Microplastic ecological risk early warning method and system based on knowledge graph and machine learning

Abstract

The present application provides a microplastic ecological risk early warning method and system based on a knowledge graph and machine learning, the microplastic ecological risk early warning method comprising: collecting microplastic ecological related data under different data sources, and preprocessing the microplastic ecological related data to form a standardized data set with a unified space-time framework and a correlation relationship; using the standardized data set with a unified space-time framework and a correlation relationship to construct a microplastic ecological risk knowledge graph with microplastics, marine organisms, exposure pathways and toxicity effects as core entities, and associated space-time dynamic data; generating a microplastic concentration mechanism early warning threshold according to the input microplastic concentration data of the target sea area and the target biological species, and using the trained machine learning early warning model to obtain the microplastic concentration prediction value corresponding to the microplastic type and the optimal early warning threshold to obtain the risk level of the target sea area for the target biological species and the dynamic early warning threshold of the microplastic concentration.

Description

Technical Field

[0001] This invention proposes a method and system for early warning of microplastic ecological risks based on knowledge graphs and machine learning, which belongs to the field of ecological risk prediction technology. Background Technology

[0002] With the escalation of marine pollution, microplastics, due to their wide distribution, non-degradability, and high toxicity, pose a serious threat to marine ecosystems and biosecurity. Therefore, conducting microplastic ecological risk early warning has become a key requirement for marine environmental governance. Existing microplastic ecological risk early warning methods mainly rely on machine learning models, using monitoring data to train models to predict concentrations and determine risk levels. However, current microplastic ecological risk early warning methods suffer from low accuracy, weak dynamic adaptability, and an inability to meet the precise early warning needs in complex marine environments. Summary of the Invention

[0003] This invention provides a method and system for early warning of microplastic ecological risks based on knowledge graphs and machine learning, in order to solve the technical problems existing in the prior art. The technical solution adopted is as follows:

[0004] A microplastic ecological risk early warning method based on knowledge graphs and machine learning, the microplastic ecological risk early warning method includes:

[0005] Collect microplastic ecology-related data from different data sources, and preprocess the microplastic ecology-related data to form a standardized data set with a unified spatiotemporal framework and correlation relationships;

[0006] Using the standardized dataset with a unified spatiotemporal framework and correlations, a microplastic ecological risk knowledge graph is constructed, with microplastics, marine organisms, exposure pathways and toxic effects as core entities, and with spatiotemporal dynamic data as correlations.

[0007] Based on the input microplastic concentration data of the target sea area and the target biological species, the mechanism warning threshold for microplastic concentration is generated by utilizing the historical evidence of microplastic concentration data and target biological species in the microplastic ecological risk knowledge graph of the target sea area.

[0008] Based on the input microplastic concentration data of the target sea area and the target biological species, the machine learning early warning model that has been trained is used to obtain the predicted value of microplastic concentration and the optimal early warning threshold corresponding to the microplastic type. The optimal early warning threshold and the mechanism early warning threshold are then fused to obtain the risk level and dynamic early warning threshold of microplastic concentration for the target sea area for the target biological species.

[0009] Furthermore, the process of collecting microplastic ecology-related data from different data sources and preprocessing this data to form a standardized dataset with a unified spatiotemporal framework and correlations includes:

[0010] Collect microplastic ecology-related data from different data sources, including academic literature databases, environmental monitoring databases, biological species databases, and chemical toxicity databases.

[0011] The microplastic ecology-related data are cleaned, abnormal data are removed, and data is normalized to obtain the initial pre-processed microplastic ecology-related data.

[0012] The microplastic ecology-related data after initial preprocessing are subjected to data association and spatiotemporal framework unification processing to obtain a standardized data set with a unified spatiotemporal framework and correlation relationship corresponding to the microplastic ecology-related data.

[0013] Furthermore, the microplastic ecology-related data after initial preprocessing are subjected to data association and spatiotemporal framework unification processing to obtain a standardized dataset with a unified spatiotemporal framework and correlation relationships corresponding to the microplastic ecology-related data, including:

[0014] Retrieve time-related data from the initial preprocessed microplastic ecological data and add standard timestamps to all time-related data.

[0015] Data information with spatial attributes is retrieved from the microplastic ecology-related data after the initial preprocessing, and standard spatial coordinates are added to all data information with spatial attributes;

[0016] By combining standard timestamps and standard spatial coordinates, a standardized dataset with a unified spatiotemporal framework and correlations is generated to correspond to microplastic ecological data.

[0017] Furthermore, a microplastic ecological risk knowledge graph is constructed using the standardized dataset with a unified spatiotemporal framework and correlation relationships, with microplastics, marine organisms, exposure pathways, and toxic effects as core entities, and correlated with spatiotemporal dynamic data, including:

[0018] Entity data is extracted from the standardized dataset with a unified spatiotemporal framework and relationships, wherein the types of entity data include microplastics, marine organisms, exposure pathways, and toxic effects;

[0019] Create entity nodes corresponding to each type of entity data, retrieve the relationships between each type of entity data, and construct an initial microplastic ecological risk knowledge graph by combining the entity nodes corresponding to each type of entity data and their relationships.

[0020] The attribute information of the same entity in different data sources is collected and deduplicated to obtain the attribute information corresponding to each entity node, and the attribute information is populated into the initial microplastic ecological risk knowledge graph.

[0021] Concentration detection events are retrieved from a standardized dataset with a unified spatiotemporal framework and correlations. These concentration detection events are then combined with an initial microplastic ecological risk knowledge graph to generate a microplastic ecological risk knowledge graph with correlated spatiotemporal dynamic data.

[0022] Furthermore, by combining concentration detection events with the initial microplastic ecological risk knowledge graph, a microplastic ecological risk knowledge graph with associated spatiotemporal dynamic data is generated, including:

[0023] For each concentration detection event, retrieve the event data information corresponding to the concentration detection event, wherein the event data information includes at least the concentration detection time, concentration detection location, and concentration detection value;

[0024] Based on the event data information corresponding to each concentration detection event, the spatiotemporal coordinates corresponding to each concentration detection event are determined, wherein the spatiotemporal coordinates are generated according to the concentration detection time and concentration detection location;

[0025] Retrieve the entity node corresponding to the detection time of each concentration detection event in the initial microplastic ecological risk knowledge graph;

[0026] Each concentration detection event with spatiotemporal coordinates is associated with a corresponding entity node to characterize the specific environmental carrier corresponding to each entity node and the concentration detection events occurring on that carrier, thereby generating a microplastic ecological risk knowledge graph with associated spatiotemporal dynamic data.

