A collaborative analysis method for marine biodiversity big data based on federated knowledge graph

Abstract

This invention belongs to the field of data processing technology and proposes a collaborative analysis method for marine biodiversity big data based on a federated knowledge graph. Specifically, the method involves presetting marine risk types, identifying graph nodes from the federated knowledge graph, and obtaining inference confidence from each graph node. The method then uses the inference confidence obtained for each graph node within a unit time period to perform validity judgments. Finally, based on the validity judgment results, valid graph nodes are selected to obtain marine risk confidence. This method effectively utilizes confidence sequence comparisons to highlight the discontinuous and unstable characteristics of the prediction sequence structure caused by non-aligned data models, while also leveraging the prediction differences between different graph nodes.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and specifically relates to a collaborative analysis method for marine biodiversity big data based on a federated knowledge graph. Background Art

[0002] Marine biodiversity itself is highly sensitive and responsive to ecological disturbances, and has a mechanistic explanation function for various marine risks such as red tides, hypoxia, and species invasions from the perspective of the ecological chain. However, the collection of biodiversity data in the same region has the problems of collection difficulties and limited coverage of biological types. Although physical data monitoring can reduce the operational bottlenecks caused by the difficulties in biodiversity monitoring to a certain extent, the models constructed by data collection in a single area usually still have high deviation problems caused by insufficient original data. Big data collaborative analysis based on federated knowledge graphs can effectively integrate biodiversity data from different geographical regions and the inference models extracted from the database.

[0003] However, marine biodiversity data incorporates both measured and sampled data. Measured data monitors objective physical quantities and typically allows for more frequent, real-time measurements. Sampled data, on the other hand, refers to monitoring data from scientific research surveys. Due to its heavy reliance on manual labor or sophisticated equipment, monitoring intervals are long and data acquisition is infrequent. These measured and sampled data are simultaneously used in the collaborative analysis of big data within federated knowledge graphs. Compared to inference models constructed using more generalized data, this non-aligned data scenario is more prone to false short-term predictions. Furthermore, the sample density of measured data is much higher than that of sampled data, causing deep learning-based inference models to experience degraded generalization capabilities in sample-sparse intervals. This creates the risk of bias amplification in non-aligned data scenarios within the collaborative analysis of big data within federated knowledge graphs. Therefore, it is necessary to assess the effectiveness of collaboration between different graph nodes within the federated knowledge graph to prevent the risk of bias amplification. Summary of the Invention

[0004] The purpose of this invention is to propose a collaborative analysis method for marine biodiversity big data based on a federated knowledge graph to solve one or more technical problems existing in the prior art and at least provide a beneficial option or create conditions.

[0005] To achieve the above objectives, according to one aspect of the present invention, a method for collaborative analysis of marine biodiversity big data based on a federated knowledge graph is provided, the method comprising the following steps:

[0006] S100, presets the marine risk type and identifies the graph nodes from the federated knowledge graph;

[0007] S200, obtaining inference confidence from each graph node;

[0008] S300, performing validity judgment based on the inference confidence obtained from each graph node within a unit time period;

[0009] S400: Select a valid atlas node to obtain the ocean risk confidence level based on the validity judgment result.

[0010] Furthermore, in step S100, the marine risk type is preset, and the method for identifying the graph nodes from the federated knowledge graph is: the federated knowledge graph is composed of a number of graph nodes, and different graph nodes represent biodiversity monitoring databases in different geographical locations; the marine risk type is preset and the graph nodes with the risk type mark are screened.

[0011] The types of marine risks include red tides, invasive alien species, abnormal sea temperatures, ocean acidification, coral bleaching, or aquaculture leakage;

[0012] The core of the federated knowledge graph here is the collaborative reasoning and co-construction of the knowledge graph by multiple databases, because the construction of marine risk models based on a single data source has defects. First, it requires data annotation for risk types that have not been monitored until the time when the risk actually occurs, and there is an obvious shortcoming in the annotation opportunities. Second, the lack of annotation opportunities leads to insufficient robustness of the constructed reasoning model and low model accuracy.

[0013] During the pre-set marine risk type step, only one marine risk type is selected to ensure consistent inference targets for graph nodes, preventing the generalization and ambiguity of inference targets in subsequently constructed inference models, which can reduce accuracy. Each data record in the biodiversity monitoring database is timestamped, giving each feature data a temporal attribute. For any graph node, if a marine risk is determined to occur in the corresponding sea area during a certain time period, the graph node will be labeled with the marine risk during that time period. After pre-setting or selecting a unique marine risk type, graph nodes with the pre-set marine risk labeling conditions are selected from the federated knowledge graph, while graph nodes without the pre-set marine risk labeling are excluded.

