An intelligent aquaculture consultant system and method

Abstract

The application provides an intelligent aquaculture consultant system and method, belonging to the field of artificial intelligence and aquaculture technology, comprising a multi-modal input processing module, a context-aware reasoning engine, an expert collaboration platform, a self-adaptive optimization module, a predictive aquaculture management module, a knowledge graph construction module, and a continuous learning and knowledge iteration module. The application adopts the above-mentioned intelligent aquaculture consultant system and method, can understand and solve complex aquaculture problems, provide accurate personalized suggestions, predict and prevent potential risks, realize intelligent transmission and application of aquaculture knowledge, and provide strong support for improving aquaculture efficiency and success rate.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and aquaculture technology, and in particular relates to an intelligent advisory system and method for aquaculture. Background Technology

[0002] Aquaculture is a crucial way to address global protein supply and a pillar industry in many coastal and inland regions. However, the aquaculture industry currently faces numerous challenges, including uneven technology dissemination, difficulty in acquiring professional knowledge, and complex risk management. With increasing global climate change and environmental pollution, aquaculture faces growing uncertainties, placing higher demands on aquaculture technology and management practices.

[0003] Traditional aquaculture technology support models mainly suffer from the following problems:

[0004] First, professional knowledge is scattered and difficult to access. Aquaculture involves knowledge from multiple disciplines such as biology, aquatic chemistry, ecology, and nutrition. Professional literature and experience are scattered across various channels, making it difficult for ordinary farmers to obtain and understand them comprehensively and systematically. At the same time, a large amount of valuable experience and knowledge exists in an implicit form in the minds of senior aquaculture experts, lacking an effective mechanism for extraction and inheritance.

[0005] Secondly, untimely technical support makes it difficult to cope with emergencies. Emergency situations such as changes in water quality and disease outbreaks during the aquaculture process often require rapid response. However, under the traditional technical service model, expert resources are limited and cannot meet the real-time consultation needs of a large number of farmers, resulting in missed opportunities for optimal intervention.

[0006] Third, there is a lack of personalized guidance. Optimal management plans vary significantly across different regions, water bodies, and aquaculture species, making it difficult to adapt general technical guidelines to diverse aquaculture conditions in different areas. Existing technical support systems generally lack the ability to provide customized advice based on the specific circumstances of each farmer.

[0007] Fourth, insufficient predictive capabilities hinder proactive management. The aquaculture environment is complex and ever-changing; relying solely on current data for management often leads to a reactive approach. The lack of scientific predictions of future environmental changes and aquaculture risks limits the foresight and initiative of aquaculture management.

[0008] Fifth, knowledge updates are slow. Aquaculture technology is constantly developing, and new research and methods are constantly emerging, but the traditional knowledge system has a long update cycle, making it difficult to integrate the latest research results and practical experience in a timely manner, resulting in a lag in technology promotion.

[0009] The development of technologies such as artificial intelligence, big data, and the Internet of Things has provided new possibilities for solving the aforementioned problems. Knowledge graph technology can systematically organize professional knowledge; multimodal interaction technology can achieve convenient and efficient human-computer communication; deep learning-based reasoning models are expected to simulate expert diagnosis and decision-making processes; and digital twin technology provides new methods for predicting aquaculture environments. However, the market currently lacks a comprehensive intelligent advisory system that organically integrates these advanced technologies and is specifically designed for the aquaculture sector. Summary of the Invention

[0010] The purpose of this invention is to provide an intelligent advisory system and method for aquaculture, which achieves intelligent identification and solution of aquaculture problems by constructing a professional knowledge graph and combining it with multimodal interaction technology; provides targeted suggestions by combining environmental data with context awareness technology; continuously optimizes the system through interaction through a continuous learning mechanism; and provides predictive aquaculture management through digital twin technology.

[0011] To achieve the above objectives, the present invention provides an intelligent advisory system for aquaculture, including...

[0012] The multimodal input processing module is used to collect, analyze, and digitize user-end input and real-time acquired data; user-end input includes aquaculture goals or aquaculture status;

[0013] The context-aware reasoning engine is used to extract features from the implementation data and aquaculture targets processed by the multimodal input processing module, analyze the correlation between environmental data and aquaculture data, and obtain the initial reasoning results.

[0014] The expert collaboration platform is used to absorb user actions collected by the multimodal input module, and to update the knowledge graph construction module through the continuous learning and knowledge iteration module; it is also used to assist the continuous learning and knowledge iteration module in knowledge conflict detection, knowledge fusion, and value assessment.

[0015] The adaptive optimization module is used to compare the consistency between user-implemented operations and system suggestions, and optimize the model parameters of relevant core modules of the context-aware inference engine based on the consistency results; it is used to adaptively adjust the context-aware inference engine according to the region and season; and it is used to cache the optimization and adjustment schemes locally to ensure offline operation.

[0016] The predictive aquaculture management module is used to predict key aquaculture indicators in the future by constructing a virtual aquaculture environment model and performing time series analysis, and automatically generate risk prevention and optimization management suggestions based on the prediction results;

[0017] The personalized suggestion generation module is used to generate implementation suggestions based on real-time collected data, knowledge graph data, and aquaculture goals using reinforcement learning algorithms;

[0018] The knowledge graph construction module provides a theoretical foundation for the context-aware reasoning engine, the predictive aquaculture management module, and the personalized suggestion generation module.