[0027] Furthermore, based on the input microplastic concentration data of the target sea area and the target biological species, and utilizing historical evidence of microplastic concentration data and target biological species in the microplastic ecological risk knowledge graph, a mechanism-based early warning threshold for microplastic concentration is generated, including:

[0028] Based on the input target marine microplastic concentration data and target biological species, the entity node corresponding to the target biological species is located in the microplastic ecological risk knowledge graph, and the entity node is used as the core biological species node.

[0029] Using the core biological species node as the center, retrieve and extract entity nodes that are related to the target biological species corresponding to the core biological species node and match the input target sea area, and use them as candidate microplastic nodes.

[0030] Retrieve the attribute information corresponding to each candidate microplastic node, and obtain microplastic characteristic parameters from the attribute information corresponding to the candidate microplastic node. The microplastic characteristic parameters include microplastic particle size, microplastic concentration data, and microplastic distribution area.

[0031] The microplastic concentration data corresponding to each candidate microplastic node is compared with the microplastic concentration data of the input target sea area to obtain the comparison results;

[0032] Based on the comparison results, candidate microplastic nodes whose microplastic concentration data and the input target sea area microplastic concentration data are within a preset difference range are selected from the candidate microplastic nodes and used as core microplastic nodes corresponding to the target biological species.

[0033] Based on the microplastic characteristic parameters corresponding to the core microplastic nodes, and combined with the species concentration and distribution area of ​​the target biological species at each historical monitoring time, a mechanism warning threshold for microplastic concentration is generated.

[0034] Furthermore, based on the microplastic characteristic parameters corresponding to the core microplastic nodes, and combined with the species concentration and distribution area of ​​the target biological species at each historical monitoring time, a mechanism-based early warning threshold for microplastic concentration is generated, including:

[0035] Retrieve the standard timestamp corresponding to each historical monitoring moment of the target biological species, and extract the microplastic feature parameters corresponding to each core microplastic node under the standard timestamp based on the standard timestamp corresponding to the historical monitoring moment.

[0036] By utilizing the microplastic characteristic parameters of each core microplastic node, as well as the species concentration and distribution area of ​​the target biological species at each historical monitoring time, the mechanism warning threshold for generating microplastic concentration is calculated.

[0037] Furthermore, based on the input microplastic concentration data of the target sea area and the target biological species, a pre-trained machine learning early warning model is used to obtain the predicted microplastic concentration value and optimal early warning threshold corresponding to the microplastic type, including:

[0038] The microplastic concentration data and target biological species of the target sea area are input into a machine learning early warning model that has been trained.

[0039] The trained machine learning early warning model is used to obtain the predicted concentration values ​​and optimal early warning thresholds for each type of microplastic.

[0040] Furthermore, by fusing the optimal early warning threshold and the mechanistic early warning threshold, dynamic early warning thresholds for the risk level and microplastic concentration of the target marine area are obtained, including:

[0041] The optimal early warning threshold and the mechanism early warning threshold are retrieved, and the optimal early warning threshold and the mechanism early warning threshold are fused according to the fusion strategy to generate a dynamic early warning threshold;

[0042] The predicted concentration of microplastics corresponding to the type of microplastic is compared with the dynamic warning threshold. If the predicted concentration of microplastics is not lower than the dynamic warning threshold, the risk level of the target sea area for the target biological species is determined to be high; otherwise, the risk level of the target sea area for the target biological species is determined to be low.

[0043] A microplastic ecological risk early warning system based on knowledge graphs and machine learning, the microplastic ecological risk early warning system includes:

[0044] The data acquisition and preprocessing module is used to collect microplastic ecology-related data from different data sources and preprocess the microplastic ecology-related data to form a standardized data set with a unified spatiotemporal framework and correlation.

[0045] The knowledge graph construction module is used to construct a microplastic ecological risk knowledge graph with microplastics, marine organisms, exposure pathways and toxic effects as core entities and associated with spatiotemporal dynamic data using the standardized data set with a unified spatiotemporal framework and correlation relationships.

[0046] The mechanism early warning module is used to generate a mechanism early warning threshold for microplastic concentration based on the input microplastic concentration data of the target sea area and the target biological species, using historical evidence of microplastic concentration data of the target sea area and the target biological species in the microplastic ecological risk knowledge graph.

[0047] The integrated early warning module is used to obtain the predicted value of microplastic concentration and the optimal early warning threshold corresponding to the microplastic type based on the input microplastic concentration data of the target sea area and the target biological species, using a pre-trained machine learning early warning model. It then uses the optimal early warning threshold and the mechanism early warning threshold to perform a fusion judgment to obtain the dynamic early warning threshold of the risk level and microplastic concentration of the target sea area for the target biological species.

[0048] Beneficial effects of this invention:

[0049] The microplastic ecological risk early warning method and system proposed in this invention, based on knowledge graphs and machine learning, combines early warning thresholds with optimal early warning thresholds. This overcomes the shortcomings of insufficient adaptability when relying solely on theoretical analysis and avoids the problem of low threshold determination accuracy due to poor representativeness when using machine learning models alone. Furthermore, by fusing early warning thresholds obtained from both methods, the system conforms to toxicological principles while closely reflecting the actual conditions of specific marine areas, effectively improving the accuracy of threshold setting and its high degree of bidirectional matching with toxicological principles and actual marine conditions. On the other hand, this embodiment can adaptively adjust to the environmental characteristics and biological sensitivity of different marine areas. The generated dynamic early warning thresholds can be updated in real time according to the spatiotemporal changes in microplastic concentration in the marine area, thereby accurately determining the risk level and effectively meeting the needs for refined and dynamic risk early warning in various complex marine environments, thus effectively improving the accuracy and dynamic adaptability of early warnings. Attached Figure Description

[0050] Figure 1 This is a flowchart of the method described in this invention;

[0051] Figure 2 This is a system block diagram of the system described in this invention. Detailed Implementation

[0052] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0053] This embodiment proposes a microplastic ecological risk early warning method based on knowledge graphs and machine learning, such as Figure 1 As shown, the microplastic ecological risk early warning method includes:

[0054] Collect microplastic ecology-related data from different data sources, and preprocess the microplastic ecology-related data to form a standardized data set with a unified spatiotemporal framework and correlation relationships;

[0055] Using the standardized dataset with a unified spatiotemporal framework and correlations, a microplastic ecological risk knowledge graph is constructed, with microplastics, marine organisms, exposure pathways and toxic effects as core entities, and with spatiotemporal dynamic data as correlations.