[0014] Furthermore, in step S200, the method for obtaining reasoning confidence from each graph node includes: obtaining a scheduling feature list of a preset ocean risk at the current graph node, the scheduling feature list is a collection of features used to construct a prediction model, sending the scheduling feature list and the ocean risk type to each graph node for distributed modeling and obtaining a prediction model, and inputting the prediction model into local historical data to obtain the reasoning confidence at any time point.

[0015] Characteristic data include measured data and sampled data. Measured data refers to real-time monitoring data of objective physical quantities through sensors, such as temperature, pH, hyperspectral data, remote sensing data, or image data; sampled data refers to monitoring data that relies on scientific research surveys, such as species distribution, individual number, phytoplankton species and number, zooplankton species and number, etc.

[0016] In an embodiment, an example of a distributed modeling request is as follows: modeling request {risk label: {red tide}, scheduling feature set: {phytoplankton density, water temperature, pH, salinity, NO3, chlorophyll concentration}}.

[0017] The process of sending the scheduling feature list and ocean risk type to each graph node and building a prediction model is as follows: a unified request containing the scheduling feature list and ocean risk type is sent to each graph node through the FL-Coordinator federated controller or the SPARQL-Over-Federation federated coordination protocol. After receiving the request, any graph node instantiates a local model according to the risk type: extracts data corresponding to the features in the scheduling feature list in the local historical data as training data, and builds a local model with the ocean risk type as the prediction feature. The local model adopts any one of the decision tree model, random forest model or deep learning model, and returns the final model obtained by training as the prediction model to the current graph node that initiates the scheduling;

[0018] Decision tree models or random forest models are preferred here because they are more tolerant of modeling tasks with a small number of features or missing features. Therefore, they are very suitable for distributed modeling tasks with incomplete feature data types and scheduling feature lists. The more missing features, the more suitable. If the feature data type completely satisfies the scheduling feature list, deep learning is preferred because neural networks have strong fitting capabilities and support dynamic feature combinations, integrating related data to improve prediction accuracy.

[0019] Inference confidence refers to the confidence that the predicted feature is true, which represents the possibility of the current risk occurring.

[0020] Furthermore, in step S300, the method for determining the validity of the inference confidence obtained from each graph node within a unit time period is:

[0021] Assume a time period as the collection period Copt, Copt∈[0.5,2.5] years. During the collection period, the inference confidence is obtained every 5 to 10 natural days, and the time scale of obtaining the inference confidence is recorded as the measurement point;

[0022] During the acquisition period, the inference confidence of each measurement point constitutes an inference sequence;

[0023] Calculate the standard deviation of the inference sequence of each node and record it as a floating value. If the floating value of any node is greater than the upper quartile of the floating values ​​of all nodes, then mark the node as a high-noise node. Otherwise, mark it as a node to be estimated. The node validity of the high-noise node is false, and the node is unreliable.

[0024] The reasoning confidence of each node to be evaluated constitutes the benchmark sequence Rf.Ls;

[0025] The maximum value in the Rf.Ls sequence is recorded as the reference inflection point, and each measurement point between any reference inflection point and the first reference inflection point in the reverse time direction is recorded as a confidence analysis domain;

[0026] Among them, the measurement points at the beginning and end of the Rf.Ls sequence that are not covered by any confidence resolution domain are respectively assigned to the confidence resolution domain closest to their time.

[0027] In the Rf.Ls sequence, the robust reliability value of the current node is calculated based on the inter-domain change rate and inter-domain consistency of each confidence resolution domain;

[0028] Among them, the inter-domain change rate Rt.Cb, Rt.Cb of each confidence resolution domain is calculated k =max{((Cb k,i -Mea.Cb k-1 ) / Mea.Cb k-1 )}; where k is the serial number of each confidence resolution domain in the Rf.Ls sequence, i is the serial number of the element in each confidence resolution domain in the Rf.Ls sequence, Rt.Cb k is the inter-domain change rate of the kth confidence resolution domain, Cb k,i is the i-th element in the k-th confidence resolution domain, Mea.Cb k-1 is the average value of the k-1th confidence resolution domain, and max{} is the function for obtaining the maximum value;

[0029] Calculate the inter-domain consistency Ovr.ap, Ovr.ap of each confidence resolution domain k =|CI k ∩CI k-1 | / min(|CI k |,|CI k-1 |); among them, Ovr.ap k is the inter-domain consistency of the kth confidence resolution domain, CI k is the interval of the inference confidence of the k-th confidence resolution domain, and min{} is the function for obtaining the minimum value;

[0030] Among them, the previous confidence resolution domain of the first confidence resolution domain is the last confidence resolution domain in the Rf.Ls sequence;

[0031] The robust reliability value of the current node is calculated based on the inter-domain change rate and inter-domain consistency of each confidence resolution domain. The specific implementation formula is mean{(1-|Rt.Cb k |)+ Ovr.ap k )}; where mean{} is the mean value function;

[0032] The inter-domain change rate reflects the degree of change in a node within each confidence resolution domain, measuring the degree to which the current confidence resolution domain deviates from its historical trend. If the inter-domain change rate is generally large, it indicates that the node's inference confidence fluctuates significantly, and the node is at risk of being unreliable. The inter-domain consistency measures the consistency of the inference confidence time series over different time periods. If the inference confidence intervals of two inference confidence intervals overlap significantly, it indicates that the node's inference confidence changes steadily. The robust reliability value reflects the stability of the node. The smaller the inter-domain change rate, the more stable the node; the greater the inter-domain consistency, the more stable the node; and the higher the overall robust reliability value, the more reliable the node's inference confidence.