[0019] The continuous learning and knowledge iteration module is used to acquire expert experience from the expert collaboration platform and use it to update the knowledge graph construction module; when discovering potential new knowledge points in the system or identifying knowledge conflicts, it collaborates with the expert collaboration platform to provide authoritative verification and arbitration.

[0020] Preferably, the ontology-based modeling method used in the knowledge graph construction module includes the following steps:

[0021] The knowledge in the field of aquaculture is divided into core categories, and subcategories are set up under each core category.

[0022] Establish core category attribute sets for various entities based on the core category definition;

[0023] Establish the relationship between core categories and core category attribute sets to obtain the corresponding semantic relationships;

[0024] By extracting and integrating basic academic knowledge, applied knowledge, and expert experience from expert collaboration platforms, professional knowledge is obtained.

[0025] Using core categories and core category attributes as knowledge graph nodes, the semantic relationships between core categories and core category attribute sets are analyzed. Based on the acquired professional knowledge, connections between knowledge graph nodes are established to form a first-level knowledge network. On the basis of the first-level knowledge network, connections between sub-class structures are established to form a second-level knowledge network. A hierarchical design is adopted to construct general knowledge and regionally specialized knowledge separately, and automatic fusion is achieved through association rules to balance the generality and specificity of knowledge. Together with the first-level knowledge network and the second-level knowledge network, a knowledge graph is formed.

[0026] Preferably, the multimodal input processing module comprises the following units:

[0027] Set up a data acquisition unit, including user input on the user terminal and setting up acquisition devices for real-time data acquisition;

[0028] An image processing unit is set up to process images from user input and real-time acquired data;

[0029] A video analysis unit is set up to process video streams from user input and real-time data acquisition, dynamically analyze the group behavior patterns of farmed organisms, and maintain video analysis performance under insufficient lighting conditions through an adaptive brightness enhancement algorithm.

[0030] An audio processing unit is set up to process audio data from user input and real-time data acquisition, extract acoustic features, and analyze the operating status of aquaculture equipment and the sounds of abnormal aquatic animal activities.

[0031] A text processing unit is set up to combine optical character recognition and natural language processing technologies to perform digital processing and structured extraction on user input and real-time collected data, and to transform unstructured text into structured data through named entity recognition and relation extraction technologies.

[0032] The multimodal fusion unit maps feature vectors from different modalities to a unified semantic space, enhancing the complementarity of information between modalities. At the same time, it uses a modality-specific confidence scoring mechanism to dynamically weight the contribution of each modality, outputting a unified semantic representation.

[0033] Preferably, the context-aware reasoning engine includes:

[0034] The data processed by the multimodal input processing module is transformed into a graph structure and aligned with the knowledge graph nodes;

[0035] Feature extraction is performed on the graph structure, and the association patterns of graph structure features and knowledge graph nodes are analyzed. Causal chains are constructed for graph structure features and knowledge graph nodes, and weights are calculated.

[0036] By combining the temporal and spatial information of the causal chain with the structural features of the graph, a comprehensive representation of the current state of the aquaculture environment is constructed.

[0037] Through reverse reasoning and template filling, targeted solutions and operational suggestions are generated.

[0038] Preferably, the personalized suggestion generation module includes:

[0039] The reinforcement learning algorithm is used to generate implementation suggestions based on user input, real-time data collection and real-time graphs. Each implementation suggestion contains three levels of implementation details: the first level is the setting of core parameters, the second level is the operation method description, and the third level is the expected effect and risk assessment.

[0040] The reinforcement learning algorithm adopts a policy optimization framework based on deep Q-networks. It uses real-time monitoring data and regional characteristics of the knowledge graph to model a Markov decision process. The user input is used as the reward function. The system selects the best case in the breeding history of the knowledge graph. By learning from the best case, it forms an optimal policy library for different situations. At the same time, the Monte Carlo tree search method is introduced to evaluate the long-term decision path to ensure the long-term effectiveness of the suggestions. The system updates the Q value in real time from user feedback and continuously optimizes the decision model.

[0041] Preferably, the continuous learning and knowledge iteration module includes:

[0042] The system processes user interaction records, expert knowledge supplements from the expert collaboration platform, knowledge graphs, and real-time collected data using a multi-stage pipeline architecture to form a knowledge base.

[0043] The multi-stage pipeline architecture processes data cleaning and standardization using a multimodal input processing module, feature extraction and semantic annotation using a context-aware reasoning engine, and knowledge conflict detection, knowledge fusion and value assessment using an expert collaboration platform.

[0044] Meanwhile, the knowledge base also has a local device-based hierarchical storage architecture to store data processed by the multi-stage pipeline architecture locally.

[0045] The knowledge base also includes a central server that processes data from a multi-stage pipeline architecture and updates it in real time, supporting rollback functionality based on time series or specific event sequences.

[0046] Preferably, the predictive aquaculture management module includes:

[0047] Spatial scanning of the aquaculture farm generates a three-dimensional terrain model. Data collected in real time by the multimodal input processing module is used to obtain water quality parameters, meteorological parameters, and biological characteristic parameters of the aquaculture farm.

[0048] By utilizing a three-dimensional terrain model and water quality parameters, meteorological parameters, and biological characteristic parameters, a coupled model of hydrodynamics, water quality change, biological growth, and meteorological influence is constructed, namely a virtual aquaculture environment model, to achieve real-time mapping between the virtual and real environments;

[0049] The virtual aquaculture environment model uses a multi-scale fusion method for prediction over time series data;

[0050] The virtual aquaculture environment model is optimized through a multi-objective decision-making algorithm, taking the aquaculture objectives input by the user as the optimization objective and considering the constraints set by the user.