[0056] Based on the input microplastic concentration data of the target sea area and the target biological species, the mechanism warning threshold for microplastic concentration is generated by utilizing the historical evidence of microplastic concentration data and target biological species in the microplastic ecological risk knowledge graph of the target sea area.

[0057] Based on the input microplastic concentration data of the target sea area and the target biological species, the machine learning early warning model that has been trained is used to obtain the predicted value of microplastic concentration and the optimal early warning threshold corresponding to the microplastic type. The optimal early warning threshold and the mechanism early warning threshold are then fused to obtain the risk level and dynamic early warning threshold of microplastic concentration for the target sea area for the target biological species.

[0058] This embodiment collects multi-source microplastic ecological-related data and preprocesses it to form a standardized dataset. It then constructs a knowledge graph containing core entities such as microplastics and marine organisms, and dynamically correlates them in time and space. On the one hand, it uses historical evidence in the knowledge graph to deduce the mechanism warning threshold. On the other hand, it obtains the concentration prediction value and the optimal warning threshold through a trained machine learning model. Finally, it integrates the two types of thresholds to achieve risk level determination and dynamic warning threshold generation. The core is to integrate the mechanism knowledge of the knowledge graph with the data-driven capability of machine learning to achieve multi-source data correlation and mechanism data fusion warning.

[0059] Meanwhile, by combining early warning thresholds with optimal early warning thresholds, this approach not only overcomes the shortcomings of relying solely on theoretical analysis for adaptability but also avoids the problem of low threshold determination accuracy due to poor representativeness caused by using only machine learning models. Furthermore, by fusing the early warning thresholds obtained from both methods, they conform to toxicological principles while closely reflecting the actual conditions of specific sea areas, effectively improving the accuracy of threshold settings and their high degree of bidirectional matching with toxicological principles and actual sea conditions. On the other hand, this embodiment can adaptively adjust to the environmental characteristics and biological sensitivity of different sea areas. The generated dynamic early warning thresholds can be updated in real time according to the spatiotemporal changes in microplastic concentration in the sea area, thereby accurately determining the risk level and effectively meeting the needs for refined and dynamic risk early warning in various complex marine environments, thus effectively improving the accuracy and dynamic adaptability of early warnings.

[0060] In one embodiment of the present invention, the step of collecting microplastic ecology-related data from different data sources and preprocessing the microplastic ecology-related data to form a standardized dataset with a unified spatiotemporal framework and correlation relationships includes:

[0061] Collect microplastic ecology-related data from different data sources, including but not limited to academic literature databases, environmental monitoring databases, biological species databases, and chemical toxicity databases;

[0062] The microplastic ecology-related data are cleaned, abnormal data are removed, and data is normalized to obtain the initial pre-processed microplastic ecology-related data.

[0063] The microplastic ecology-related data after initial preprocessing are subjected to data association and spatiotemporal framework unification processing to obtain a standardized data set with a unified spatiotemporal framework and correlation relationship corresponding to the microplastic ecology-related data.

[0064] Specifically, the preprocessed microplastic ecology-related data undergoes data association and spatiotemporal framework unification processing to obtain a standardized dataset with a unified spatiotemporal framework and correlation relationships, including:

[0065] Data with time attributes are retrieved from the microplastic ecological-related data after initial preprocessing, and a standard timestamp, such as UTC time, is added to all data with time attributes; wherein, the data with time attributes includes, but is not limited to, monitoring data, simulation data, and experimental data;

[0066] Data information with spatial attributes is retrieved from the initial preprocessed microplastic ecological data, and standard spatial coordinates are added to all data information with spatial attributes; wherein, the data information with spatial attributes includes, but is not limited to, monitoring points, simulation grids, and species distribution ranges, etc.

[0067] By combining standard timestamps and standard spatial coordinates, a standardized dataset with a unified spatiotemporal framework and correlations is generated to correspond to microplastic ecological data.

[0068] In this embodiment, firstly, microplastic-related ecological data are collected from multiple sources, including academic literature databases and environmental monitoring databases. Initial preprocessing is completed through data cleaning, anomaly removal, and normalization. Next, standard timestamps are added to time-attribute data, and standard spatial coordinates are added to spatial-attribute data to achieve data association and construct a unified spatiotemporal framework, forming a standardized data set. Then, based on this set, a microplastic ecological risk knowledge graph is constructed, containing core entities such as microplastics and marine organisms, and their spatiotemporal dynamic associations. Subsequently, historical evidence of target sea areas and target organisms in the knowledge graph is used to deduce early warning thresholds. Simultaneously, a trained machine learning model processes the input data to obtain predicted microplastic concentration values ​​and optimal early warning thresholds. Finally, a fusion strategy integrates the two types of thresholds to complete risk level determination and generate dynamic early warning thresholds. This embodiment ensures data comprehensiveness by covering multi-dimensional data sources, improves data purity through preprocessing, and breaks down data barriers using standard spatiotemporal identifiers to form a structured and associated data set, providing accurate data information for subsequent steps.

[0069] One embodiment of the present invention utilizes the standardized dataset with a unified spatiotemporal framework and correlations to construct a microplastic ecological risk knowledge graph with microplastics, marine organisms, exposure pathways, and toxic effects as core entities, and correlated with spatiotemporal dynamic data, including:

[0070] Entity data is extracted from the standardized dataset with a unified spatiotemporal framework and relationships, wherein the types of entity data include microplastics, marine organisms, exposure pathways, and toxic effects;

[0071] Create entity nodes corresponding to each type of entity data, retrieve the relationships between each type of entity data, and construct an initial microplastic ecological risk knowledge graph by combining the entity nodes corresponding to each type of entity data and their relationships.