[0033] Obtain the difference between the mean and standard deviation of the robust reliability values ​​of all nodes to be evaluated, and record it as the reliability limit. If the robust reliability value of any node to be evaluated is greater than the reliability limit, the node validity of the node to be evaluated is true and the node is reliable; otherwise, the node validity is false and the node is unreliable.

[0034] Since the validity judgment of each node requires processing the confidence resolution domain, it can effectively quantify the fluctuation of the inference confidence of the marine biodiversity nodes, identify potential abnormal nodes in advance, and avoid the risk of inference deviations being amplified during the collaborative analysis process, thereby reducing the interference of abnormal nodes on the overall federated knowledge graph inference results. However, the acquisition of robust and reliable values ​​is overly dependent on the local fluctuation characteristics of the inference confidence changes, which easily leads to the sensitivity of node stability assessment to sample density, and then there is a significant deviation between the obtained node validity and the actual stability level, resulting in decision-making errors, especially in abnormally dense periods, making the problem more prominent. However, the existing technology cannot effectively compensate for the phenomenon of stability misjudgment caused by sample sparsity and non-aligned data. In order to eliminate this influence, the present invention proposes a more preferred solution as follows:

[0035] Furthermore, in step S300, the method for determining the validity of the inference confidence obtained from each graph node within a unit time period is:

[0036] Assume a time period as the collection period Copt, Copt∈[2,3] years. During the collection period, the inference confidence is obtained every 5 to 10 natural days, and the time scale of obtaining the inference confidence is recorded as the measurement point;

[0037] During the acquisition period, the inference confidence of each measurement point constitutes the inference sequence ER.Ls;

[0038] The Fourier transform algorithm is used on the inference sequence to obtain the energy spectrum density after Fourier transform and form a fluctuating energy sequence. Among them, in the result of the Fourier transform algorithm, the frequency is arranged from low to high, and the corresponding energy spectrum density is also arranged from small to large according to the frequency.

[0039] The elements in the second half of the fluctuation energy sequence are summed and recorded as the high-frequency energy value, and the ratio of the high-frequency energy value to the sum of all elements in the fluctuation energy sequence is recorded as the fluctuation proportion Fpo;

[0040] Because measured data in marine biodiversity monitoring is frequent and sampled data is sparse, fusion modeling in federated knowledge graphs can produce excessive short-term fluctuations in inference. These fluctuations do not reflect actual ecological disturbances, but rather "false signals" caused by data gaps or inadequate model fit. Therefore, we introduce the "fluctuation ratio" as a metric to determine the "short-term instability" of a node. Essentially, it characterizes the proportion of "short-term abnormal fluctuations caused by non-ecological disturbance factors" in the node's inference sequence from a frequency domain perspective.

[0041] By performing Fourier transform on the inference sequence of each node and converting it into the frequency domain, we can obtain a complex number sequence in the frequency domain. The square of the modulus of the complex number represents the energy of each frequency component. To judge the fluctuation of inference confidence, we mainly observe the distribution of energy in frequency. If the energy is concentrated in low frequency, it means that the overall signal is stable. If the energy is distributed in high frequency, it means that there is significant fluctuation in the signal, and the reliability of the node will be affected by its fluctuation. Then, the fluctuation ratio of each node is used to screen out nodes with severe fluctuations. A high fluctuation ratio means that the inference result of the node has severe, unstable, and short-term impact-like jumps in time series, and often does not have the stable response characteristics at the ecosystem level. Its credibility is low. Special processing of the screened nodes can ensure the reasoning stability and robustness of the collaborative modeling of the federated knowledge graph in a "cross-regional, heterogeneous, and non-aligned data environment."

[0042] If the fluctuation ratio of any node is greater than the upper quartile of the fluctuation ratio of all nodes, the node is marked as a post-node, otherwise it is marked as a pre-node;

[0043] According to the inference confidence, the stationary level of each node is obtained;

[0044] The specific implementation formula of the stationary level is mean(DVF(ER.Ls)) / max(DVF(ER.Ls)); the DVF(ER.Ls) function is a set of absolute values ​​of the difference between each element in the ER.Ls sequence and the previous element, mean() and max() are functions for calculating the average value and obtaining the maximum value, respectively;

[0045] The inference sequence is used to calculate the Pearson correlation coefficient between any subsequent node and all preceding nodes. If there is a preceding node with a Pearson correlation coefficient greater than 0.7, the stationary level of the preceding node is updated to the minimum value between its original stationary level and the stationary level of the subsequent node.