[0051] Preferably, the adaptive optimization module includes:

[0052] When comparing the consistency between user actions and system suggestions, the core aquaculture parameters output by the personalized suggestion generation module are automatically adjusted using a two-dimensional matrix of region and season. The adjustment range is determined based on statistical analysis of historical data. At the same time, the early warning threshold of the predictive aquaculture management module and the judgment threshold of the context-aware inference engine are automatically updated based on the latest weather forecast data.

[0053] The key parameters of each optimization, the graph neural network parameters of the context-aware inference engine, the reinforcement learning model weights of the personalized suggestion generation module, the time series prediction model parameters of the predictive aquaculture management module, and the updated content of the knowledge base are all cached locally. The adaptive optimization module, the predictive aquaculture management module, and the continuous learning and knowledge iteration module all have local caching functions and automatically update data through a differentiated synchronization mechanism.

[0054] A smart advisory method for aquaculture includes the following steps:

[0055] The multimodal input processing module collects user input and real-time acquired data, and performs analysis and digital processing.

[0056] The context-aware reasoning engine extracts features from the processed data and aquaculture objectives, analyzes the correlation between the environment and aquaculture data, and draws initial reasoning results to provide preliminary basis for aquaculture decision-making.

[0057] The expert collaboration platform incorporates user operations, updates the knowledge graph through continuous learning and knowledge iteration modules, assists in knowledge conflict detection, integration, and value assessment, and ensures the accuracy and timeliness of knowledge.

[0058] The adaptive optimization module compares the consistency between user operations and system suggestions, optimizes the parameters of the context-aware inference engine model, and adjusts the inference engine according to regional and seasonal adaptability. The optimization and adjustment schemes are cached locally to ensure offline operation and improve system adaptability and stability.

[0059] The predictive aquaculture management module constructs a virtual aquaculture environment model, predicts key future aquaculture indicators based on time series analysis, and automatically generates risk prevention and optimization management suggestions to help users address potential problems in advance.

[0060] The personalized suggestion generation module generates implementation suggestions based on real-time collected data, knowledge graph data, and aquaculture goals, using reinforcement learning algorithms to meet users' personalized needs and improve aquaculture efficiency.

[0061] The knowledge graph construction module provides the theoretical foundation for other modules, while the continuous learning and knowledge iteration module acquires expert experience to update the knowledge graph, collaboratively handles new knowledge points and knowledge conflicts, and ensures the continuous improvement and development of the knowledge graph.

[0062] Therefore, the present invention employs the above-mentioned intelligent advisory system and method for aquaculture, and the technical effects are as follows:

[0063] First, this invention constructs a systematic aquaculture knowledge graph, integrating scattered professional knowledge and experience into a structured knowledge network. Through ontological modeling methods, it systematically covers key entities such as farmed species, diseases, environmental parameters, and treatment plans, along with their complex relationships, making professional knowledge more standardized, structured, and accessible, providing comprehensive and systematic knowledge support for aquaculture farmers.

[0064] Secondly, this invention achieves multimodal interaction and intelligent recognition capabilities, supporting multiple input formats such as images, videos, audio, and text. Users can intuitively describe problems by uploading images of the breeding environment, videos of equipment operation, or breeding records. The system can accurately identify the breeder species, disease symptoms, and environmental anomalies, greatly reducing the threshold for technical communication and improving the accuracy of problem diagnosis.

[0065] Third, the context-aware reasoning engine of this invention possesses diagnostic and decision-making capabilities similar to those of an expert. Through graph neural network technology, the system can analyze the complex relationships between environmental parameters, symptom manifestations, and historical data, simulating the expert's thinking process and completing the reasoning from problem description to solution in a short time, providing timely and professional technical support to farmers.

[0066] Fourth, this invention provides highly personalized aquaculture recommendations, generating customized solutions based on the user's aquaculture history, regional characteristics, and aquaculture goals. The system considers multiple aquaculture aspects such as water quality management, feed feeding, and disease prevention and control, providing detailed parameter settings, operating procedures, and expected results to help farmers develop scientific and reasonable aquaculture strategies.

[0067] Fifth, this invention possesses continuous learning and knowledge iteration capabilities, enabling it to constantly absorb new research findings, expert experience, and user feedback. Through a federated learning architecture, the system extracts valuable information from each interaction and updates the knowledge base while protecting user data privacy, thus continuously improving the system's professional capabilities.

[0068] Sixth, this invention enables predictive aquaculture management by constructing a virtual model of the aquaculture environment using digital twin technology to predict future changes in aquaculture indicators and potential risks. This forward-looking management approach enables farmers to proactively prevent problems, optimize resource allocation, and improve aquaculture efficiency and economic benefits.

[0069] Finally, the system of this invention possesses an adaptive optimization mechanism and local caching capabilities, enabling it to continuously optimize model parameters based on user feedback and aquaculture results, and automatically adjust reference thresholds according to regional and seasonal changes, ensuring the system's applicability in various environments. Simultaneously, the local caching mechanism guarantees that the system can still function normally in aquaculture sites with unstable networks, meeting practical application needs. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of an intelligent aquaculture advisory system according to the present invention;

[0071] Figure 2 A schematic diagram of the knowledge graph construction module;

[0072] Figure 3 This is a schematic diagram of the multimodal input processing module;

[0073] Figure 4 This is a schematic diagram of a context-aware reasoning engine;

[0074] Figure 5 A diagram illustrating the module for generating personalized suggestions;

[0075] Figure 6 This is a module for continuous learning and knowledge iteration.