[0072] The attribute information of the same entity in different data sources is collected and deduplicated to obtain the attribute information corresponding to each entity node, and the attribute information is populated into the initial microplastic ecological risk knowledge graph.

[0073] Concentration detection events are retrieved from a standardized dataset with a unified spatiotemporal framework and correlations. These concentration detection events are then combined with an initial microplastic ecological risk knowledge graph to generate a microplastic ecological risk knowledge graph with correlated spatiotemporal dynamic data.

[0074] In this embodiment, firstly, microplastic ecological data are collected from multiple sources, including academic literature databases and environmental monitoring databases. Initial preprocessing is performed through data cleaning, anomaly removal, and normalization. Then, standard timestamps are added to time-attribute data, and standard spatial coordinates are added to spatial-attribute data, forming a standardized data set with a unified spatiotemporal framework and relationships. Next, when constructing a knowledge graph based on this set, four types of entity data—microplastics, marine organisms, exposure pathways, and toxic effects—are extracted, and corresponding entity nodes are created. An initial graph is constructed by combining the relationships between entities. Then, multi-source attribute information of the same entity is collected, deduplicated, and filled into the initial graph. Subsequently, concentration detection events from the standardized data are retrieved to generate a microplastic ecological risk knowledge graph with associated spatiotemporal dynamic data. Then, historical evidence of target sea areas and target organisms in the knowledge graph is used to deduce the mechanism warning threshold. Simultaneously, a trained machine learning model processes the input data to obtain predicted microplastic concentration values ​​and optimal warning thresholds. Finally, a fusion strategy is used to integrate the two types of thresholds to complete the risk level determination and generate dynamic warning thresholds.

[0075] Meanwhile, this embodiment achieves high-quality integration of multi-source heterogeneous data. By covering multiple data sources and standardizing the processing, it breaks down the spatiotemporal barriers of data and forms a structured and related data set. At the same time, by using a unified spatiotemporal framework and dynamic knowledge association, it ensures that the early warning results are consistent with the spatiotemporal characteristics of the target sea area, effectively improving the accuracy of subsequent knowledge graph construction.

[0076] One embodiment of the present invention utilizes concentration detection events combined with an initial microplastic ecological risk knowledge graph to generate a microplastic ecological risk knowledge graph with associated spatiotemporal dynamic data, including:

[0077] For each concentration detection event, retrieve the event data information corresponding to the concentration detection event, wherein the event data information includes at least the concentration detection time, concentration detection location, and concentration detection value;

[0078] Based on the event data information corresponding to each concentration detection event, the spatiotemporal coordinates corresponding to each concentration detection event are determined, wherein the spatiotemporal coordinates are generated according to the concentration detection time and concentration detection location;

[0079] Retrieve the entity node corresponding to the detection time of each concentration detection event in the initial microplastic ecological risk knowledge graph;

[0080] Each concentration detection event with spatiotemporal coordinates is associated with a corresponding entity node to characterize the specific environmental carrier corresponding to each entity node and the concentration detection events occurring on that carrier, thereby generating a microplastic ecological risk knowledge graph with associated spatiotemporal dynamic data.

[0081] This embodiment uses an initial microplastic ecological risk knowledge graph as its basic framework, and expands the static knowledge graph into a spatiotemporally dynamic knowledge graph by using concentration detection events as the association carrier of dynamic spatiotemporal information. First, event data information such as detection time, location, and value are extracted from the concentration detection events. Second, each detection event is assigned a unique spatiotemporal coordinate based on the time and location information. Then, the corresponding entity nodes in the initial knowledge graph are matched according to the detection time. Finally, the detection events with spatiotemporal coordinates are associated with their respective entity nodes, binding the ecological risk attributes of the entity nodes with the spatiotemporally dynamic data of the specific environmental carrier, ultimately generating a microplastic ecological risk knowledge graph with associated spatiotemporally dynamic data.

[0082] This embodiment breaks through the static attribute limitations of the initial knowledge graph, achieving deep integration of microplastic ecological risk knowledge with spatiotemporal dynamic data, enabling the knowledge graph to possess dynamic representation capabilities in the spatiotemporal dimension. Simultaneously, by precisely associating concentration detection events with entity nodes through spatiotemporal coordinates, it clarifies the specific environmental carriers and spatiotemporal background corresponding to the ecological risks of entity nodes, effectively improving the relevance and directionality of risk knowledge. Furthermore, by mining the spatiotemporal attributes of concentration detection event data, discrete detection data is transformed into dynamic node association information in the knowledge graph, effectively enhancing and enriching the expressive dimensions of microplastic ecological risk knowledge. This achieves a full-dimensional representation of microplastic ecological risks from static entity relationships to dynamic spatiotemporal evolution, providing more complete knowledge support for subsequent risk assessment and trend prediction.

[0083] In one embodiment of the present invention, based on input microplastic concentration data of a target sea area and target biological species, a mechanism-based early warning threshold for microplastic concentration is generated using historical evidence of microplastic concentration data and target biological species in the microplastic ecological risk knowledge graph of the target sea area, including:

[0084] Based on the input target marine microplastic concentration data and target biological species, the entity node corresponding to the target biological species is located in the microplastic ecological risk knowledge graph, and the entity node is used as the core biological species node.

[0085] Using the core biological species node as the center, retrieve and extract entity nodes that are related to the target biological species corresponding to the core biological species node and match the input target sea area, and use them as candidate microplastic nodes.

[0086] Retrieve the attribute information corresponding to each candidate microplastic node, and obtain microplastic characteristic parameters from the attribute information corresponding to the candidate microplastic node. The microplastic characteristic parameters include microplastic particle size, microplastic concentration data, and microplastic distribution area.

[0087] The microplastic concentration data corresponding to each candidate microplastic node is compared with the microplastic concentration data of the input target sea area to obtain the comparison results;

[0088] Based on the comparison results, candidate microplastic nodes whose microplastic concentration data and the input target sea area microplastic concentration data are within a preset difference range are selected from the candidate microplastic nodes and used as core microplastic nodes corresponding to the target biological species.

[0089] Based on the microplastic characteristic parameters corresponding to the core microplastic nodes, and combined with the species concentration and distribution area of ​​the target biological species at each historical monitoring time, a mechanism warning threshold for microplastic concentration is generated.

