Big data-based fishery digital platform management system and method

Abstract

The present application relates to the technical field of fishery big data management, in particular to a fishery digital platform management system and method based on big data, comprising: collecting a plurality of fishery monitoring source heterogeneous data sets, using an improved spatiotemporal clustering algorithm introducing ship behavior patterns and marine geographic fence constraints to process fishing boat track point data, identifying fishing boat operation modes and suspected illegal operation areas. Multidimensional time series correlation analysis is carried out on fishing ground environmental sensing data, key environmental factor sequences are mined, and fishing catch transaction flow data is subjected to feature mining to generate price fluctuation and supply and demand relationship features. A fishery knowledge graph is constructed by fusing multiple types of features, a multidimensional comprehensive management plan is generated through a fishery management strategy model, and the plan is disassembled into digital instructions and distributed to corresponding business terminals. The method realizes the integration and analysis of multi-source fishery heterogeneous data, and improves the collaborative ability of fishing boat supervision, environmental analysis and market analysis.

Description

Technical Field

[0001] This invention relates to the field of fishery big data management technology, and in particular to a fishery digital platform management system and method based on big data. Background Technology

[0002] The fisheries industry suffers from fragmented data sources, encompassing diverse and heterogeneous data such as fishing vessel trajectories, fishing ground environmental sensors, catch transaction records, and aquaculture area video surveillance. Traditional fisheries management models employ independent storage and separate analysis for each type of monitoring data, lacking a unified integration mechanism. This results in fragmented data resources, hindering the formation of a comprehensive fisheries data system. Furthermore, fishing vessel trajectory data analysis often relies on common clustering algorithms. These algorithms lack constraints related to vessel behavior characteristics and marine geographic boundaries, making it difficult to differentiate operational patterns and define the scope of illegal activities from massive datasets.

[0003] Fishing ground environmental data mostly remains at the level of simple time-series change statistics, without conducting multi-dimensional time-series correlation mining, making it impossible to screen out environmental factors that affect fishery output. Fish catch transaction data only records basic information and lacks the ability to deeply mine market price fluctuations and supply and demand characteristics, so the potential laws governing industry operation cannot be effectively extracted.

[0004] The existing fisheries management system lacks a method for integrating and modeling multi-dimensional fisheries characteristic information, making it impossible to build a knowledge system framework capable of supporting the relationships between multiple elements. Management strategy formulation relies heavily on manual experience and cannot generate systematic management solutions based on the correlation results of multi-source data. Management instructions also cannot be broken down according to business levels and accurately distributed to fishing vessel terminals, fishing port supervision terminals, and market announcement terminals, resulting in significant shortcomings in the industry's comprehensive digital management and control capabilities. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a big data-based digital platform management system and method for fisheries.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a fisheries digital platform management method based on big data, comprising: Acquire heterogeneous fishery data sets from multiple fishery monitoring sources, including fishing vessel trajectory point data, fishing ground environmental sensor data, catch transaction flow data, and aquaculture area video surveillance data; An improved spatiotemporal clustering algorithm is applied to the fishing vessel trajectory point data to identify fishing vessel operation patterns and suspected illegal operation areas. The improved spatiotemporal clustering algorithm constrains the clustering process based on vessel behavior patterns and marine geofence information. Multidimensional time-series correlation analysis was performed on the environmental sensor data of the fishing grounds to extract the sequence of key environmental factors affecting the catch. Market dynamic feature mining was performed on the transaction flow data of the catch to generate price fluctuation patterns and supply and demand relationship features. By integrating the aforementioned fishing vessel operation modes, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics, a fisheries knowledge graph for comprehensive analysis is constructed. Based on the fisheries knowledge graph, a comprehensive management plan for fishing vessel production, resource conservation, and market regulation is generated by calling a preset fisheries management strategy model. The comprehensive management plan is broken down into executable digital instructions and distributed to the corresponding fishing vessel terminals, fishing port supervision terminals, and market announcement terminals.

[0007] As a further aspect of the present invention, an improved spatiotemporal clustering algorithm is applied to the fishing vessel trajectory point data to identify fishing vessel operation patterns and suspected illegal operation areas, including: The fishing vessel trajectory point data is cleaned and interpolated to form a continuous spatiotemporal trajectory sequence of fishing vessels; The improved spatiotemporal clustering algorithm is used to analyze the spatiotemporal trajectory sequence of the fishing vessel. When calculating the similarity between trajectory points, the improved spatiotemporal clustering algorithm comprehensively considers spatial distance, temporal proximity, heading speed similarity, and whether it is located within a preset no-fishing zone or protected area geographical fence. Based on the similarity, density clustering is performed on the spatiotemporal trajectory points of fishing vessels. Trajectory points with high spatiotemporal density and similar behavioral patterns are aggregated into clusters. Each cluster represents a stable fishing vessel operation mode, including: trawling, purse seine, or transit. Clusters of trajectory points that are located within or at the edge of the geographical fence of a fishing ban area or protected area and whose behavior patterns do not conform to routine operations or navigation are identified and marked as suspected illegal operation areas.

[0008] As a further aspect of the present invention, the improved spatiotemporal clustering algorithm constrains the clustering process based on ship behavior patterns and marine geographic fence information, including: When calculating the similarity between trajectory points, a behavioral pattern similarity component is introduced. The behavioral pattern similarity is calculated by comparing the similarity of the local trajectories around the two points in terms of heading and velocity distribution. Introduce a geofence constraint component to determine whether two trajectory points are both within a specific marine geofence, or within fences with opposing properties. The behavioral pattern similarity component, the geofence constraint component, and the basic spatiotemporal distance component are weighted and fused to form a comprehensive similarity metric, which is used in the subsequent density clustering process. During the clustering result evaluation phase, for each identified cluster, the consistency of the behavior patterns of its internal trajectory points and the consistency of the geofence attributes are checked. Clusters with consistency below the threshold are split or removed to ensure the accuracy of the identification of the fishing vessel operation patterns and suspected illegal operation areas.

[0009] As a further aspect of the present invention, multidimensional time-series correlation analysis is performed on the fishery environment sensing data to extract key environmental factor sequences affecting catch yield, including: The environmental sensing data of the fishing grounds includes time-series measurements of water temperature, salinity, chlorophyll concentration, dissolved oxygen, and ocean current speed. The time-series measurements of various environmental factors are standardized and outlier removal is performed to obtain regularized time-series data of environmental factors; Calculate the cross-correlation between time series data of different environmental factors, and analyze the time-lag correlation between each environmental factor and historical catch data of the same period. Environmental factors that show correlation with catch data at a specific time lag are selected as candidate key environmental factors; Principal component analysis is performed on the time series data of the candidate key environmental factors to extract a few principal components that can explain most of the data variance. The time series variation sequence of the principal components is then used as the key environmental factor sequence.