[0076] Figure 7 This is a predictive aquaculture management module;

[0077] Figure 8 This is an adaptive optimization module. Detailed Implementation

[0078] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0079] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0080] Example 1

[0081] like Figure 1 As shown, an intelligent aquaculture advisory system includes several main functional modules: a multimodal input processing module, a context-aware reasoning engine, an expert collaboration platform, an adaptive optimization module, a predictive aquaculture management module, a knowledge graph construction module, and a continuous learning and knowledge iteration module. These modules are interconnected through a data interaction interface to jointly complete the intelligent identification, analysis, and resolution of aquaculture problems.

[0082] like Figure 2 As shown, the knowledge graph construction module provides a theoretical basis for the context-aware reasoning engine, the predictive aquaculture management module, and the personalized suggestion generation module.

[0083] The ontological modeling method used in the knowledge graph construction module includes the following steps:

[0084] The knowledge in the field of aquaculture is divided into core categories, and subcategories are set up under each core category.

[0085] Establish core category attribute sets for various entities based on the core category definition;

[0086] Establish the relationship between core categories and core category attribute sets to obtain the corresponding semantic relationships;

[0087] By extracting and integrating basic academic knowledge, applied knowledge, and expert experience from expert collaboration platforms, professional knowledge is obtained.

[0088] Using core categories and core category attributes as knowledge graph nodes, the semantic relationships between core categories and core category attribute sets are analyzed. Based on the acquired professional knowledge, connections between knowledge graph nodes are established to form a first-level knowledge network. On the basis of the first-level knowledge network, connections between sub-class structures are established to form a second-level knowledge network. A hierarchical design is adopted to construct general knowledge and regionally specialized knowledge separately, and automatic fusion is achieved through association rules to balance the generality and specificity of knowledge. Together with the first-level knowledge network and the second-level knowledge network, a knowledge graph is formed.

[0089] The knowledge graph construction module employs ontology-based modeling to systematically organize knowledge in the aquaculture field. First, a top-level ontology framework is constructed, dividing aquaculture knowledge into five core categories: species classification, environmental parameters, disease pathology, treatment plans, and aquaculture techniques. Taking species classification as an example, a three-tiered subclass structure is established, comprising species (e.g., fish, crustaceans), biological characteristics (e.g., growth cycle, diet), and aquaculture characteristics (e.g., density requirements, economic value).

[0090] In the entity attribute definition phase, the system establishes standardized attribute sets for various entities. Taking tilapia as an example, its species entity includes 15 basic attributes such as scientific name, distribution area, growth cycle, optimal water temperature, dissolved oxygen requirement, and pH tolerance range; the streptococcal disease entity includes 12 attributes such as symptom manifestations (protruding eyes, bleeding points on the body surface), pathogenesis, transmission route, and susceptible stage; the dissolved oxygen environmental parameter entity includes 8 attributes such as parameter definition, measurement method, normal range, and danger threshold.

[0091] The relation type construction defines 20 semantic relations, including "cause" (e.g., low dissolved oxygen causes hypoxia), "applies to" (e.g., chloramphenicol is applicable to bacterial septicemia), "affects" (e.g., water temperature affects feed intake), "is a kind of" (e.g., grass carp is a kind of cyprinid fish), and "requires" (e.g., shrimp larvae cultivation requires light regulation), which are used to describe complex relationships between different entities.

[0092] Knowledge extraction and integration automatically acquires professional knowledge from three types of knowledge sources using natural language processing technology: basic academic knowledge is extracted from 50 professional textbooks such as "Aquaculture" and "Fish Pathology" and 500 core journal articles published by the China Fisheries Society; applied knowledge is extracted from 300 technical manuals published by the National Fisheries Technology Extension Center; and experiential knowledge is extracted from 50 senior aquaculture experts through structured interviews. The system uses named entity recognition and relation extraction algorithms to transform unstructured text into entities and relations in a knowledge graph.

[0093] The knowledge network structure covers 100 key aquaculture species (including grass carp, tilapia, and shrimp) in major aquaculture areas of China, 200 common diseases (including bacterial, viral, and parasitic diseases), 50 environmental parameters (water quality, bottom sediment, and meteorology), and 300 treatment options (drug, physical, and biological methods), forming a knowledge network of 10,200 entity nodes and 32,500 relational edges, providing a solid knowledge foundation for the system's reasoning and suggestion generation.

[0094] like Figure 3 As shown, the multimodal input processing module is used to collect, analyze, and digitize user-end input and real-time collected data; user-end input includes aquaculture goals or aquaculture status;

[0095] The multimodal input processing module consists of the following units:

[0096] Set up a data acquisition unit, including user input on the user terminal and setting up acquisition devices for real-time data acquisition;

[0097] An image processing unit is set up to process images from user input and real-time acquired data;

[0098] A video analysis unit is set up to process video streams from user input and real-time data acquisition, dynamically analyze the group behavior patterns of farmed organisms, and maintain video analysis performance under insufficient lighting conditions through an adaptive brightness enhancement algorithm.

[0099] An audio processing unit is set up to process audio data from user input and real-time data acquisition, extract acoustic features, and analyze the operating status of aquaculture equipment and the sounds of abnormal aquatic animal activities.