[0090] This embodiment uses a microplastic ecological risk knowledge graph based on spatiotemporal dynamic data as its core support. It focuses on microplastic concentration data in target sea areas and target biological species, combining node localization, association retrieval, parameter extraction, data comparison, node screening, and threshold generation to generate a microplastic concentration mechanism early warning threshold. First, based on the input information, the core biological species node corresponding to the target biological species is located in the knowledge graph. Second, using this node as the center, related entity nodes matching the microplastic type in the target sea area are retrieved as candidate microplastic nodes. Then, the attribute information of the candidate nodes is retrieved, extracting characteristic parameters such as microplastic particle size, concentration, and distribution area. Subsequently, the concentration data of the candidate nodes is compared with the input concentration data of the target sea area, and nodes with a difference within a preset range of 3%-7% are selected as core microplastic nodes. Finally, combining the characteristic parameters of the core microplastic nodes with the historical monitoring species concentration and distribution area of ​​the target biological species, a mechanism early warning threshold is generated.

[0091] This embodiment deeply integrates early warning threshold generation with microplastic characteristics and historical biological monitoring data based on historical evidence and relationships in a knowledge graph. This approach differs from traditional experience-based threshold settings, giving the thresholds a clear representation of ecological mechanisms. Simultaneously, by filtering core microplastic nodes within a preset difference range of 3%-7%, it achieves precise matching between microplastics in the target sea area and historical data in the knowledge graph, effectively improving the targeting and reliability of threshold generation. Furthermore, by fully exploring the relationships between biological species and microplastic types in the knowledge graph, as well as historical attribute data, it transforms static knowledge associations into a dynamic basis for early warning threshold generation. This approach effectively improves the accuracy and sensitivity of the mechanistic early warning threshold in reflecting the potential risk threshold of microplastic concentrations in the target sea area to target biological species.

[0092] In one embodiment of the present invention, based on the microplastic characteristic parameters corresponding to the core microplastic node, and combined with the species concentration and distribution area of ​​the target biological species obtained at various historical monitoring times, a mechanism-based early warning threshold for microplastic concentration is generated, including:

[0093] Retrieve the standard timestamp corresponding to each historical monitoring moment of the target biological species, and extract the microplastic feature parameters corresponding to each core microplastic node under the standard timestamp based on the standard timestamp corresponding to the historical monitoring moment.

[0094] By utilizing the microplastic characteristic parameters of each core microplastic node, as well as the species concentration and distribution area of ​​the target biological species at each historical monitoring time, the mechanism warning threshold for generating microplastic concentration is calculated.

[0095] The process for obtaining the mechanism warning threshold corresponding to the microplastic concentration is as follows:

[0096] The species concentration and distribution area of ​​the target biological species are retrieved from each historical monitoring period, and the species concentration and distribution area obtained from each historical monitoring period are normalized to obtain the normalized species concentration and distribution area.

[0097] The average values ​​of species concentration and distribution area are obtained by using the normalized species concentration and distribution area corresponding to each historical monitoring time of the target biological species.

[0098] An ecological state regulation coefficient Ψ = exp[-λ×C / (1+S)] is generated using the average values ​​of the species concentration and distribution area corresponding to the target biological species; where λ represents the ecological sensitivity coefficient, used to control the degree of influence of species aggregation on risk perception, and its value ranges from 0.5 to 1.2; C represents the average value of the species concentration of the target biological species; and S represents the average value corresponding to the distribution area of ​​the target biological species.

[0099] The microplastic feature vector for each core microplastic node is formed by using the microplastic feature parameters corresponding to the core microplastic node, and the security reference vector to which the microplastic type belongs is retrieved for the core microplastic node.

[0100] Calculate the Mahalanobis distance D between the microplastic feature vector and the safety reference vector corresponding to the core microplastic node, and use the Mahalanobis distance D between the microplastic feature vector and the safety reference vector to obtain the feature concentration benchmark parameter Γ corresponding to the microplastic type of the core microplastic node. z =C mp ×(1+η×D); where C mp This represents the microplastic concentration data corresponding to the core microplastic node; η represents the characteristic deviation influence coefficient, with a value range of 0.7-1.3.

[0101] The weight values ​​corresponding to the core microplastic nodes are combined with the feature concentration benchmark parameters corresponding to the microplastic types of the core microplastic nodes to generate a weighted average, thus generating the fused feature concentration parameter Γ corresponding to the microplastic type. r The weight value corresponding to the core microplastic node is set based on the risk contribution of different core microplastic nodes to the target organism, and the value range is 0-1.

[0102] The ecological state regulation coefficient Ψ and the fusion characteristic concentration parameter Г corresponding to the microplastic type are used. r Obtain the mechanism warning threshold Г=Г corresponding to the microplastic concentration. r ×Ф(Г r×Ψ); where Ф() represents a nonlinear output function, and the structure of the nonlinear output function is Ф(x)=x / [1+(x / Г). max ) 2 ] 0.5 ;Г max This indicates the maximum concentration of microplastics in historical monitoring data.

[0103] This embodiment uses the characteristic parameters of core microplastic nodes and historical monitoring data of target biological species as dual core inputs. Through a hierarchical logical process involving timestamp alignment, data normalization, coefficient modeling, vector operations, parameter fusion, and nonlinear mapping, it achieves the quantitative generation of microplastic concentration mechanism warning thresholds. First, it retrieves the standard timestamps of historical monitoring of the target organism and extracts the characteristic parameters of the core microplastic nodes at the corresponding timestamps. Second, it normalizes the historical species concentration and distribution area of ​​the target organism, calculates their average value, and substitutes it into a formula to generate an ecological state regulation coefficient. Then, it converts the core microplastic characteristic parameters into feature vectors, calculates the Mahalanobis distance using its safety reference vector, and further generates characteristic concentration benchmark parameters. Subsequently, it performs a weighted average of each core microplastic node based on risk contribution weights to obtain fused characteristic concentration parameters. Finally, it determines the mechanism warning threshold for microplastic concentration by mapping the product of the fused characteristic concentration parameters and the ecological state regulation coefficient using a nonlinear output function Ф(·).