[0046] Among them, if there are multiple predecessor nodes with Pearson correlation coefficients greater than 0.7, the plateau level of the predecessor node corresponding to the maximum value of the Pearson correlation coefficient is updated; a Pearson correlation coefficient greater than 0.7 indicates that there is a strong positive correlation between the two nodes.

[0047] Obtain the predecessor node that is closest in value to the average value of the stationary levels of all predecessor nodes and record it as the credible predecessor node;

[0048] The node validity of the trusted predecessor node is true, and the node is reliable;

[0049] Calculate the volatility risk Faus of other nodes based on the trusted front-end nodes:

[0050] ;

[0051] Where j1 is an accumulated variable, Hqq is the number of measurement points obtained during the acquisition period, ER.Ls(j1) and MER.Ls(j1) are the inference confidence of the j1-th measurement point in the inference sequence of the current node and the trusted predecessor node, respectively, MER and P.MER are the stationary levels of the current node and the trusted predecessor node, respectively, and exp() is an exponential function with the natural constant e as the base;

[0052] If the volatility risk of any node is less than the 80% quantile of the volatility risk of all nodes, the node validity of the node is true and the node is reliable; otherwise, the node validity is false and the node is unreliable.

[0053] Beneficial effects: Since the validity judgment process is fed back by each graph node, the inference confidence of the local data is not predicted using the prediction result of only one time point, but rather taking multiple inference confidences of consecutive time periods for overall decision-making. Therefore, it can effectively utilize confidence sequence comparison and highlight the discontinuous and unstable characteristics of the prediction sequence structure brought by the non-aligned data model, while also utilizing the prediction differences of different graph nodes.

[0054] Furthermore, in step S400, a method for selecting a valid graph node to obtain the ocean risk confidence level according to the validity judgment result is:

[0055] The graph nodes with true validity of each node are obtained as valid nodes, and each valid node is voted on the occurrence of ocean risks at the current location. The results with a rate higher than 50% are obtained and sent to the client. If it is higher than 50%, the risk is warned to occur, otherwise the risk does not occur.

[0056] Preferably, all undefined variables in the present invention, if not clearly defined, can be manually set thresholds.

[0057] The present invention also provides a marine biodiversity big data collaborative analysis system based on a federated knowledge graph. The marine biodiversity big data collaborative analysis system based on a federated knowledge graph includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the marine biodiversity big data collaborative analysis method based on the federated knowledge graph are implemented. The marine biodiversity big data collaborative analysis system based on the federated knowledge graph can be run on computing devices such as desktop computers, laptops, PDAs, and cloud data centers. The executable system may include, but is not limited to, a processor, a memory, and a server cluster. The processor executes the computer program and runs in a unit of the system.

[0058] The beneficial effects of the present invention are: since the validity judgment process is fed back by each graph node, the inference confidence of the predicted local data is not predicted using only the prediction result of one time point, but a comprehensive decision is made based on multiple inference confidences of consecutive time periods. Therefore, it can effectively utilize the confidence sequence comparison and highlight the discontinuous and unstable characteristics of the prediction sequence structure brought by the non-aligned data model, while also utilizing the prediction differences of different graph nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The above and other features of the present invention will become more apparent through a detailed description of the embodiments shown in conjunction with the accompanying drawings. In the drawings of the present invention, the same reference numerals represent the same or similar elements. Obviously, the drawings described below are only some embodiments of the present invention. It is possible for a person skilled in the art to derive other drawings based on these drawings without inventive effort. In the drawings:

[0060] Figure 1 Shown is a flow chart of the collaborative analysis method of marine biodiversity big data based on federated knowledge graph;

[0061] Figure 2Shown is the structural diagram of the marine biodiversity big data collaborative analysis system based on the federated knowledge graph. DETAILED DESCRIPTION

[0062] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of this application can be combined with each other unless there is a conflict.

[0063] like Figure 1 The following is a flowchart of the collaborative analysis method of marine biodiversity big data based on the federated knowledge graph. Figure 1 To illustrate a method for collaborative analysis of marine biodiversity big data based on a federated knowledge graph according to an embodiment of the present invention, the method comprises the following steps:

[0064] S100, presets the marine risk type and identifies the graph nodes from the federated knowledge graph;

[0065] S200, obtaining inference confidence from each graph node;

[0066] S300, performing validity judgment based on the inference confidence obtained from each graph node within a unit time period;

[0067] S400: Select a valid atlas node to obtain the ocean risk confidence level based on the validity judgment result.