[0100] A text processing unit is set up to combine optical character recognition and natural language processing technologies to perform digital processing and structured extraction on user input and real-time collected data, and to transform unstructured text into structured data through named entity recognition and relation extraction technologies.

[0101] The multimodal fusion unit maps feature vectors from different modalities to a unified semantic space, enhancing the complementarity of information between modalities. At the same time, it uses a modality-specific confidence scoring mechanism to dynamically weight the contribution of each modality, outputting a unified semantic representation.

[0102] The multimodal input processing module supports various input formats such as images, videos, audio, and text, enabling comprehensive perception of the aquaculture scene.

[0103] The image processing unit employs a deep convolutional neural network based on the ResNet-152 and EfficientNet-B7 architecture to process aquaculture environment images with a resolution of 3 megapixels (2048×1536). For example, if a user uploads an image of fish in a fishpond, the system can identify grass carp in the image and detect red spots and fin rot symptoms on the fish's surface, initially diagnosing it as "red spot disease," with an accuracy rate of 95%. The image processing unit uses a five-level multi-scale feature pyramid network to simultaneously analyze visual information from three regions: the water surface (detecting floating debris and oil film), the water body (observing transparency and color), and the pond bottom (identifying the state of the bottom sediment), providing a basis for a comprehensive assessment of the aquaculture environment.

[0104] The video analytics unit supports high frame rate video streaming processing of 60 frames per second to analyze the group behavior patterns of farmed organisms. For example, by analyzing the swimming speed, grouping degree, and surfacing behavior of grass carp schools, the system can assess the dissolved oxygen status of the water; by monitoring changes in feeding activity, it can determine feed palatability and fish health. The video analytics unit employs an architecture combining 3D convolutional neural networks and long short-term memory networks, maintaining good analytical performance even under low-light conditions through an adaptive brightness enhancement algorithm.

[0105] The audio processing unit uses 16-bit quantization and a 48kHz sampling rate to collect sounds from the aquaculture environment, enabling it to identify equipment malfunctions and changes in aquatic animal behavior. For example, through sound feature analysis, the system can detect abnormal noise caused by an imbalance in the aerator impeller, providing early warning of equipment failure; it can also identify the distinctive splashing sounds emitted by certain fish when oxygen is insufficient, serving as an early warning signal for water quality problems.

[0106] The text processing unit combines optical character recognition (OCR) and natural language processing (NLP) technologies to digitize user-submitted aquaculture records. For example, if a user uploads a handwritten feeding record, the system can automatically identify information such as date, feed type, and amount fed, converting it into structured data and storing it in the system to provide a foundation for subsequent analysis. The text processing unit supports extracting key information from unstructured descriptions, such as "Yesterday morning, several fish were found surfacing in the northwest corner of the pond, and there was foam on the surface," which can be parsed into a structured representation of location, time, and symptoms.

[0107] The multimodal fusion unit employs a Transformer-based cross-modal attention mechanism to integrate information from different modalities into a unified semantic representation. For example, when a user simultaneously provides pond images, fish behavior videos, and water quality description text, the system can comprehensively analyze this information to form a complete understanding of the aquaculture situation. Even when some modal information is missing, the system can still perform analysis using the existing modalities, maintaining over 85% of its performance.

[0108] like Figure 4 As shown, the context-aware reasoning engine is used to extract features from the implementation data and aquaculture targets processed by the multimodal input processing module, analyze the correlation between environmental data and aquaculture data, and obtain the initial reasoning results.

[0109] Context-aware reasoning engines include:

[0110] The data processed by the multimodal input processing module is transformed into a graph structure and aligned with the knowledge graph nodes;

[0111] Feature extraction is performed on the graph structure, and the association patterns of graph structure features and knowledge graph nodes are analyzed. Causal chains are constructed for graph structure features and knowledge graph nodes, and weights are calculated.

[0112] By combining the temporal and spatial information of the causal chain with the structural features of the graph, a comprehensive representation of the current state of the aquaculture environment is constructed.

[0113] Through reverse reasoning and template filling, targeted solutions and operational suggestions are generated.

[0114] The context-aware reasoning engine is based on knowledge graphs and environmental data, and uses graph neural network technology to build reasoning models to achieve intelligent diagnosis and solution generation for aquaculture problems.

[0115] The inference model employs a hierarchical graph neural network architecture, specifically designed to solve complex problems in aquaculture. For example, if a user reports that their farmed grass carp are exhibiting symptoms such as slow swimming, loss of appetite, and pale gills, the system can analyze these symptoms in conjunction with environmental parameters (water temperature 24℃, dissolved oxygen 3.5mg / L, pH 8.2), and through inference determine that the possible problem is hypoxia or bacterial gill disease, and provide targeted treatment suggestions.

[0116] The five layers of the graph convolutional network work together: the input encoding layer transforms the user-described symptoms and environmental data into a graph structure, aligning it with relevant nodes in the knowledge graph; the feature extraction layer learns the association features between symptoms, environmental parameters, and possible diseases through graph convolution operations; the relational reasoning layer analyzes the causal chains between different entities, such as the reasoning path of low dissolved oxygen → stress response → decreased immunity → bacterial infection; the context integration layer combines spatiotemporal information such as recent meteorological changes (e.g., continuous high temperatures) to construct a more complete state representation; and the solution generation layer generates a comprehensive solution based on the reasoning results, including emergency oxygenation, water quality improvement, and preventive medication.