[0104] Meanwhile, this embodiment achieves precise spatiotemporal alignment of microplastic characteristic parameters with biological historical monitoring data through timestamps. Combined with normalization and Mahalanobis distance, it eliminates interference from differences in data dimensionality and distribution, effectively improving the accuracy of threshold quantification. Furthermore, using ecological state adjustment coefficients for mechanistic threshold setting effectively enhances the responsiveness of biological ecological states to the dynamic correction performance of risk thresholds and the sensitivity of adaptive control. Simultaneously, the feature deviation influence coefficient and weight values ​​reflect the differences in risk contribution of different microplastic types, enabling threshold generation to effectively improve its characterization of microplastic characteristic mechanisms. In addition, this embodiment employs weighted averaging to effectively fuse the characteristic parameters of multiple core microplastic nodes. A nonlinear output function constrains and maps the fused parameters, effectively avoiding excessive influence of extreme values ​​on the threshold, thereby improving the rationality and accuracy of mechanistic early warning threshold setting.

[0105] In one embodiment of the present invention, based on input microplastic concentration data of a target sea area and target biological species, a pre-trained machine learning early warning model is used to obtain predicted microplastic concentration values ​​and optimal early warning thresholds corresponding to different microplastic types, including:

[0106] The microplastic concentration data and target biological species of the target sea area are input into a machine learning early warning model that has been trained. Specifically, the machine learning early warning model is a machine learning model with an XGBoost structure.

[0107] The trained machine learning early warning model is used to obtain the predicted concentration values ​​and optimal early warning thresholds for each type of microplastic.

[0108] Then, by fusing the optimal early warning threshold and the mechanistic early warning threshold, dynamic early warning thresholds for the risk level and microplastic concentration of the target marine area are obtained, including:

[0109] The optimal early warning threshold and the mechanism early warning threshold are retrieved, and the optimal early warning threshold and the mechanism early warning threshold are fused according to the fusion strategy to generate a dynamic early warning threshold;

[0110] The predicted concentration of microplastics corresponding to the type of microplastic is compared with the dynamic warning threshold. If the predicted concentration of microplastics is not lower than the dynamic warning threshold, the risk level of the target sea area for the target biological species is determined to be high; otherwise, the risk level of the target sea area for the target biological species is determined to be low.

[0111] The fusion strategy is as follows: when the optimal early warning threshold is lower than the mechanism early warning threshold, the dynamic early warning threshold R = (Y × k + Г × s) × α + 0.5 × |Y - Г| × (1 - α), where Y represents the optimal early warning threshold; Г represents the mechanism early warning threshold; k and s represent the target biological species sensitivity correction coefficient and the target marine environment adaptation coefficient, respectively, with values ​​ranging from 0.7 to 1.3 and 0.8 to 1.2, and the specific values ​​of the target biological species sensitivity correction coefficient and the target marine environment adaptation coefficient are obtained based on experience; α represents the threshold fusion weight coefficient, with a value of 0.6.

[0112] When the optimal early warning threshold is not lower than the mechanistic early warning threshold, the dynamic early warning threshold R = (Y × Г). 0.5 .

[0113] In this embodiment, firstly, the microplastic concentration data of the target sea area and the target biological species are input into a pre-trained XGBoost structured machine learning early warning model, which outputs the concentration prediction value and the optimal early warning threshold corresponding to the microplastic type. Secondly, the optimal early warning threshold and the aforementioned mechanism early warning threshold are retrieved, and a corresponding fusion strategy is selected based on their values: when the optimal early warning threshold is lower than the mechanism early warning threshold, a weighted fusion formula including a biological sensitivity correction coefficient, a marine environment adaptation coefficient, and a fusion weight coefficient is used to calculate the dynamic early warning threshold; when the optimal early warning threshold is not lower than the mechanism early warning threshold, the geometric mean of the two is used to calculate the dynamic early warning threshold. Finally, the concentration prediction value is compared with the dynamic early warning threshold, and the risk level is determined to be low or high based on whether the prediction value is lower than the threshold.

[0114] This embodiment employs a dual-threshold fusion strategy to achieve the complementary advantages of mechanistic thresholds and optimal thresholds from machine learning. By combining species sensitivity, marine environment adaptability correction coefficients, and scenario-specific fusion formulas, the dynamic early warning threshold possesses scenario-based dynamic adaptability. Utilizing the strong nonlinear fitting capability of the machine learning model, it outputs both the predicted concentration value and the optimal threshold. Combined with the ecological mechanism support of the mechanistic threshold, this effectively improves the accuracy of risk prediction and the rationality of the early warning threshold. Simultaneously, by directly comparing the predicted concentration value with the dynamic early warning threshold, a binary determination of risk level is achieved, simplifying the early warning decision-making process and effectively improving the efficiency and clarity of risk level determination. Furthermore, the parallel output of the dual models and the strategy-specific fusion mechanism effectively avoid the limitations of a single model threshold, minimizing the impact of extreme situations on the early warning results and effectively improving the overall robustness of the microplastic ecological risk early warning system. Furthermore, the fusion strategy in this embodiment achieves a deep complementarity between the data-driven advantage of the optimal early warning threshold from machine learning and the ecological mechanism support advantage of the early warning threshold from the mechanism perspective through scenario-specific fusion logic. At the same time, through the refined correction of the target biological species sensitivity correction coefficient and the target marine environment adaptation coefficient, as well as the quantitative control of the fixed fusion weight coefficient, the generated dynamic early warning threshold can not only accurately adapt to different data characteristics and ecological scenarios, but also has stable and reliable quantitative attributes. This effectively improves the rationality, accuracy, and matching of the dynamic early warning threshold setting with the actual scenario, thereby effectively improving the accuracy of the determination of the microplastic ecological risk level of the target marine area for the target biological species.

[0115] This invention proposes a microplastic ecological risk early warning system based on knowledge graphs and machine learning, such as... Figure 2 As shown, the microplastic ecological risk early warning system includes:

[0116] The data acquisition and preprocessing module is used to collect microplastic ecology-related data from different data sources and preprocess the microplastic ecology-related data to form a standardized data set with a unified spatiotemporal framework and correlation.

[0117] The knowledge graph construction module is used to construct a microplastic ecological risk knowledge graph with microplastics, marine organisms, exposure pathways and toxic effects as core entities and linked with spatiotemporal dynamic data using the standardized data set with a unified spatiotemporal framework and correlation relationships.