[0068] Furthermore, in step S100, the marine risk type is preset, and the method for identifying the graph nodes from the federated knowledge graph is: the federated knowledge graph is composed of a number of graph nodes, and different graph nodes represent biodiversity monitoring databases in different geographical locations; the marine risk type is preset and the graph nodes with the risk type mark are screened.

[0069] The types of marine risks include red tides, invasive alien species, abnormal sea temperatures, ocean acidification, coral bleaching, or aquaculture leakage;

[0070] The core of the federated knowledge graph here is the collaborative reasoning and co-construction of the knowledge graph by multiple databases, because the construction of marine risk models based on a single data source has defects. First, it requires data annotation for risk types that have not been monitored until the time when the risk actually occurs, and there is an obvious shortcoming in the annotation opportunities. Second, the lack of annotation opportunities leads to insufficient robustness of the constructed reasoning model and low model accuracy.

[0071] During the pre-set marine risk type step, only one marine risk type is selected to ensure consistent inference targets for graph nodes, preventing the generalization and ambiguity of inference targets in subsequently constructed inference models, which can reduce accuracy. Each data record in the biodiversity monitoring database is timestamped, giving each feature data a temporal attribute. For any graph node, if a marine risk is determined to occur in the corresponding sea area during a certain time period, the graph node will be labeled with the marine risk during that time period. After pre-setting or selecting a unique marine risk type, graph nodes with the pre-set marine risk labeling conditions are selected from the federated knowledge graph, while graph nodes without the pre-set marine risk labeling are excluded.

[0072] Furthermore, in step S200, the method for obtaining reasoning confidence from each graph node includes: obtaining a scheduling feature list of a preset ocean risk at the current graph node, the scheduling feature list is a collection of features used to construct a prediction model, sending the scheduling feature list and the ocean risk type to each graph node for distributed modeling and obtaining a prediction model, and inputting the prediction model into local historical data to obtain the reasoning confidence at any time point.

[0073] Characteristic data include measured data and sampled data. Measured data refers to real-time monitoring data of objective physical quantities through sensors, such as temperature, pH, hyperspectral data, remote sensing data, or image data; sampled data refers to monitoring data that relies on scientific research surveys, such as species distribution, individual number, phytoplankton species and number, zooplankton species and number, etc.

[0074] In an embodiment, an example of a distributed modeling request is as follows: modeling request {risk label: {red tide}, scheduling feature set: {phytoplankton density, water temperature, pH, salinity, NO3, chlorophyll concentration}}.

[0075] The process of sending the scheduling feature list and ocean risk type to each graph node and building a prediction model is as follows: a unified request containing the scheduling feature list and ocean risk type is sent to each graph node through the FL-Coordinator federated controller or the SPARQL-Over-Federation federated coordination protocol. After receiving the request, any graph node instantiates a local model according to the risk type: extracts data corresponding to the features in the scheduling feature list in the local historical data as training data, and builds a local model with the ocean risk type as the prediction feature. The local model adopts any one of the decision tree model, random forest model or deep learning model, and returns the final model obtained by training as the prediction model to the current graph node that initiates the scheduling;

[0076] Decision tree models or random forest models are preferred here because they are more tolerant of modeling tasks with a small number of features or missing features. Therefore, they are very suitable for distributed modeling tasks with incomplete feature data types and scheduling feature lists. The more missing features, the more suitable. If the feature data type completely satisfies the scheduling feature list, deep learning is preferred because neural networks have strong fitting capabilities and support dynamic feature combinations, integrating related data to improve prediction accuracy.

[0077] Inference confidence refers to the confidence that the predicted feature is true, which represents the possibility of the current risk occurring.

[0078] Furthermore, in step S300, the method for determining the validity of the inference confidence obtained from each graph node within a unit time period is:

[0079] Assume a time period as the collection period Copt, Copt∈[0.5,2.5] years. During the collection period, the inference confidence is obtained every 5 to 10 natural days, and the time scale of obtaining the inference confidence is recorded as the measurement point;

[0080] During the acquisition period, the inference confidence of each measurement point constitutes an inference sequence;

[0081] Calculate the standard deviation of the inference sequence of each node and record it as a floating value. If the floating value of any node is greater than the upper quartile of the floating values ​​of all nodes, then mark the node as a high-noise node. Otherwise, mark it as a node to be estimated. The node validity of the high-noise node is false, and the node is unreliable.

[0082] The reasoning confidence of each node to be evaluated constitutes the benchmark sequence Rf.Ls;

[0083] The maximum value in the Rf.Ls sequence is recorded as the reference inflection point, and each measurement point between any reference inflection point and the first reference inflection point in the reverse time direction is recorded as a confidence analysis domain;

[0084] Among them, the measurement points at the beginning and end of the Rf.Ls sequence that are not covered by any confidence resolution domain are respectively assigned to the confidence resolution domain closest to their time.