[0117] The 256 feature channels enable the model to capture complex knowledge representations within aquaculture: 64 channels represent the physiological characteristics, growth stages, and sensitive indicators of species such as grass carp; 64 channels represent the normal ranges, trends, and interactions of environmental parameters such as water temperature, dissolved oxygen, and ammonia nitrogen; 64 channels encode the symptoms, onset conditions, and transmission mechanisms of various diseases; and 64 channels represent the applicable conditions, operational methods, and expected effects of management measures such as oxygenation, water conditioning, and medication. This high-dimensional feature representation ensures that the model remains sensitive to subtle but critical changes in the aquaculture process.

[0118] Correlation analysis is achieved through a multi-head graph attention mechanism, with eight attention heads focusing on different types of correlation patterns. For example, when analyzing the problem of grass carp surfacing, different attention heads may focus on different relationships, such as the direct impact of ammonia nitrogen concentration, the synergistic effect of water temperature and dissolved oxygen, and the indirect impact of excessive algae growth. Through Shapley value analysis, the system identified insufficient dissolved oxygen (65%), excessively high water temperature (23%), and elevated ammonia nitrogen (12%) as the three main factors causing the problem, and prioritized these factors according to their degree of impact in the generated solution.

[0119] The reasoning process employs a two-stage strategy to ensure timely response. For example, if a user reports that shrimp in the pond are turning red and exhibiting reduced activity, the system will make a preliminary judgment within 300 milliseconds, stating that "it may be a bacterial disease, and it is recommended to immediately test the water quality and shrimp." Simultaneously, deep reasoning will be initiated, and within the next 700 milliseconds, the system will complete the probability analysis and differential diagnosis of specific diseases such as red body disease and hepatopancreatic necrosis, ultimately generating a complete set of recommendations including emergency treatment, medication plans, and subsequent management.

[0120] like Figure 5 As shown, the personalized suggestion generation module is used to generate implementation suggestions based on real-time collected data, knowledge graph data, and aquaculture goals using reinforcement learning algorithms;

[0121] The personalized suggestion generation module includes:

[0122] The reinforcement learning algorithm is used to generate implementation suggestions based on user input, real-time data collection and real-time graphs. Each implementation suggestion contains three levels of implementation details: the first level is the setting of core parameters, the second level is the operation method description, and the third level is the expected effect and risk assessment.

[0123] The reinforcement learning algorithm adopts a policy optimization framework based on deep Q-networks. It uses real-time monitoring data and regional characteristics of the knowledge graph to model a Markov decision process. The user input is used as the reward function. The system selects the best case in the breeding history of the knowledge graph. By learning from the best case, it forms an optimal policy library for different situations. At the same time, the Monte Carlo tree search method is introduced to evaluate the long-term decision path to ensure the long-term effectiveness of the suggestions. The system updates the Q value in real time from user feedback and continuously optimizes the decision model.

[0124] The personalized suggestion generation system uses reinforcement learning algorithms to generate targeted suggestions based on the user's breeding history, regional characteristics, and breeding goals, providing scientific guidance for the entire breeding process.

[0125] The system generates personalized recommendations based on data integration across four dimensions: analyzing the user's aquaculture data from the past three years, such as the average survival rate of tilapia for a southern farmer being 85%, the feed conversion ratio being 1.6, and the main disease being streptococcal disease; considering regional characteristics, such as the long high-temperature period in the region during summer, high rainfall, and groundwater as the water source; combining aquaculture goals, such as a planned harvest time of 120 days, a target size of 500 grams per fish, and a cost control of 12 yuan per kilogram; and integrating real-time monitoring data, such as the current water temperature of 28℃, pH value of 7.8, and dissolved oxygen of 5.6 mg / L.

[0126] Reinforcement learning algorithms model the aquaculture process as a Markov decision process, learning from over 10,000 successful cases to form an optimal strategy library for different scenarios. For example, the feeding strategy developed by the system for the aforementioned farmers takes into account the dissolved oxygen risk caused by high temperatures in southern summers, suggesting feeding twice a day, in the early morning and evening, controlling the amount of each feeding to no more than 60% of the total daily feed, and appropriately reducing the amount of feed before predicted rainfall. These strategies are evaluated for long-term effects through Monte Carlo tree search, ensuring that the pursuit of growth rate does not increase disease risk.

[0127] The system provides parameterized guidance for 15 key aquaculture stages, but dynamically adjusts its focus based on current conditions. For example, in the early stages of aquaculture, the emphasis is on water quality management and seedling cultivation; in the mid-growth stage, the focus is on feed delivery and disease control; and in the pre-harvest stage, the focus is on growth regulation and economic benefit analysis. This dynamic adjustment mechanism ensures that the system's recommendations always align with the actual needs of the aquaculture cycle.

[0128] Based on the urgency and complexity of the inquiries, the system employs a three-tiered response mechanism. For example, when a user reports an urgent issue such as "sudden mass mortality in the shrimp pond," the system generates an emergency plan within 5 seconds, including specific measures and operational steps such as immediate full-pond aeration, emergency 20% water replacement, and the use of bottom-improving bacteria. For routine inquiries like "how to improve feed conversion ratio," the system generates standardized suggestions within 10 seconds, including parameter ranges such as feed brand selection, feeding frequency (4-6 times / day), and feed particle size (3-5mm). For strategic questions like "how to plan for the next quarter's farming," the system generates a comprehensive plan within 30 seconds, encompassing multiple aspects such as seedling selection, stocking density, vaccination, and phased feeding.