[0118] The mechanism early warning module is used to generate a mechanism early warning threshold for microplastic concentration based on the input microplastic concentration data of the target sea area and the target biological species, using historical evidence of microplastic concentration data and target biological species in the microplastic ecological risk knowledge graph.

[0119] The integrated early warning module is used to obtain the predicted value of microplastic concentration and the optimal early warning threshold corresponding to the microplastic type based on the input microplastic concentration data of the target sea area and the target biological species, using a pre-trained machine learning early warning model. It then uses the optimal early warning threshold and the mechanism early warning threshold to perform a fusion judgment to obtain the dynamic early warning threshold of the risk level and microplastic concentration of the target sea area for the target biological species.

[0120] This embodiment collects multi-source microplastic ecological-related data and preprocesses it to form a standardized dataset. It then constructs a knowledge graph containing core entities such as microplastics and marine organisms, and dynamically correlates them in time and space. On the one hand, it uses historical evidence in the knowledge graph to deduce the mechanism warning threshold. On the other hand, it obtains the concentration prediction value and the optimal warning threshold through a trained machine learning model. Finally, it integrates the two types of thresholds to achieve risk level determination and dynamic warning threshold generation. The core is to integrate the mechanism knowledge of the knowledge graph with the data-driven capability of machine learning to achieve multi-source data correlation and mechanism data fusion warning.

[0121] Meanwhile, by combining early warning thresholds with optimal early warning thresholds, this approach not only overcomes the shortcomings of relying solely on theoretical analysis for adaptability but also avoids the problem of low threshold determination accuracy due to poor representativeness caused by using only machine learning models. Furthermore, by fusing the early warning thresholds obtained from both methods, they conform to toxicological principles while closely reflecting the actual conditions of specific sea areas, effectively improving the accuracy of threshold settings and their high degree of bidirectional matching with toxicological principles and actual sea conditions. On the other hand, this embodiment can adaptively adjust to the environmental characteristics and biological sensitivity of different sea areas. The generated dynamic early warning thresholds can be updated in real time according to the spatiotemporal changes in microplastic concentration in the sea area, thereby accurately determining the risk level and effectively meeting the needs for refined and dynamic risk early warning in various complex marine environments, thus effectively improving the accuracy and dynamic adaptability of early warnings.

[0122] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A microplastic ecological risk early warning method based on knowledge graphs and machine learning, characterized in that, The microplastic ecological risk early warning method includes: Collect microplastic ecology-related data from different data sources, and preprocess the microplastic ecology-related data to form a standardized data set with a unified spatiotemporal framework and correlation relationships; Using the standardized dataset with a unified spatiotemporal framework and correlations, a microplastic ecological risk knowledge graph is constructed, with microplastics, marine organisms, exposure pathways and toxic effects as core entities, and with spatiotemporal dynamic data as correlations. Based on the input microplastic concentration data of the target sea area and the target biological species, and utilizing historical evidence of microplastic concentration data and target biological species in the microplastic ecological risk knowledge graph, a mechanistic warning threshold for microplastic concentration is generated. This includes: based on the input microplastic concentration data of the target sea area and the target biological species, locating the entity node corresponding to the target biological species in the microplastic ecological risk knowledge graph, and designating the entity node as a core biological species node; retrieving and extracting entity nodes corresponding to microplastic types that are associated with the target biological species corresponding to the core biological species node and match the input target sea area, using these as candidate microplastic nodes; retrieving the attribute information corresponding to each candidate microplastic node, and analyzing the data from the candidate microplastic... Microplastic characteristic parameters are obtained from the attribute information corresponding to the plastic nodes. These parameters include microplastic particle size, microplastic concentration data, and the area of ​​the microplastic distribution range. The microplastic concentration data corresponding to each candidate microplastic node is compared with the microplastic concentration data of the input target sea area to obtain comparison results. Based on the comparison results, candidate microplastic nodes whose microplastic concentration data differs from the input target sea area microplastic concentration data within a preset range are selected as core microplastic nodes corresponding to the target biological species. Based on the microplastic characteristic parameters corresponding to the core microplastic nodes, combined with the species concentration and distribution area of ​​the target biological species obtained at each historical monitoring time, a mechanism-based early warning threshold for microplastic concentration is generated. Based on the input microplastic concentration data of the target sea area and the target biological species, the machine learning early warning model that has been trained is used to obtain the predicted value of microplastic concentration and the optimal early warning threshold corresponding to the microplastic type. The optimal early warning threshold and the mechanism early warning threshold are then fused to obtain the risk level and dynamic early warning threshold of microplastic concentration for the target sea area for the target biological species.

2. The microplastic ecological risk early warning method according to claim 1, characterized in that, The process involves collecting microplastic ecology-related data from different data sources and preprocessing this data to form a standardized dataset with a unified spatiotemporal framework and correlations, including: Collect microplastic ecology-related data from different data sources, including academic literature databases, environmental monitoring databases, biological species databases, and chemical toxicity databases; The microplastic ecology-related data are cleaned, abnormal data is removed, and data is normalized to obtain the initial pre-processed microplastic ecology-related data. The microplastic ecology-related data after initial preprocessing are subjected to data association and spatiotemporal framework unification processing to obtain a standardized data set with a unified spatiotemporal framework and correlation relationship corresponding to the microplastic ecology-related data.

3. The microplastic ecological risk early warning method according to claim 2, characterized in that, The microplastic ecology-related data after initial preprocessing are subjected to data association and spatiotemporal framework unification processing to obtain a standardized dataset with a unified spatiotemporal framework and correlation relationships, including: Retrieve time-related data from the initial preprocessed microplastic ecological data and add standard timestamps to all time-related data. Data information with spatial attributes is retrieved from the microplastic ecology-related data after the initial preprocessing, and standard spatial coordinates are added to all data information with spatial attributes; By combining standard timestamps and standard spatial coordinates, a standardized dataset with a unified spatiotemporal framework and correlations is generated to correspond to microplastic ecological data.