[0085] In the Rf.Ls sequence, the robust reliability value of the current node is calculated based on the inter-domain change rate and inter-domain consistency of each confidence resolution domain;

[0086] Among them, the inter-domain change rate Rt.Cb, Rt.Cb of each confidence resolution domain is calculated k =max{((Cb k,i -Mea.Cb k-1 ) / Mea.Cb k-1)}; where k is the serial number of each confidence resolution domain in the Rf.Ls sequence, i is the serial number of the element in each confidence resolution domain in the Rf.Ls sequence, Rt.Cb k is the inter-domain change rate of the kth confidence resolution domain, Cb k,i is the i-th element in the k-th confidence resolution domain, Mea.Cb k-1 is the average value of the k-1th confidence resolution domain, and max{} is the function for obtaining the maximum value;

[0087] Calculate the inter-domain consistency Ovr.ap, Ovr.ap of each confidence resolution domain k =|CI k ∩CI k-1 | / min(|CI k |,|CI k-1 |); among them, Ovr.ap k is the inter-domain consistency of the kth confidence resolution domain, CI k is the interval of the inference confidence of the k-th confidence resolution domain, and min{} is the function for obtaining the minimum value;

[0088] Among them, the previous confidence resolution domain of the first confidence resolution domain is the last confidence resolution domain in the Rf.Ls sequence;

[0089] The robust reliability value of the current node is calculated based on the inter-domain change rate and inter-domain consistency of each confidence resolution domain. The specific implementation formula is mean{(1-|Rt.Cb k |)+ Ovr.ap k )}; where mean{} is the mean value function;

[0090] The inter-domain change rate reflects the degree of change in a node within each confidence resolution domain, measuring the degree to which the current confidence resolution domain deviates from its historical trend. If the inter-domain change rate is generally large, it indicates that the node's inference confidence fluctuates significantly, and the node is at risk of being unreliable. The inter-domain consistency measures the consistency of the inference confidence time series over different time periods. If the inference confidence intervals of two inference confidence intervals overlap significantly, it indicates that the node's inference confidence changes steadily. The robust reliability value reflects the stability of the node. The smaller the inter-domain change rate, the more stable the node; the greater the inter-domain consistency, the more stable the node; and the higher the overall robust reliability value, the more reliable the node's inference confidence.

[0091] Obtain the difference between the mean and standard deviation of the robust reliability values ​​of all nodes to be evaluated, and record it as the reliability limit. If the robust reliability value of any node to be evaluated is greater than the reliability limit, the node validity of the node to be evaluated is true and the node is reliable; otherwise, the node validity is false and the node is unreliable.

[0092] Since the validity judgment of each node requires processing the confidence resolution domain, it can effectively quantify the fluctuation of the inference confidence of the marine biodiversity nodes, identify potential abnormal nodes in advance, and avoid the risk of inference deviations being amplified during the collaborative analysis process, thereby reducing the interference of abnormal nodes on the overall federated knowledge graph inference results. However, the acquisition of robust and reliable values ​​is overly dependent on the local fluctuation characteristics of the inference confidence changes, which easily leads to the sensitivity of node stability assessment to sample density, and then there is a significant deviation between the obtained node validity and the actual stability level, resulting in decision-making errors, especially in abnormally dense periods, making the problem more prominent. However, the existing technology cannot effectively compensate for the phenomenon of stability misjudgment caused by sample sparsity and non-aligned data. In order to eliminate this influence, the present invention proposes a more preferred solution as follows:

[0093] Furthermore, in step S300, the method for determining the validity of the inference confidence obtained from each graph node within a unit time period is:

[0094] Assume a time period as the collection period Copt, Copt∈[2,3] years. During the collection period, the inference confidence is obtained every 5 to 10 natural days, and the time scale of obtaining the inference confidence is recorded as the measurement point;

[0095] During the acquisition period, the inference confidence of each measurement point constitutes the inference sequence ER.Ls;

[0096] The Fourier transform algorithm is used on the inference sequence to obtain the energy spectrum density after Fourier transform and form a fluctuating energy sequence. Among them, in the result of the Fourier transform algorithm, the frequency is arranged from low to high, and the corresponding energy spectrum density is also arranged from small to large according to the frequency.

[0097] The elements in the second half of the fluctuation energy sequence are summed and recorded as the high-frequency energy value, and the ratio of the high-frequency energy value to the sum of all elements in the fluctuation energy sequence is recorded as the fluctuation proportion Fpo;

[0098] Because measured data in marine biodiversity monitoring is frequent and sampled data is sparse, fusion modeling in federated knowledge graphs can produce excessive short-term fluctuations in inference. These fluctuations do not reflect actual ecological disturbances, but rather "false signals" caused by data gaps or inadequate model fit. Therefore, we introduce the "fluctuation ratio" as a metric to determine the "short-term instability" of a node. Essentially, it characterizes the proportion of "short-term abnormal fluctuations caused by non-ecological disturbance factors" in the node's inference sequence from a frequency domain perspective.