[0010] As a further aspect of the present invention, market dynamic feature mining is performed on the fishery transaction data to generate price fluctuation patterns and supply-demand relationship characteristics, including: The transaction data of the fish catch is analyzed to extract the species, specifications, transaction time, transaction price, transaction quantity and buyer and seller information for each transaction; The transaction data is aggregated according to variety, specification and preset time window to generate price time series and transaction volume time series with different time granularities; The price time series is decomposed to separate long-term trend, seasonal cycle, cyclical fluctuation and irregular residual components, and the price fluctuation pattern is extracted from the seasonal cycle and cyclical fluctuation. The leading-lag relationship between the trading volume time series and the corresponding price time series is analyzed, and market concentration is analyzed in combination with buyer and seller information, thereby generating the supply and demand relationship characteristics that reflect the real-time market conditions and potential changes.

[0011] As a further aspect of the present invention, a fisheries knowledge graph for comprehensive analysis is constructed by integrating the aforementioned fishing vessel operation patterns, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics, including: Define the entity types and relationship types of the fisheries knowledge graph. Entity types include fishing vessels, operation modes, sea areas, environmental factors, catch species, and market price patterns. Relationship types include engaging in, being located in, being affected by, being associated with, and being manifested as. The fishing vessel operation mode is instantiated as an operation mode entity and connected to the specific fishing vessel entity through the operation relationship. Instantiate the suspected illegal operation area as a sea area entity with illegal attributes; Each principal component of the key environmental factor sequence is treated as an environmental factor entity and connected to the relevant marine area entities through location or influence relationships. The price fluctuation pattern and supply and demand characteristics are structured, instantiated into a market price pattern entity, and connected with specific fish catch entities through association relationships. By utilizing graph embedding technology, all entities and relationships are represented and learned in a low-dimensional vector space to form the fisheries knowledge graph containing rich semantic and relational information.

[0012] As a further aspect of the present invention, based on the aforementioned fisheries knowledge graph, a comprehensive management plan for fishing vessel production, resource conservation, and market regulation is generated by invoking a preset fisheries management strategy model, including: From the fisheries knowledge graph, extract subgraph structures related to the current management objectives, which include improving production efficiency, protecting fishery resources, or stabilizing market prices; The extracted subgraph structure is input into the fisheries management strategy model, which is a decision-generating network based on graph neural networks. The fisheries management strategy model performs information propagation and aggregation on the subgraph structure, and finally generates a series of probability distributions of management actions for fishing vessel entities, sea area entities, and catch species entities in the graph at the output layer. Based on the probability distribution, management actions with probability values ​​exceeding a threshold are selected and combined with the management rule base to form specific and operable management measures. These management measures together constitute the comprehensive management plan covering guidance on fishing vessel production, resource conservation recommendations, and market regulation policies.

[0013] As a further aspect of the present invention, the training process of the fisheries management strategy model includes: Collect historical fisheries knowledge graph data and records of management measures taken during the corresponding periods, as well as their subsequent evaluation data, to form a training sample; By inputting historical fisheries knowledge graph data into the fisheries management strategy model to be trained, the probability distribution of predicted management actions can be obtained. The predicted probability distribution of management actions is compared with the actual management measures taken in history, and the subsequent evaluation data is taken into account to construct the loss function of the model. The loss function encourages the model to recommend management measures that have achieved positive and effective results in history. The parameters of the fisheries management strategy model are iteratively optimized using the backpropagation algorithm until the model converges, enabling it to generate effective management plans based on the input map data.

[0014] As a further aspect of the present invention, the comprehensive management plan is broken down into executable digital instructions and distributed to corresponding fishing vessel terminals, fishing port supervision terminals, and market announcement terminals, including: Analyze each management measure in the comprehensive management plan to identify its target execution terminal type, execution content, triggering conditions, and execution parameters; Based on different target terminal types, management measures are translated into standardized instruction formats that can be recognized and executed by the terminals. The standardized instruction format includes an instruction header, an instruction body, and a checksum. Assign a unique instruction identifier to each digitized instruction generated by translation, and set the effective time or triggering logic of the instruction based on the triggering conditions; Through the communication gateway of the fisheries digital platform, the digital instructions are pushed to the online fishing vessel terminals, fishing port supervision terminals, and market announcement terminals according to their target execution terminal addresses.

[0015] As a further aspect of the present invention, the present invention also includes a fishery digital platform management system based on big data, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the fishery digital platform management method based on big data as described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: An improved spatiotemporal clustering algorithm, incorporating vessel behavior patterns and marine geographic fencing information, is used to analyze fishing vessel trajectory data. This algorithm standardizes the constraint logic of spatial boundaries and behavior patterns during clustering, regulates the criteria for trajectory clustering, and partitions the data according to marine control boundaries and vessel operation characteristics, thereby distinguishing fishing vessel operational behavior types and delineating activity areas. Multidimensional time-series correlation analysis is conducted on fishing ground environmental sensor data to extract the change sequences of key environmental factors affecting catch yields from continuous time-series data. Market dynamic characteristics are mined from catch transaction data to identify price fluctuation patterns and supply-demand relationship characteristics, expanding the dimensional range of fisheries data analysis.

[0017] By integrating fishing vessel operation patterns, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics, a fisheries knowledge graph can be built. This graph can establish a stable correlation framework among various elements such as fisheries production, marine environment, market transactions, and resource management, solidify the inherent correlation logic among various data elements, and form a standardized knowledge system that can support comprehensive judgment.

[0018] By leveraging a fisheries knowledge graph and invoking pre-defined fisheries management strategy models, a comprehensive management plan covering fishing vessel production, resource conservation, and market regulation is generated. This plan can form a systematic management plan based on the relationships between multiple elements. The comprehensive management plan is broken down into standardized digital instructions, which are distributed to fishing vessel terminals, fishing port supervision terminals, and market announcement terminals according to their business affiliations. This adapts to the business carrying specifications of different terminals, connecting the entire process from data analysis and strategy generation to terminal execution, forming a digital management and control mechanism for the entire fisheries chain. Attached Figure Description

