correction function, where is the period length constant, is the joint probability density function, Node and nodes At the moment The marginal probability density of is the graph attention expansion function, is a trainable vector parameter, is the node timing mapping matrix, is the node feature representation, Represented as a node The node number of an adjacent monitoring factor, node The set of neighbor nodes of represents the Granger-based lagged causal weight function, defined as: in, represents the lag order index, represents the maximum lag order; represents the time position correction coefficient, which is defined as: represents the mutual information enhancement term; represents the graph attention expansion function, which is defined as: , representing a node right The ecological feedback impact probability score; By comparing and analyzing the propagation paths and structural impacts of the EFMG atlas, a sliding window local time series model is constructed. For each time point, a counterfactual evolution path is generated after exclusion. The feedback propagation differences between the original path and the counterfactual path on the EFMG atlas structure are compared. The disturbance moment is identified through the counterfactual evolution path. Combined with the similarity of historical events and spatial consistency information, the warning credibility is evaluated and a graded warning response is carried out.

2. The method for detecting anomalies in time-series marine ecological environment early warning monitoring data according to claim 1, characterized in that: The data set preprocessing includes using the K-nearest neighbor algorithm or Bayesian inference method to fill missing values ​​when the monitoring factors are missing at a time point, using the Z score or interquartile range method to screen initial outliers for the time series of each monitoring factor, using a unified timestamp mechanism to complete the time alignment of the monitoring factors, and using minimum-maximum normalization or standard deviation normalization to perform unified numerical conversion.

3. The method for detecting anomalies in time-series marine ecological environment early warning monitoring data according to claim 2, characterized in that: The construction of the EFMG graph includes the following steps: nodes represent various monitoring factors, edges represent feedback relationships, and edge weights are determined by the fusion of causal analysis, information relevance, time lag, and graph attention mechanism. The EFMG graph dynamically updates and changes with the monitoring time window.

4. The method for detecting abnormal points in time-series marine ecological environment early warning monitoring data according to claim 3, characterized in that: The EFMG map is used to represent the feedback path between monitoring factors. During the construction process, the indicator evolution process is modeled based on a sliding window, and the lagged causal relationship and collaborative information strength are combined to determine whether there is a feedback connection between monitoring factors.

5. The method for detecting abnormal points in time-series marine ecological environment early warning monitoring data according to claim 4, characterized in that: The generation of the excluded counterfactual evolution path for each time point includes using a sliding window local time series model to infer and generate an alternative path for the case where the point did not occur under the given before and after information of the sliding window.

6. The method for detecting abnormal points in time-series marine ecological environment early warning monitoring data according to claim 5, characterized in that: The generation of the excluded counterfactual evolution path for each time point includes comparing the propagation results of the original sequence and the counterfactual path in the EFMG graph feedback structure, identifying the time point, constructing a candidate anomaly set to evaluate the warning credibility, and performing a graded warning response.

7. The method for detecting abnormal points in time-series marine ecological environment early warning monitoring data according to claim 6, characterized in that: The evaluation of warning credibility and the graded warning response include outputting a warning credibility score for each candidate anomaly point based on four factors: abnormal disturbance score, EFMG spectrum structure change, historical event similarity and spatial consistency, and dividing the warning into multiple levels according to the scoring results and triggering the corresponding warning response strategy.

8. A system using the method for detecting abnormal points in time-series marine ecological environment early warning monitoring data according to any one of claims 1 to 7, characterized in that: It includes a data collection and standardization processing module, an ecological factor-feedback mapping map construction module, and an anomaly identification module based on counterfactual generation and map perturbation; The data acquisition and standardization processing module is used to build a unified data input basis; The ecological factor-feedback mapping diagram construction module is used to depict the coupling dynamic relationship between different monitoring indicators in the ecosystem; The anomaly recognition based on counterfactual generation and graph disturbance is used to identify disturbance points in monitoring data, determine the degree of impact on the ecological feedback system, and form a basis for early warning decision-making.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for detecting anomaly points in time-series marine ecological environment early warning monitoring data described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting anomaly points in temporal marine ecological environment early warning monitoring data described in any one of claims 1 to 7 are implemented.

Images