4. The microplastic ecological risk early warning method according to claim 1, characterized in that, A microplastic ecological risk knowledge graph is constructed using the standardized dataset with a unified spatiotemporal framework and correlations. This graph is centered on microplastics, marine organisms, exposure pathways, and toxic effects, and is linked to spatiotemporally dynamic data. The graph includes: Entity data is extracted from the standardized dataset with a unified spatiotemporal framework and relationships, wherein the types of entity data include microplastics, marine organisms, exposure pathways, and toxic effects; Create entity nodes corresponding to each type of entity data, retrieve the relationships between each type of entity data, and construct an initial microplastic ecological risk knowledge graph by combining the entity nodes corresponding to each type of entity data and their relationships. The attribute information of the same entity in different data sources is collected and deduplicated to obtain the attribute information corresponding to each entity node, and the attribute information is populated into the initial microplastic ecological risk knowledge graph. Concentration detection events are retrieved from a standardized dataset with a unified spatiotemporal framework and correlations. These concentration detection events are then combined with an initial microplastic ecological risk knowledge graph to generate a microplastic ecological risk knowledge graph with correlated spatiotemporal dynamic data.

5. The microplastic ecological risk early warning method according to claim 4, characterized in that, By combining concentration detection events with an initial microplastic ecological risk knowledge graph, a microplastic ecological risk knowledge graph with associated spatiotemporal dynamic data is generated, including: For each concentration detection event, retrieve the event data information corresponding to the concentration detection event, wherein the event data information includes at least the concentration detection time, concentration detection location, and concentration detection value; Based on the event data information corresponding to each concentration detection event, the spatiotemporal coordinates corresponding to each concentration detection event are determined, wherein the spatiotemporal coordinates are generated according to the concentration detection time and concentration detection location; Retrieve the entity node corresponding to the detection time of each concentration detection event in the initial microplastic ecological risk knowledge graph; Each concentration detection event with spatiotemporal coordinates is associated with a corresponding entity node to characterize the specific environmental carrier corresponding to each entity node and the concentration detection events occurring on that carrier, thereby generating a microplastic ecological risk knowledge graph with associated spatiotemporal dynamic data.

6. The microplastic ecological risk early warning method according to claim 1, characterized in that, Based on the microplastic characteristic parameters corresponding to the core microplastic nodes, and combined with the species concentration and distribution area of ​​the target biological species at each historical monitoring time, a mechanism-based early warning threshold for microplastic concentration is generated, including: Retrieve the standard timestamp corresponding to each historical monitoring moment of the target biological species, and extract the microplastic feature parameters corresponding to each core microplastic node under the standard timestamp based on the standard timestamp corresponding to the historical monitoring moment. By utilizing the microplastic characteristic parameters of each core microplastic node, as well as the species concentration and distribution area of ​​the target biological species at each historical monitoring time, the mechanism warning threshold for generating microplastic concentration is calculated.

7. The microplastic ecological risk early warning method according to claim 1, characterized in that, Based on the input microplastic concentration data of the target sea area and the target biological species, a pre-trained machine learning early warning model is used to obtain the predicted microplastic concentration value and optimal early warning threshold corresponding to the microplastic type, including: The microplastic concentration data and target biological species of the target sea area are input into a machine learning early warning model that has been trained. The trained machine learning early warning model is used to obtain the predicted concentration values ​​and optimal early warning thresholds for each type of microplastic.

8. The microplastic ecological risk early warning method according to claim 1, characterized in that, By fusing optimal and mechanistic warning thresholds, dynamic warning thresholds for risk levels and microplastic concentrations of target biological species in the target sea area are obtained, including: The optimal early warning threshold and the mechanism early warning threshold are retrieved, and the optimal early warning threshold and the mechanism early warning threshold are fused according to the fusion strategy to generate a dynamic early warning threshold; The predicted concentration of microplastics corresponding to the type of microplastic is compared with the dynamic warning threshold. If the predicted concentration of microplastics is not lower than the dynamic warning threshold, the risk level of the target sea area for the target biological species is determined to be high; otherwise, the risk level of the target sea area for the target biological species is determined to be low.

9. A microplastic ecological risk early warning system based on knowledge graphs and machine learning, used to execute the method described in any one of claims 1 to 8, characterized in that, The microplastic ecological risk early warning system includes: The data acquisition and preprocessing module is used to collect microplastic ecology-related data from different data sources and preprocess the microplastic ecology-related data to form a standardized data set with a unified spatiotemporal framework and correlation. The knowledge graph construction module is used to construct a microplastic ecological risk knowledge graph with microplastics, marine organisms, exposure pathways and toxic effects as core entities and associated with spatiotemporal dynamic data using the standardized data set with a unified spatiotemporal framework and correlation relationships. The mechanism early warning module is used to generate a mechanism early warning threshold for microplastic concentration based on the input microplastic concentration data and target biological species of the target sea area, and using historical evidence of microplastic concentration data and target biological species in the microplastic ecological risk knowledge graph. This includes: locating the entity node corresponding to the target biological species in the microplastic ecological risk knowledge graph based on the input microplastic concentration data and target biological species, and designating the entity node as a core biological species node; retrieving and extracting entity nodes corresponding to microplastic types that are associated with the target biological species corresponding to the core biological species node and match the input target sea area, centered on the core biological species node, as candidate microplastic nodes; and retrieving the attribute information corresponding to each candidate microplastic node from... Microplastic characteristic parameters are obtained from the attribute information corresponding to the candidate microplastic nodes. These parameters include microplastic particle size, microplastic concentration data, and the area of ​​the microplastic distribution range. The microplastic concentration data corresponding to each candidate microplastic node is compared with the microplastic concentration data of the input target sea area to obtain comparison results. Based on the comparison results, candidate microplastic nodes whose microplastic concentration data differs from the input target sea area microplastic concentration data within a preset range are selected as core microplastic nodes corresponding to the target biological species. Based on the microplastic characteristic parameters corresponding to the core microplastic nodes, combined with the species concentration and distribution area of ​​the target biological species obtained at each historical monitoring time, a mechanism-based early warning threshold for microplastic concentration is generated. The integrated early warning module is used to obtain the predicted value of microplastic concentration and the optimal early warning threshold corresponding to the microplastic type based on the input microplastic concentration data of the target sea area and the target biological species, using a pre-trained machine learning early warning model. It then uses the optimal early warning threshold and the mechanism early warning threshold to perform a fusion judgment to obtain the dynamic early warning threshold of the risk level and microplastic concentration of the target sea area for the target biological species.

Images