[0019] Figure 1 This is a state diagram of the fishery digital platform management method based on big data as described in this invention; Figure 2 A flowchart for identifying fishing vessel operation patterns and suspected illegal operation areas; Figure 3 This is a flowchart for extracting key environmental factor sequences. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0022] See Figure 1 This invention provides a fisheries digital platform management method based on big data, the specific method including: Through multiple data interfaces on the platform, a heterogeneous fisheries data set is continuously acquired from various fisheries monitoring sources, including Automatic Identification System (AIS), shipborne sensors, shore-based sensor networks, aquatic product trading market information systems, and aquaculture area video surveillance systems. This set specifically includes vessel trajectory point data recording vessel location, time, speed, and heading; fishing ground environmental sensor data from marine buoys and remote sensing satellites; catch transaction flow data recording transaction details; and video surveillance data from nearshore and deep-sea aquaculture areas. This heterogeneous data is processed in parallel. For the vessel trajectory point data, an improved spatiotemporal clustering algorithm is applied. This algorithm incorporates constraints from vessel behavior patterns and marine geofencing information into traditional clustering, thereby identifying different fishing vessel operation patterns and suspected illegal operation areas. Multidimensional time-series correlation analysis is performed on the fishing ground environmental sensor data to extract key environmental factor sequences significantly correlated with changes in catch volume. Market dynamic feature mining is performed on the catch transaction flow data to generate price fluctuation patterns and supply-demand relationship characteristics that characterize market patterns. This paper deeply integrates the analyzed fishing vessel operation patterns, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics. By defining entities and relationships, a fisheries knowledge graph reflecting the multi-dimensional relationship between fisheries production, environment, and market is constructed. Based on the constructed fisheries knowledge graph, a pre-set, trained fisheries management strategy model is invoked. This model can reason about the graph information to generate a comprehensive management plan that simultaneously covers fishing vessel production scheduling suggestions, fisheries resource conservation measures, and market regulation strategies. This comprehensive management plan is then structurally decomposed into specific digital instructions that can be understood and executed by different terminals. Through the platform's instruction distribution system, these instructions are distributed to the onboard terminals of target fishing vessels, the monitoring and command terminals of fishing ports, and the information dissemination terminals of the market, thus forming a data-driven closed loop from data perception to management execution.

[0023] In one embodiment of the present invention, the process of performing an improved spatiotemporal clustering algorithm on fishing vessel trajectory point data is specifically implemented as follows. (See reference...) Figure 2The received raw fishing vessel trajectory data is cleaned to remove obviously erroneous points, and interpolation smoothing is performed on missing trajectory points caused by signal loss, forming a spatiotemporally continuous sequence of fishing vessel trajectories suitable for analysis. An improved spatiotemporal clustering algorithm is then used to analyze this sequence. This algorithm employs a comprehensive metric when calculating the similarity between any two trajectory points. This metric not only calculates the spatial Euclidean distance and time difference between the two points but also introduces a comparison of heading and speed similarity, and examines whether the two points are located within the same pre-defined no-fishing zone or protected area's geographic fence. The heading and speed similarity is calculated by comparing the histogram of heading distribution and speed statistical characteristics of local trajectory segments within a time window before and after each point. The geographic fence constraint component is assigned a value by determining whether the two points are located within a marine geographic polygon area with specific management attributes. The spatial distance, temporal proximity, heading and speed similarity, and geographic fence constraint components are assigned different weights and combined using a weighted fusion formula to form a comprehensive similarity score. Based on this comprehensive similarity metric, a density-based clustering algorithm is used to cluster the spatiotemporal trajectory points of fishing vessels. The algorithm groups trajectory points with high spatiotemporal density and high similarity into clusters. Each stable cluster represents a specific fishing vessel operation mode. For example, areas with continuous low-speed movement and dense trajectories may be identified as trawl operations, while fast, large-scale straight-line trajectories may be identified as transit operations. During clustering, trajectory point clusters located at or adjacent to the boundaries of prohibited fishing areas, protected areas, or other geographically defined boundaries, and whose overall behavioral pattern characteristics differ significantly from normal navigation or adjacent conventional operation areas, are marked by the system as suspected illegal operation areas. In the post-processing stage of the clustering results, to improve accuracy, an internal consistency evaluation is performed on each identified cluster. This checks whether the behavioral pattern similarity of the trajectory points within the cluster is consistent with the geographically defined boundary attributes. Clusters with consistency below a preset threshold are either split or removed from the final set of stable operation modes.

[0024] In practical implementation, the big data-based fisheries digital platform management method involves processing fishing vessel trajectory point data, which includes location, time, speed, and heading information. This data is acquired from the fishing vessel's automatic identification system or onboard sensors. The fishing vessel trajectory point data is cleaned and interpolated to form a continuous spatiotemporal trajectory sequence. The cleaning process removes outliers with latitude and longitude outside a reasonable range or reversed timestamps. The interpolation process fills in the signal loss period by linearly interpolating based on the position and time of the preceding and following trajectory points. In some embodiments, an improved spatiotemporal clustering algorithm analyzes the fishing vessel spatiotemporal trajectory sequence. When calculating the similarity between trajectory points, the improved spatiotemporal clustering algorithm comprehensively considers spatial distance, temporal proximity, heading and speed similarity, and whether the vessel is located within a preset no-fishing zone or protected area's geographical fence. Spatial distance is calculated using Euclidean distance to determine the physical distance between two points. Temporal proximity is calculated using the absolute value of the timestamp difference between the two points. Heading speed similarity is calculated by comparing the heading distribution histograms and speed statistics of local trajectories around the two points. Geofencing constraints are determined by querying a pre-defined marine geofence polygon database to determine whether two points are located in the same no-fishing zone or protected area. In an example scenario, compared to clustering algorithms that only use spatial distance, the improved spatiotemporal clustering algorithm can distinguish spatially close but different behavioral patterns of fishing vessel trajectory point data in the same sea area. For example, it can separate high-speed straight-line movement transit patterns from low-speed curved movement trawling operation patterns.

[0025] In practical implementation, the behavioral pattern similarity component is calculated by comparing the similarity of the local trajectories around two points in terms of heading and velocity distribution. The local trajectory is a sequence of trajectory points within a fixed time window before and after the current point. The heading distribution histogram divides the heading angle into multiple intervals for statistical frequency, and the velocity statistical characteristics include mean and variance. The geofence constraint component determines whether two trajectory points are both within a specific marine geofence or within fences with opposing properties, such as one point in a no-fishing zone and the other in a permitted fishing zone. The improved spatiotemporal clustering algorithm weights and fuses the behavioral pattern similarity component, the geofence constraint component, and the basic spatiotemporal distance component to form a comprehensive similarity measure for subsequent density clustering. It can be understood that the weighting coefficients for the weighted fusion are pre-set according to the actual application scenario, with weight coefficients assigned to the spatial distance component, temporal proximity component, heading and velocity similarity component, and geofence constraint component, respectively. , , and Comprehensive similarity measurement The calculation formula is expressed as: in: and Represents two trajectory points, Represents the normalized spatial distance. Indicates the temporal proximity of the normalized data. Indicates the similarity of behavioral patterns. This represents the geofence constraint component, where the sum of all weight coefficients is 1, and each component value is normalized to the range [0,1]. In another example scenario, data comparison shows that for two trajectory points that are spatially close but one is located in a protected area and the other in a regular sea area, a clustering algorithm based solely on spatial distance may classify them into the same category. However, the improved spatiotemporal clustering algorithm reduces similarity by using the geofence constraint component, thus avoiding incorrect clustering.