[0099] By performing Fourier transform on the inference sequence of each node and converting it into the frequency domain, we can obtain a complex number sequence in the frequency domain. The square of the modulus of the complex number represents the energy of each frequency component. To judge the fluctuation of inference confidence, we mainly observe the distribution of energy in frequency. If the energy is concentrated in low frequency, it means that the overall signal is stable. If the energy is distributed in high frequency, it means that there is significant fluctuation in the signal, and the reliability of the node will be affected by its fluctuation. Then, the fluctuation ratio of each node is used to screen out nodes with severe fluctuations. A high fluctuation ratio means that the inference result of the node has severe, unstable, and short-term impact-like jumps in time series, and often does not have the stable response characteristics at the ecosystem level. Its credibility is low. Special processing of the screened nodes can ensure the reasoning stability and robustness of the collaborative modeling of the federated knowledge graph in a "cross-regional, heterogeneous, and non-aligned data environment."

[0100] If the fluctuation ratio of any node is greater than the upper quartile of the fluctuation ratio of all nodes, the node is marked as a post-node, otherwise it is marked as a pre-node;

[0101] According to the inference confidence, the stationary level of each node is obtained;

[0102] The specific implementation formula of the stationary level is mean(DVF(ER.Ls)) / max(DVF(ER.Ls)); the DVF(ER.Ls) function is a set of absolute values ​​of the difference between each element in the ER.Ls sequence and the previous element, mean() and max() are functions for calculating the average value and obtaining the maximum value, respectively;

[0103] The inference sequence is used to calculate the Pearson correlation coefficient between any subsequent node and all preceding nodes. If there is a preceding node with a Pearson correlation coefficient greater than 0.7, the stationary level of the preceding node is updated to the minimum value between its original stationary level and the stationary level of the subsequent node.

[0104] Among them, if there are multiple predecessor nodes with Pearson correlation coefficients greater than 0.7, the plateau level of the predecessor node corresponding to the maximum value of the Pearson correlation coefficient is updated; a Pearson correlation coefficient greater than 0.7 indicates that there is a strong positive correlation between the two nodes.

[0105] Obtain the predecessor node that is closest in value to the average value of the stationary levels of all predecessor nodes and record it as the credible predecessor node;

[0106] The node validity of the trusted predecessor node is true, and the node is reliable;

[0107] Calculate the volatility risk Faus of other nodes based on the trusted front-end nodes:

[0108] ;

[0109] Where j1 is an accumulated variable, Hqq is the number of measurement points obtained during the acquisition period, ER.Ls(j1) and MER.Ls(j1) are the inference confidence of the j1-th measurement point in the inference sequence of the current node and the trusted predecessor node, respectively, MER and P.MER are the stationary levels of the current node and the trusted predecessor node, respectively, and exp() is an exponential function with the natural constant e as the base;

[0110] If the volatility risk of any node is less than the 80% quantile of the volatility risk of all nodes, the node validity of the node is true and the node is reliable; otherwise, the node validity is false and the node is unreliable.

[0111] Beneficial effects: Since the validity judgment process is fed back by each graph node, the inference confidence of the local data is not predicted using the prediction result of only one time point, but rather taking multiple inference confidences of consecutive time periods for overall decision-making. Therefore, it can effectively utilize confidence sequence comparison and highlight the discontinuous and unstable characteristics of the prediction sequence structure brought by the non-aligned data model, while also utilizing the prediction differences of different graph nodes.

[0112] Furthermore, in step S400, a method for selecting a valid graph node to obtain the ocean risk confidence level according to the validity judgment result is:

[0113] The graph nodes with true validity of each node are obtained as valid nodes, and each valid node is voted on the occurrence of ocean risks at the current location. The results with a rate higher than 50% are obtained and sent to the client. If it is higher than 50%, the risk is warned to occur, otherwise the risk does not occur.

[0114] The embodiment of the present invention provides a collaborative analysis system for marine biodiversity big data based on a federated knowledge graph, such as Figure 2 The figure shows a structural diagram of the marine biodiversity big data collaborative analysis system based on a federated knowledge graph of the present invention. The marine biodiversity big data collaborative analysis system based on a federated knowledge graph of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned embodiment of the marine biodiversity big data collaborative analysis method based on a federated knowledge graph are implemented.

[0115] The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to run in a unit of the system.

[0116] The marine biodiversity big data collaborative analysis system based on the federated knowledge graph can be run on computing devices such as desktop computers, laptops, PDAs, and cloud servers. The marine biodiversity big data collaborative analysis system based on the federated knowledge graph can be operated on systems that may include, but are not limited to, processors and memories. Those skilled in the art will understand that the examples are merely examples of marine biodiversity big data collaborative analysis systems based on the federated knowledge graph and do not constitute a limitation on the marine biodiversity big data collaborative analysis system based on the federated knowledge graph. The system may include more or fewer components than the example, or a combination of certain components, or different components. For example, the marine biodiversity big data collaborative analysis system based on the federated knowledge graph may also include input and output devices, network access devices, buses, and the like.