[0129] Each recommendation provides three levels of implementation details. Taking water quality adjustment recommendations as an example, the first level sets core parameters, such as "add 50 kg / mu of quicklime in water at a depth of 1.5 meters"; the second level provides instructions on operation, such as "dissolve quicklime in clean water and evenly sprinkle it along the edge of the pond, avoiding areas where fish congregate, and operate between 9 and 11 am"; the third level provides expected effects and risk assessment, such as "expecting a pH increase of 0.5-1.0 and an alkalinity increase of 50 mg / L within 24 hours, which may lead to a temporary reduction in plankton, and it is recommended to supplement with microbial agents after 2 days." This layered presentation allows users with different experience levels to find suitable operational guidance.

[0130] The visual interface features a responsive design, delivering optimal performance across various devices. The system transforms complex parameter adjustment suggestions into intuitive charts, such as water quality parameter trend graphs (showing changes in dissolved oxygen and pH over the past 7 days and predicted for the next 3 days), feed feeding curves (displaying the correspondence with fish growth stages), and disease risk heat maps (intuitively displaying the probability of disease risk under different combinations of environmental parameters). This allows fish farmers to readily understand the scientific basis and expected effects of the system's recommendations.

[0131] like Figure 6 As shown, the continuous learning and knowledge iteration module is used to acquire expert experience from the expert collaboration platform and use it to update the knowledge graph construction module; when discovering potential new knowledge points in the system or identifying knowledge conflicts, it collaborates with the expert collaboration platform to provide authoritative verification and arbitration.

[0132] The continuous learning and knowledge iteration module includes:

[0133] The system processes user interaction records, expert knowledge supplements from the expert collaboration platform, knowledge graphs, and real-time collected data using a multi-stage pipeline architecture to form a knowledge base.

[0134] The multi-stage pipeline architecture processes data cleaning and standardization using a multimodal input processing module, feature extraction and semantic annotation using a context-aware reasoning engine, and knowledge conflict detection, knowledge fusion and value assessment using an expert collaboration platform.

[0135] Meanwhile, the knowledge base also has a local device-based hierarchical storage architecture to store data processed by the multi-stage pipeline architecture locally.

[0136] The knowledge base also includes a central server that processes data from a multi-stage pipeline architecture and updates it in real time, supporting rollback functionality based on time series or specific event sequences.

[0137] The continuous learning and knowledge iteration module adopts a federated learning architecture, which can extract valuable information from each interaction and update the knowledge base while protecting user data privacy, thus enabling the continuous evolution of system capabilities.

[0138] We continuously acquire new knowledge from five types of data streams: collecting approximately 12,000 user interaction records monthly, such as inquiries from farmers about "how to improve the survival rate of Litopenaeus vannamei during the high-temperature season" and subsequent operational feedback; obtaining approximately 400 pieces of high-quality supplementary knowledge through an expert collaboration platform, such as "the usage methods and precautions of new water quality stabilizers"; automatically crawling the latest research and technical reports from 40 journals, including the Journal of Fisheries of China, and 15 industry websites, such as China Fisheries Portal, weekly; integrating real-time data uploaded by IoT devices, with each farm generating approximately 2GB of raw data daily, including 24-hour water temperature, dissolved oxygen, and pH value change curves; and collecting regional epidemic and environmental monitoring data, such as "early warning of hemorrhagic disease outbreak in crucian carp in South China".

[0139] The data processing adopts a multi-stage pipeline architecture: the data cleaning stage filters out abnormal values ​​from water quality sensors and incomplete user feedback; the feature extraction stage converts descriptive text such as "water quality improved significantly after feeding EM bacteria" into a structured relationship of "EM bacteria → water quality improvement (effect = significant)"; knowledge conflict detection identifies a potential contradiction between two suggestions for "feeding probiotics during high-temperature periods," one recommending "once in the morning and once in the evening" and the other recommending "feeding only in the evening"; knowledge fusion evaluation determines that the latter is more applicable in high-temperature areas in the south, scoring 85 points, while the former scores 65 points. The system retains the knowledge point with the higher score and records the applicable conditions.

[0140] The federated learning architecture ensures user data privacy and security. For example, environmental data and operational records of a certain aquaculture farm are always stored locally. The system trains the model locally and extracts abstract patterns and rules, such as discovering rules like "under the conditions of <region = Pearl River Delta, season = summer, farmed species = Litopenaeus vannamei>, reducing daily feed by 15% and increasing nighttime aeration time can improve survival rate by 8%." These abstract knowledge patterns are processed using differential privacy technology and then uploaded to a central server, where they are aggregated and analyzed with other users' experience patterns to form more reliable knowledge rules, while ensuring that the specific data of individual aquaculture farms is not leaked.

[0141] The knowledge database features a hierarchical storage architecture: the hot data layer stores frequently accessed knowledge such as "prevention and control of common diseases during the summer heat" and the latest environmental data; the warm data layer stores nearly 90 days of user consultation history and feedback; the cold data layer stores complete historical data and large multimedia resources such as teaching videos; and the metadata layer maintains the structure information and version change records of the knowledge graph. When a user inquires about questions such as "how to deal with the inability of Litopenaeus vannamei to molt," the system can complete knowledge retrieval and matching in milliseconds and provide the latest solutions.