[0026] In practical implementation, based on a comprehensive similarity metric, the improved spatiotemporal clustering algorithm employs density clustering to cluster the spatiotemporal trajectory points of fishing vessels. Density clustering groups trajectory points with high spatiotemporal density and similar behavioral patterns into clusters. Each cluster represents a stable fishing vessel operation mode, including trawl operations, purse seine operations, or transit operations. Trawl operation modes correspond to clusters with low-speed movement and dense trajectories, purse seine operation modes correspond to clusters with circular or encircling trajectories, and transit operations correspond to clusters with high-speed linear movement. Clusters of trajectory points identified within or at the edge of prohibited fishing areas or protected area geofences, and whose behavioral patterns do not conform to routine operations or navigation, are marked as suspected illegal operation areas. In some embodiments, during the clustering result evaluation phase, the consistency of behavioral patterns of the trajectory points within each identified cluster is checked against the consistency of geofence attributes. Behavioral pattern consistency is measured by calculating the average similarity of behavioral patterns among all point pairs within the cluster, while geofence attribute consistency is determined by checking whether the points within the cluster primarily belong to the same geofence type. Clusters with consistency below a threshold are split or removed. The splitting operation divides the cluster into multiple sub-clusters based on the dimension with low consistency, while the removal operation removes inconsistent clusters from the results. The threshold is understood to be set based on historical data or expert knowledge; for example, the behavioral pattern consistency threshold is set to 0.7, and the geofence attribute consistency requirement is that more than 80% of the points within a cluster belong to the same fence type. In practical implementation, through the above process, the improved spatiotemporal clustering algorithm outputs fishing vessel operation patterns and suspected illegal operation areas. For example, in a data comparison, the traditional clustering algorithm misclassifies trajectories near prohibited fishing areas as normal operations, while the improved spatiotemporal clustering algorithm splits them into two clusters—compliant navigation and suspected illegal operations—through consistency checks.

[0027] In one embodiment of the present invention, the process of performing multidimensional time-series correlation analysis on fishery environment sensing data is specifically implemented as follows. (See reference...) Figure 3Fishing ground environmental sensor data typically includes time-series measurements of multiple environmental parameters such as water temperature, salinity, chlorophyll concentration, dissolved oxygen, and ocean current velocity. This data is input in time-series format. The raw time-series measurements of each environmental factor are standardized to eliminate dimensional differences, and statistical methods are used to identify and remove outliers caused by sensor malfunctions or transmission errors, resulting in regularized environmental factor time-series data. Cross-correlation between these regularized environmental factor time-series data is calculated to analyze whether there are unidirectional or inverse relationships. Simultaneously, time-lag correlation analysis is performed between these environmental factor time-series data and historical catch data series for the same period, calculating the correlation coefficients between environmental factors and catch at different time offsets. Through analysis, environmental factors that show a significant correlation with catch data at specific time lags, such as leading catch changes by several days or weeks, are selected as candidate key environmental factors. Since multicollinearity may exist among these candidate key environmental factors, principal component analysis is further performed on their time-series data matrices. Principal component analysis transforms multiple correlated environmental factor variables into a set of linearly uncorrelated principal component variables. The top few principal components with eigenvalues ​​exceeding a set threshold and cumulative variance contribution rates exceeding a preset proportion are selected. These principal components can explain most of the variation information in the original environmental data. Extracting the score sequence of each principal component over time constitutes the key environmental factor sequence for subsequent analysis. This sequence comprehensively reflects the changes in the core environmental drivers affecting fishery productivity.

[0028] In practical implementation, the big data-based fisheries digital platform management method involves processing environmental sensor data from fishing grounds. This data includes time-series measurements of water temperature, salinity, chlorophyll concentration, dissolved oxygen, and ocean current velocity. These data originate from monitoring buoys deployed in the target fishing grounds, remote sensing satellite inversion products, and aerial survey data. The raw time-series measurements of various environmental factors are standardized. The Z-score method is used to eliminate dimensional differences between parameters such as water temperature, salinity, chlorophyll concentration, dissolved oxygen, and ocean current velocity. A statistical box plot method is applied to identify and remove outliers caused by momentary sensor malfunctions or data transmission errors, thus obtaining well-ordered time-series data of environmental factors. In one example scenario, processing one year's worth of water temperature sensor data for a certain sea area revealed instantaneous spikes in the raw data due to transmission interference. The standardization and outlier removal process removed these data points that deviated from the normal range by more than three standard deviations, resulting in a smooth daily average water temperature sequence.

[0029] In practice, the cross-correlation between time-series data of different environmental factors is calculated, and the Pearson correlation coefficient is used to analyze the pairwise linear correlation between time-series data of water temperature, salinity, chlorophyll concentration, dissolved oxygen, and ocean current velocity. The time-lag correlation between each environmental factor and historical catch data from the same period is analyzed; the catch data comes from fishing vessel production reports from the same sea area. Time-lag correlation analysis is achieved by calculating the correlation coefficients between the environmental factor series and the catch series at different time offsets. In some embodiments, the time-lag correlation coefficient... The calculation formula is: in: Indicates a certain environmental factor over time Time series data, Indicates time Fish catch data, Describing covariance, and They represent sequence sum Standard deviation of the sequence The time lag can be positive or negative. A positive time lag indicates that changes in environmental factors precede changes in catch volume. Data comparison and analysis revealed that when chlorophyll concentration time-series data for a certain sea area led catch volume data by four weeks, the correlation coefficient reached 0.65, while the immediate correlation coefficient between water temperature time-series data and catch volume data was only 0.3. This suggests that chlorophyll concentration is a more significant leading indicator.

[0030] In specific implementation, environmental factors that show a significant correlation with catch data at a specific time lag are selected as candidate key environmental factors, and a threshold for the absolute value of the correlation coefficient is set, for example, greater than 0.5. Optionally, for multiple environmental factors with multiple correlations, such as water temperature time series data and salinity time series data being highly correlated, they are included in the candidate set simultaneously. Principal component analysis is performed on the time series data of the candidate key environmental factors. Principal component analysis transforms the multiple environmental factor variables that may have been correlated into linearly uncorrelated principal component variables through orthogonal transformation. In some embodiments, principal component analysis extracts a few principal components that can explain most of the data variance, selecting principal components with eigenvalues ​​greater than 1, whose cumulative variance contribution rate must exceed a preset proportion, for example, 85%. The first principal component is usually a weighted sum of multiple environmental factors, reflecting the most important cooperative change pattern. It can be understood that the time series change sequence of the principal components, i.e., the principal component score sequence at each time point, is used as the key environmental factor sequence. In practice, through the above process, for example in a data comparison scenario, after principal component analysis of the original five candidate environmental factors, the first two principal components explain 90% of the variance. Using the score sequence of these two principal components as the key environmental factor sequence can more stably characterize the comprehensive changes of the core environmental drivers affecting fishery productivity than using a single original environmental factor sequence.