[0117] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor serves as the control center of the operation system of the marine biodiversity big data collaborative analysis system based on the federated knowledge graph, and utilizes various interfaces and lines to connect the various parts of the entire marine biodiversity big data collaborative analysis system based on the federated knowledge graph.

[0118] The memory can be used to store the computer programs and / or modules. The processor implements the various functions of the federated knowledge graph-based marine biodiversity big data collaborative analysis system by running or executing the computer programs and / or modules stored in the memory and accessing the data stored in the memory. The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function or an image playback function); the data storage area may store data generated based on the use of the mobile phone (such as audio data and a phone book). Furthermore, the memory may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0119] Although the present invention has been described in considerable detail and with particularity with respect to several embodiments, it is not intended to limit the present invention to any of these details or embodiments or any particular embodiment, so as to effectively encompass the intended scope of the present invention. In addition, the present invention has been described above with respect to embodiments foreseen by the inventors for the purpose of providing a useful description, and those insubstantial modifications of the present invention that are not currently foreseen may still represent equivalent modifications of the present invention.

Claims

1. A collaborative analysis method for marine biodiversity big data based on a federated knowledge graph, characterized by: The method comprises the following steps: S100, presets the marine risk type and identifies the graph nodes from the federated knowledge graph; S200, obtaining inference confidence from each graph node; S300, performing validity judgment based on the inference confidence obtained from each graph node within a unit time period; S400, identifying marine risks through atlas nodes with true validity judgment results; In step S300, the method for judging the validity of the inference confidence obtained from each graph node within a unit time period is as follows: set a time period as a collection period, obtain the inference confidence at equal intervals within the collection period, and record the time when the inference confidence is obtained as a measurement point; The inference confidence of all measurement points in the acquisition period is made into an inference sequence, and the standard deviation of the inference sequence is calculated and recorded as a floating value. If the floating value of any node is greater than the upper quartile of the floating values ​​of all nodes, the node is marked as a high-noise node, otherwise it is marked as a node to be estimated, and the node validity of the high-noise node is marked as false. The inference sequence of the node to be estimated is used as the benchmark sequence, the maximum value in the benchmark sequence is recorded as the benchmark inflection point, and the period between any benchmark inflection point and the first benchmark inflection point in the reverse time direction is recorded as a confidence resolution domain; In the benchmark sequence, the graph confidence structure value of the current node is calculated based on the inter-domain change rate and inter-domain consistency of each confidence resolution domain; If the graph confidence structure value of any node to be estimated is greater than the reliability limit, the node validity of the node to be estimated is true, otherwise the node validity is false.

2. The method for collaborative analysis of marine biodiversity big data based on federated knowledge graph according to claim 1 is characterized in that: In step S100, the marine risk type is preset, and a method for identifying a graph node from a federated knowledge graph is as follows: the federated knowledge graph is composed of a plurality of graph nodes, and different graph nodes represent biodiversity monitoring databases at different geographical locations; Preset ocean risk types and filter graph nodes marked with this ocean risk type.

3. The method for collaborative analysis of marine biodiversity big data based on federated knowledge graph according to claim 1, characterized in that: In step S200, the method for obtaining reasoning confidence from each graph node includes: obtaining a scheduling feature list of preset ocean risks at the current graph node, the scheduling feature list is a collection of features used to build a prediction model, sending the scheduling feature list and ocean risk type to each graph node for distributed modeling and obtaining a prediction model, and inputting the prediction model into local historical data to obtain the reasoning confidence at any time point.

4. The method for collaborative analysis of marine biodiversity big data based on federated knowledge graph according to claim 1, characterized in that: In step S400, the method for identifying marine risks through the graph nodes with true validity judgment results is as follows: the historical scheduling feature data of the current graph node is input into the prediction model of each graph node with true validity, and the cross-reasoning confidence of each graph node to the current graph node is obtained; the cross-reasoning confidence of each credible graph node is weighted and fused with its ecological coupling degree to obtain the federated reasoning confidence of the current graph node; If the confidence of the federated reasoning exceeds the preset risk threshold, it is determined that the current graph node has the ocean risk.

5. A collaborative analysis system of marine biodiversity big data based on a federated knowledge graph, characterized by: The marine biodiversity big data collaborative analysis system based on the federated knowledge graph includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the marine biodiversity big data collaborative analysis method based on the federated knowledge graph described in any one of claims 1 to 4 are implemented. The marine biodiversity big data collaborative analysis system based on the federated knowledge graph runs on desktop computers, laptop computers, PDAs, and computing devices in cloud data centers.

Images