[0031] In one embodiment of the present invention, the process of performing market dynamic feature mining on fishery transaction data is specifically implemented as follows: The structured transaction data is parsed to extract key fields for each transaction, including the type and grade of the catch, transaction timestamp, transaction price, transaction quantity, and anonymized identification information for the buyer and seller. The data is then categorized according to the type and grade of the catch and aggregated according to preset time windows, such as daily, weekly, or monthly, to generate price and volume time series at different time granularities. For the price time series, a time series decomposition method is used to decompose it into a combination of long-term trend components, seasonal cycle components, cyclical fluctuation components, and irregular residual components. From the decomposition results, seasonal fluctuation patterns with fixed periods and identifiable cyclic fluctuation patterns with variable periods are extracted, together constituting a price fluctuation pattern describing the market price change pattern of the catch. When analyzing supply and demand characteristics, the analysis focuses on two aspects: firstly, the lead-lag relationship between the trading volume time series and the corresponding price time series on the time axis, such as whether a surge in trading volume precedes a price increase; secondly, by combining buyer and seller information, market concentration indicators, such as the Herfindahl-Hirschman Index, are calculated to analyze the comparison and changing trends of market buying and selling forces. These analyses combine to generate supply and demand characteristics that reflect the immediate tightness or looseness of the market and its potential direction of change. Subsequently, by integrating fishing vessel operation patterns, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics, a fisheries knowledge graph for comprehensive analysis is constructed. Its implementation includes: pre-defining the model layer of the fisheries knowledge graph, namely entity types and relationship types. Entity types include fishing vessels, operation patterns, sea areas, environmental factors, catch species, and market price patterns. Relationship types include engaging in, located in, affected by, associated with, and manifested as. Each identified fishing vessel operation pattern is instantiated as an operation pattern entity and connected to the specific fishing vessel entity executing that pattern through the "engaging in" relationship. Suspected illegal operation areas are instantiated as sea area entities with the "illegal" attribute. Each principal component of the key environmental factor sequence is instantiated as an environmental factor entity and connected to the marine area entity in which it mainly acts through "located" or "influence" relationships. After structurally describing the extracted price fluctuation patterns and supply-demand characteristics, these are instantiated as market price pattern entities and connected to related catch species entities through "associated with" relationships. Finally, using graph embedding technology, all entities and relationships in the graph are mapped to a low-dimensional continuous vector space for representation learning. This ensures that semantically similar entities are close in distance within the vector space, thus forming a fisheries knowledge graph rich in semantic and relational information.

[0032] In practical implementation, the big data-based fisheries digital platform management method involves mining market dynamic characteristics from fishery transaction data. This data originates from data interfaces of aquatic product wholesale markets and online trading platforms. The process involves analyzing the fishery transaction data to extract the species, specifications, transaction time, transaction price, transaction quantity, and buyer / seller information for each transaction. The buyer / seller information is anonymized, retaining only the anonymized identifiers used for market structure analysis. The transaction data is aggregated by species, specifications, and preset time windows (daily, weekly, or monthly) to generate price and volume time series at different time granularities. The price time series is typically calculated using a volume-weighted average price. In one example scenario, a year's worth of ribbonfish transaction data is analyzed, aggregated by "ribbonfish - large size" and "week," generating 52-week price and volume time series.

[0033] In practical implementation, the price time series is decomposed using the classic seasonal decomposition method, separating it into a superposition of long-term trend components, seasonal cycle components, cyclical fluctuation components, and irregular residual components. The long-term trend component reflects the long-term upward or downward direction of prices, the seasonal cycle component reflects regular fluctuations with a fixed annual cycle, the cyclical fluctuation component reflects medium- to long-term fluctuations with variable cycles, and the irregular residual component reflects random disturbances. Price fluctuation patterns are extracted from the decomposed seasonal cycle and cyclical fluctuation components. These patterns can be quantified as seasonal exponential curves and characteristic parameters such as cyclical fluctuation cycle and amplitude. In some embodiments, the price fluctuation pattern is structured as a data object containing attributes such as pattern type, cycle, amplitude, and phase. It is understood that price fluctuation patterns differ across different commodities, showcasing the main component characteristics after decomposing a monthly price time series for a particular commodity. The lead-lag relationship between the trading volume time series and the corresponding price time series is analyzed. Cross-correlation analysis is used to calculate the correlation coefficient between the trading volume series and the price series at different time shifts, seeking whether changes in trading volume lead changes in price. Market concentration is analyzed by combining information from both buyers and sellers. Market concentration is measured by the Herfindahl-Hirschmann index (HHI), which is calculated using the following formula: in: This represents the total number of buyers or sellers in the market. Indicates the first The market share of each buyer's or seller's trading volume. A rise in the HHI index may indicate that market forces are becoming more concentrated. This generates supply and demand characteristics that reflect the immediate market conditions and potential changes. These characteristics can be described as patterns such as "supply shortage - price increase - volume-leading" or "market fragmentation - price stability," as shown in Table 1.

[0034] Table 1: Monthly Price Time Series Decomposition Table for a Certain Commodity In practical implementation, a fisheries knowledge graph for comprehensive analysis is constructed by integrating fishing vessel operation patterns, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics. The entity types and relationship types of the fisheries knowledge graph are defined. Entity types include fishing vessels, operation patterns, sea areas, environmental factors, catch species, and market price patterns. Relationship types include engaging in, located in, affected by, associated with, and manifested as. Fishing vessel operation patterns are instantiated as operation pattern entities and connected to specific fishing vessel entities through the engagement relationship; for example, "fishing vessel Zheyu 12345" is connected to the "trawl operation" entity through the engagement relationship. Suspected illegal operation areas are instantiated as sea area entities with illegal attributes, and the type of illegality is labeled in the entity attributes. Each principal component of the key environmental factor sequence is treated as an environmental factor entity, such as "principal component 1 - comprehensive environmental index," and connected to relevant sea area entities through location or influence relationships. In some embodiments, price fluctuation patterns and supply and demand characteristics are structured and instantiated as market price pattern entities, such as "ribbonfish - spring price increase pattern," and connected to specific catch species entities through association relationships. By utilizing graph embedding techniques, such as TransE or graph neural networks, all entities and relationships are represented in a low-dimensional vector space, forming a fisheries knowledge graph rich in semantic and relational information. The embedding representation allows the vector relationships "fishing vessel - engaged in -> trawling" and "trawling - associated with -> ribbonfish" to be reflected in the space. Optionally, the graph embedding process is accomplished by minimizing the scoring loss function of existing fact triples in the knowledge graph. In a data comparison scenario, compared to traditional relational database storage, this graph structure can more intuitively support complex relational queries such as "finding the market price pattern of species C caught by fishing vessels engaged in B operations in sea areas affected by environmental factor A."

[0035] In one embodiment of the present invention, the process of generating a comprehensive management plan based on a fisheries knowledge graph and a preset fisheries management strategy model is implemented as follows: When the management platform receives an instruction to generate a management plan, it first extracts relevant subgraph structures from the constructed fisheries knowledge graph according to the current management objectives. For example, if the objective is to improve the production efficiency of a specific sea area, a subgraph is extracted that includes entities of the sea area, nearby environmental factors, fishing vessels operating in the sea area and their operational patterns, and market price patterns of relevant catches. This extracted subgraph structure is then input into the preset fisheries management strategy model, which is a decision-generating network based on a graph neural network. The graph neural network performs multiple rounds of information propagation and aggregation on the input subgraph structure, and the feature vector of each entity node aggregates information from its neighboring nodes (connected through relationships). After processing by multi-layer graph convolution or graph attention networks, the output layer of the model generates a series of predefined probability distributions for management actions applied to entity nodes in the subgraph, such as fishing vessel entities, sea area entities, and catch species entities. Examples include the probability of actions like "suggesting a change of fishing grounds" for a specific fishing vessel entity or "suggesting increased patrols" for a specific sea area. Based on this probability distribution, the system selects valid management actions with probability values ​​exceeding a set threshold. These actions are then combined with industry standards, policies, and regulations stored in the management rule base to form specific and actionable management measures. These measures collectively constitute a comprehensive management plan covering multiple dimensions, including guidance for fishing vessel production, resource conservation recommendations, and market regulation policies. The training process of the fisheries management strategy model is as follows: historical fisheries knowledge graph data and records of management measures actually taken by fisheries management departments during the corresponding periods are collected, along with post-implementation evaluation data such as resource recovery indicators and changes in economic benefits, forming a training sample set. The fisheries knowledge graph data at a specific historical moment is input into the model to be trained, and the model outputs the predicted probability distribution of management actions. This predicted distribution is compared with the actual management measures taken at this point in history, and historical measures are labeled as "positive" or "negative" based on post-effect evaluation data to construct the model's loss function. This loss function is designed so that the model's recommended management actions tend to align with historically effective measures, while avoiding negative ones. All parameters in the model are iteratively optimized using the backpropagation algorithm until the model converges on the validation set, resulting in a model capable of generating effective management plans based on the input geographic data.

[0036] In specific implementation, the process of generating a comprehensive management plan based on a fisheries knowledge graph and a pre-defined fisheries management strategy model is as follows: When the management platform receives an instruction to generate a management plan, it extracts relevant subgraph structures from the constructed fisheries knowledge graph according to the current management objectives, which may include improving production efficiency, protecting fishery resources, or stabilizing market prices. For example, if the current management objective is "protecting a specific fishery resource in the East China Sea," then a subgraph structure is extracted from the fisheries knowledge graph that includes entities related to the sea area, its associated environmental factors, fishing vessels recently active in the sea area and their operational patterns, as well as entities representing the main catch species in the sea area and their market price patterns. This extracted subgraph structure is then input into the fisheries management strategy model, which is a decision-generating network based on a graph neural network. In some embodiments, the fisheries management strategy model adopts a graph attention network architecture, where each entity node in the subgraph structure has an initial feature vector, and relationships are treated as edges. The graph attention network performs multiple rounds of information propagation and aggregation on the input subgraph structure. In each layer, the feature vector of each entity node is weighted and aggregated with information from its neighboring nodes through an attention mechanism. After processing through multiple layers of a graph neural network, the output layer of the fisheries management strategy model generates a series of predefined probability distributions of management actions for entity nodes in the subgraph that can be subject to management actions. Optionally, the probability distributions of management actions are generated through a Softmax output layer that is related to the entity node type. For example, for fishing vessel entity nodes, possible management actions include "suggesting a change to fishing ground A", "suggesting a return to port for rest", and "suggesting a change of fishing gear"; for marine area entity nodes, possible management actions include "suggesting increased patrols" and "suggesting the establishment of a temporary protected area".

[0037] In its implementation, the fisheries management strategy model disseminates and aggregates information on a subgraph structure. Ultimately, at the output layer, the model generates a probability distribution of management actions for fishing vessel entities, sea area entities, and catch species entities within the graph. Based on this probability distribution, management actions with probabilities exceeding a threshold are selected and combined with a management rule base to form specific and actionable management measures. The management rule base stores industry regulations, policy documents, and historical experience rules, used for compliance verification and parameter refinement of the original actions output by the fisheries management strategy model. For example, if the fisheries management strategy model outputs the action "suggest turning to fishing ground A" with a probability of 0.85 (exceeding the threshold of 0.7) for a certain fishing vessel entity, the management rule base, combined with the current status of fishing ground A and the vessel's voyage, refines it into the specific measure "suggesting that vessel Zheyu 12345 turn to the sea area at longitude XXX, latitude YYY within 24 hours." These management measures collectively constitute a comprehensive management plan covering fishing vessel production guidance, resource conservation recommendations, and market regulation policies. See Table 2 for a simplified output.

[0038] Table 2: Probability Distribution of Management Actions Output by the Fisheries Management Strategy Model The training process for the fisheries management strategy model includes collecting historical fisheries knowledge graph data and records of management measures taken during the corresponding periods, along with their subsequent effect evaluation data, to form the training sample. The historical fisheries knowledge graph data is constructed from historical data using the same method. Management measure records are derived from historical management logs, and the effect evaluation data comes from resource assessment reports and economic statistics reports from a period after the implementation of the measures, labeled with "positive" or "negative" utility tags by experts. The historical fisheries knowledge graph data is input into the fisheries management strategy model to be trained, yielding the predicted probability distribution of management actions. The predicted probability distribution of management actions is compared with the actual management measures taken historically, and considering the effect evaluation data, a loss function for the fisheries management strategy model is constructed. This loss function encourages the fisheries management strategy model to recommend historically effective management measures. Here is an example: in: This represents the number of entity-action pairs in a training batch. It is a binary indicator variable that indicates whether the action was actually taken in the corresponding graph state in history (1 for taking, 0 for not taking). It is the probability of taking this action predicted by the fisheries management strategy model. and This is a weighting coefficient, the value of which is adjusted based on the post-effect evaluation label of the corresponding historical management measures. For actions with a "positive" post-effect, its... Actions with a value greater than their aftereffect are considered "negative." This weighting allows historically effective measures to receive higher weights in the loss function, driving model learning. The parameters of the fisheries management strategy model are iteratively optimized using the backpropagation algorithm until the model converges, enabling it to generate effective management plans based on the input map data. In an example scenario, data comparison shows that an untrained fisheries management strategy model outputs random actions, while a well-trained model, under the same test map input, assigns a higher probability to management actions that have historically yielded positive aftereffects.

[0039] In one embodiment of the present invention, the process of decomposing a comprehensive management plan into executable digital instructions and distributing them to corresponding terminals is implemented as follows. The comprehensive management plan output by the plan generation module is a structured document or data object. The instruction decomposition engine first parses each specific management measure in the plan, identifies the type of target execution terminal, the specific content of the execution measure, the triggering conditions of the measure, and the various parameters required for execution. The target execution terminal types mainly include fishing vessel terminals, fishing port supervision terminals, and market announcement terminals. According to different terminal types, the text or structured description of the management measures is translated into a standardized instruction format that can be recognized and executed by that type of terminal. This standardized instruction format usually includes three parts: an instruction header, an instruction body, and a check code. The instruction header contains information such as instruction type, target terminal identifier, and priority; the instruction body carries specific operation commands and parameters; and the check code is used to ensure the integrity of instruction transmission. A globally unique instruction identifier is assigned to each translated digital instruction for tracking and logging purposes. At the same time, based on the triggering conditions parsed from the management measures, the absolute time of its effectiveness or the logical conditions for its triggering are set in the instruction. Finally, through the communication gateway integrated into the fisheries digital platform, the encapsulated digital instructions, based on the target terminal address information in the instruction header, are pushed to the currently online target fishing vessel terminals, the target fishing port monitoring and command terminals, and the market information release and announcement terminals via different communication links such as satellite communication, mobile networks, or dedicated lines. Upon receiving the instructions, the terminals automatically execute them or prompt the operators to perform the corresponding operations based on the instruction content.

[0040] In practical implementation, the comprehensive management plan is broken down into executable digital instructions and distributed to the corresponding fishing vessel terminals, fishing port supervision terminals, and market announcement terminals. The specific implementation is as follows: Each management measure in the comprehensive management plan is analyzed to identify its target execution terminal type, execution content, triggering conditions, and execution parameters. The target execution terminal types include fishing vessel terminals, fishing port supervision terminals, and market announcement terminals. The execution content is the specific operation required by the management measure. The triggering condition is the time or event condition for executing the operation. The execution parameters are the specific numerical or location information required for the operation. In an example scenario, a management measure is analyzed: "It is recommended that the Zhejiang Fishery 12345 vessel turn to the sea area of ​​East Longitude XXX, North Latitude YYY within 24 hours for production." Its target execution terminal type is identified as "fishing vessel terminal," the execution content is "turn to the designated sea area," the triggering condition is "after receiving the instruction," and the execution parameters are "East Longitude XXX, North Latitude YYY." In another example scenario, data comparison shows that for the same measure of "strengthening patrols in a certain sea area", the target execution terminal type is "fishing port supervision terminal", while for the measure of "issuing a price warning for a certain commodity", the target execution terminal type is "market announcement terminal".

[0041] In practical implementation, management measures are translated into standardized instruction formats that can be recognized and executed by the terminals, based on different target terminal types. These standardized instruction formats include an instruction header, an instruction body, and a checksum. The instruction header contains metadata such as instruction type encoding, a unique identifier for the target terminal, instruction priority, and protocol version number. The instruction body carries the encoded specific operation commands and parameters. The checksum is used to verify the integrity of the instruction during transmission. In some embodiments, for turning instructions on fishing vessel terminals, the instruction body may include fields such as target latitude and longitude and suggested speed; for patrol instructions on fishing port monitoring terminals, the instruction body may include fields such as target sea area boundary coordinates and patrol intensity; for warning instructions on market announcement terminals, the instruction body may include fields such as warning type, warning level, and warning text. A unique instruction identifier is assigned to each translated digital instruction. This identifier is typically generated by combining a timestamp, instruction type, and random number to ensure global uniqueness. The effective time or triggering logic of the instruction is set based on the triggering conditions parsed from the management measures. The effective time can be an absolute future point in time, and the triggering logic can be a Boolean expression dependent on the fulfillment of specific events or conditions. Optionally, for instructions that are to be executed immediately upon receipt, their effective time is set to the instruction generation time.

[0042] In practical implementation, the communication gateway of the fisheries digital platform pushes digital instructions to online fishing vessel terminals, fishing port monitoring terminals, and market announcement terminals according to their target execution terminal addresses. Fishing vessel terminals typically access the system via satellite communication or mobile networks, fishing port monitoring terminals via wired broadband or dedicated networks, and market announcement terminals via the internet. The communication gateway is responsible for adapting to different communication protocols and links. The integrity of the digital instructions is guaranteed by a checksum, which is obtained by calculating the instruction body content using a specific algorithm. This can be understood as a method for calculating the instruction checksum. The formula is: in: This represents a cryptographic hash function (such as SHA-256). The instruction body portion represents the digitized instructions. This represents a pre-shared key. This indicates a connection operation. The receiving terminal recalculates the checksum using the same algorithm and compares it with the received checksum to verify that the instruction has not been tampered with during transmission. In some embodiments, the communication gateway employs an asynchronous confirmation mechanism. For instructions sent to the fishing vessel terminal, the instruction remains in a "sending" state in the gateway until an acknowledgment is received from the terminal. If no acknowledgment is received within a timeout period, a retransmission is triggered. In a specific example scenario, compared to traditional broadcast text notifications, this structured digital instruction can be automatically parsed by the terminal and trigger standard processing procedures within the terminal. For example, after receiving a standardized steering instruction, the fishing vessel terminal can directly import it into the autopilot system to generate a course, without requiring crew members to manually input coordinates.

[0043] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A fisheries digital platform management method based on big data, characterized in that, The method includes: Acquire heterogeneous fishery data sets from multiple fishery monitoring sources, including fishing vessel trajectory point data, fishing ground environmental sensor data, catch transaction flow data, and aquaculture area video surveillance data; An improved spatiotemporal clustering algorithm is applied to the fishing vessel trajectory point data to identify fishing vessel operation patterns and suspected illegal operation areas. The improved spatiotemporal clustering algorithm constrains the clustering process based on vessel behavior patterns and marine geofence information. Multidimensional time-series correlation analysis was performed on the environmental sensor data of the fishing grounds to extract the sequence of key environmental factors affecting the catch. Market dynamic feature mining was performed on the transaction flow data of the catch to generate price fluctuation patterns and supply and demand relationship features. By integrating the aforementioned fishing vessel operation modes, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics, a fisheries knowledge graph for comprehensive analysis is constructed. Based on the fisheries knowledge graph, a comprehensive management plan for fishing vessel production, resource conservation, and market regulation is generated by calling a preset fisheries management strategy model. The comprehensive management plan is broken down into executable digital instructions and distributed to the corresponding fishing vessel terminals, fishing port supervision terminals, and market announcement terminals.

2. The fisheries digital platform management method based on big data according to claim 1, characterized in that, An improved spatiotemporal clustering algorithm is applied to the fishing vessel trajectory point data to identify fishing vessel operation patterns and suspected illegal operation areas, including: The fishing vessel trajectory point data is cleaned and interpolated to form a continuous spatiotemporal trajectory sequence of fishing vessels; The improved spatiotemporal clustering algorithm is used to analyze the spatiotemporal trajectory sequence of the fishing vessel. When calculating the similarity between trajectory points, the improved spatiotemporal clustering algorithm comprehensively considers spatial distance, temporal proximity, heading speed similarity, and whether it is located within a preset no-fishing zone or protected area geographical fence. Based on the similarity, density clustering is performed on the spatiotemporal trajectory points of fishing vessels. Trajectory points with high spatiotemporal density and similar behavioral patterns are aggregated into clusters. Each cluster represents a stable fishing vessel operation mode, including: trawling, purse seine, or transit. Clusters of trajectory points that are located within or at the edge of the geographical fence of a fishing ban area or protected area and whose behavior patterns do not conform to routine operations or navigation are identified and marked as suspected illegal operation areas.

3. The fishery digital platform management method based on big data according to claim 2, characterized in that, The improved spatiotemporal clustering algorithm constrains the clustering process based on ship behavior patterns and marine geographic fence information, including: When calculating the similarity between trajectory points, a behavioral pattern similarity component is introduced. The behavioral pattern similarity is calculated by comparing the similarity of the local trajectories around the two points in terms of heading and velocity distribution. Introduce a geofence constraint component to determine whether two trajectory points are both within a specific marine geofence, or within fences with opposing properties. The behavioral pattern similarity component, the geofence constraint component, and the basic spatiotemporal distance component are weighted and fused to form a comprehensive similarity metric, which is used in the subsequent density clustering process. During the clustering result evaluation phase, for each identified cluster, the consistency of the behavior patterns of its internal trajectory points and the consistency of the geofence attributes are checked. Clusters with consistency below the threshold are split or removed to ensure the accuracy of the identification of the fishing vessel operation patterns and suspected illegal operation areas.

4. The fisheries digital platform management method based on big data according to claim 1, characterized in that, Multidimensional time-series correlation analysis was performed on the environmental sensor data of the fishing grounds to extract the key environmental factor sequences affecting the catch, including: The environmental sensing data of the fishing grounds includes time-series measurements of water temperature, salinity, chlorophyll concentration, dissolved oxygen, and ocean current speed. The time-series measurements of various environmental factors are standardized and outlier removal is performed to obtain regularized time-series data of environmental factors; Calculate the cross-correlation between time series data of different environmental factors, and analyze the time-lag correlation between each environmental factor and historical catch data of the same period. Environmental factors that show correlation with catch data at a specific time lag are selected as candidate key environmental factors; Principal component analysis is performed on the time series data of the candidate key environmental factors to extract a few principal components that can explain most of the data variance. The time series variation sequence of the principal components is then used as the key environmental factor sequence.

5. The fisheries digital platform management method based on big data according to claim 1, characterized in that, Market dynamic feature mining is performed on the aforementioned fishery transaction data to generate price fluctuation patterns and supply and demand characteristics, including: The transaction data of the fish catch is analyzed to extract the species, specifications, transaction time, transaction price, transaction quantity and buyer and seller information for each transaction; The transaction data is aggregated according to variety, specification and preset time window to generate price time series and transaction volume time series with different time granularities; The price time series is decomposed to separate long-term trend, seasonal cycle, cyclical fluctuation and irregular residual components, and the price fluctuation pattern is extracted from the seasonal cycle and cyclical fluctuation. The leading-lag relationship between the trading volume time series and the corresponding price time series is analyzed, and market concentration is analyzed in combination with buyer and seller information, thereby generating the supply and demand relationship characteristics that reflect the real-time market conditions and potential changes.

6. The fisheries digital platform management method based on big data according to claim 1, characterized in that, By integrating the aforementioned fishing vessel operation patterns, suspected illegal operation areas, key environmental factor sequences, price fluctuation patterns, and supply and demand characteristics, a fisheries knowledge graph for comprehensive analysis is constructed, including: Define the entity types and relationship types of the fisheries knowledge graph. Entity types include fishing vessels, operation modes, sea areas, environmental factors, catch species, and market price patterns. Relationship types include engaging in, being located in, being affected by, being associated with, and being manifested as. The fishing vessel operation mode is instantiated as an operation mode entity and connected to the specific fishing vessel entity through the operation relationship. Instantiate the suspected illegal operation area as a sea area entity with illegal attributes; Each principal component of the key environmental factor sequence is treated as an environmental factor entity and connected to the relevant marine area entities through location or influence relationships. The price fluctuation pattern and supply and demand characteristics are structured, instantiated into a market price pattern entity, and connected with specific fish catch entities through association relationships. By utilizing graph embedding technology, all entities and relationships are represented and learned in a low-dimensional vector space to form the fisheries knowledge graph containing rich semantic and relational information.

7. The fisheries digital platform management method based on big data according to claim 1, characterized in that, Based on the aforementioned fisheries knowledge graph, a comprehensive management plan is generated by invoking a pre-defined fisheries management strategy model, targeting fishing vessel production, resource conservation, and market regulation. This plan includes: From the fisheries knowledge graph, extract subgraph structures related to the current management objectives, which include improving production efficiency, protecting fishery resources, or stabilizing market prices; The extracted subgraph structure is input into the fisheries management strategy model, which is a decision-generating network based on graph neural networks. The fisheries management strategy model performs information propagation and aggregation on the subgraph structure, and finally generates a series of probability distributions of management actions for fishing vessel entities, sea area entities, and catch species entities in the graph at the output layer. Based on the probability distribution, management actions with probability values ​​exceeding a threshold are selected and combined with the management rule base to form specific and operable management measures. These management measures together constitute the comprehensive management plan covering guidance on fishing vessel production, resource conservation recommendations, and market regulation policies.

8. The fishery digital platform management method based on big data according to claim 7, characterized in that, The training process of the fisheries management strategy model includes: Collect historical fisheries knowledge graph data and records of management measures taken during the corresponding periods, as well as their subsequent evaluation data, to form a training sample; By inputting historical fisheries knowledge graph data into the fisheries management strategy model to be trained, the probability distribution of predicted management actions can be obtained. The predicted probability distribution of management actions is compared with the actual management measures taken in history, and the subsequent evaluation data is taken into account to construct the loss function of the model. The loss function encourages the model to recommend management measures that have achieved positive and effective results in history. The parameters of the fisheries management strategy model are iteratively optimized using the backpropagation algorithm until the model converges, enabling it to generate effective management plans based on the input map data.

9. The fishery digital platform management method based on big data according to claim 1, characterized in that, The comprehensive management plan is broken down into executable digital instructions and distributed to corresponding fishing vessel terminals, fishing port supervision terminals, and market announcement terminals, including: Analyze each management measure in the comprehensive management plan to identify its target execution terminal type, execution content, triggering conditions, and execution parameters; Based on different target terminal types, management measures are translated into standardized instruction formats that can be recognized and executed by the terminals. The standardized instruction format includes an instruction header, an instruction body, and a checksum. Assign a unique instruction identifier to each digitized instruction generated by translation, and set the effective time or triggering logic of the instruction based on the triggering conditions; Through the communication gateway of the fisheries digital platform, the digital instructions are pushed to the online fishing vessel terminals, fishing port supervision terminals, and market announcement terminals according to their target execution terminal addresses.

10. A fisheries digital platform management system based on big data, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the fishery digital platform management method based on big data as described in any one of claims 1 to 9.

Images