[0142] Knowledge base updates are implemented through a tiered iterative strategy: when critical information such as new methods for detecting highly pathogenic viruses is discovered, the system completes an emergency update within 5 minutes; regular incremental updates are performed every 24 hours, such as adding newly discovered "water quality-disease correlation" knowledge points to the knowledge graph; and the knowledge graph structure is optimized every 30 days, such as reorganizing the "disease prevention and control" classification system to improve retrieval efficiency. The system automatically retains a 30-day version history; for example, if a newly added water treatment method is found to have side effects, it can quickly roll back to the knowledge state before the problem was discovered, ensuring system reliability.

[0143] like Figure 7 As shown, the predictive aquaculture management module is used to predict key aquaculture indicators in the future by constructing a virtual aquaculture environment model and performing time series analysis, and automatically generate risk prevention and optimization management suggestions based on the prediction results.

[0144] The predictive aquaculture management module includes:

[0145] Spatial scanning of the aquaculture farm generates a three-dimensional terrain model. Data collected in real time by the multimodal input processing module is used to obtain water quality parameters, meteorological parameters, and biological characteristic parameters of the aquaculture farm.

[0146] By utilizing a three-dimensional terrain model and water quality parameters, meteorological parameters, and biological characteristic parameters, a coupled model of hydrodynamics, water quality change, biological growth, and meteorological influence is constructed, namely a virtual aquaculture environment model, to achieve real-time mapping between the virtual and real environments;

[0147] The virtual aquaculture environment model uses a multi-scale fusion method for prediction over time series data;

[0148] The virtual aquaculture environment model is optimized through a multi-objective decision-making algorithm, taking the aquaculture objectives input by the user as the optimization objective and considering the constraints set by the user.

[0149] The predictive aquaculture management module is based on digital twin technology, which constructs a virtual model of the aquaculture environment and uses time series analysis to predict key future aquaculture indicators, thereby enabling a shift from passive response to proactive management.

[0150] Digital twin technology first uses lidar to perform high-precision spatial scanning of the aquaculture farm, reconstructing a three-dimensional terrain with a resolution of 0.1 meters, including the precise outline and depth distribution map of a 5-acre shrimp pond with a depth of 3 meters. Then, a sensor network is deployed, including an underwater sensor array (monitoring parameters such as temperature, dissolved oxygen, and pH at different water layers), a weather station (recording factors such as air temperature, light intensity, and wind speed), and an aquatic organism monitoring system (tracking shrimp growth and behavior). Finally, a coupled model system incorporating hydrodynamics, water quality changes, biological growth, and meteorological influences is constructed, updating data every 5 minutes to synchronize the virtual model with the actual aquaculture environment.

[0151] The virtual aquaculture environment model consists of five sub-models: the aquatic environment sub-model displays the regional distribution of dissolved oxygen dropping to the critical value of 3.5 mg / L at 6 AM in the form of a 3D heatmap, and simulates the effects of different aerator layout schemes; the meteorological impact sub-model predicts that heavy rainfall in the next 3 days will lead to a risk of a 0.8 decrease in pH and a 25% increase in ammonia nitrogen; the biological population sub-model shows the deviation of the growth curve of Litopenaeus vannamei from the standard growth curve and correlates it with a decrease in feeding activity; the feed dynamics sub-model calculates the sedimentation path and decomposition rate of the fed feed and identifies potential dead zones; and the equipment operation sub-model monitors the abnormal 20% efficiency decrease of aerator No. 2 and simulates the potential impact of further deterioration of the malfunction on the aquatic environment. Through this virtual environment, farmers can intuitively understand the complex system dynamics and test different management strategies without interfering with actual production.

[0152] Time series analysis employs a multi-scale fusion approach for forecasting: short-term forecasts (within 7 days) use an LSTM deep learning model to accurately predict "the lowest dissolved oxygen level of 2.8 mg / L will occur between 2-4 AM on Wednesday"; medium-term forecasts (within 14 days) combine wavelet decomposition and random forest algorithms to predict "water temperature will steadily rise to 32℃ within the next 10 days, and the algal community will change"; long-term forecasts (within 30 days) use a seasonal ARIMA model to predict "the cultured organisms will reach 24g / tail by the end of this month, requiring adjustment of feed particle size." The system achieves prediction accuracy rates of 92%, 88%, and 85% for 12 indicators, including dissolved oxygen, pH, and ammonia nitrogen, respectively, providing a scientific basis for aquaculture management decisions.

[0153] The risk control mechanism achieves proactive management based on threshold early warning and trend analysis. The system defines three risk levels: a "Caution" level warning is triggered when dissolved oxygen is predicted to drop to 3.5-4.0 mg / L within the next 48 hours, suggesting appropriate control of feeding; a "Warning" level warning is triggered when dissolved oxygen is predicted to drop to 2.5-3.5 mg / L, suggesting increased nighttime aeration and preparation of emergency equipment; and an "Emergency" level warning is triggered when dissolved oxygen is predicted to fall below 2.5 mg / L, suggesting immediate implementation of comprehensive aeration and emergency water exchange measures. The system identifies the cumulative risk of a slow increase in hydrogen sulfide concentration in the aquaculture pond and provides bottom sediment improvement suggestions 10 days before reaching a dangerous level, preventing a crisis.

[0154] The optimized management recommendations are generated through a multi-objective decision-making algorithm, balancing economic benefits, resource consumption, and risk control. The system provides three types of management recommendations: For a predicted period of high temperatures, a preventative recommendation is generated: "Starting next week, increase the shade net coverage area to 40% of the water surface and adjust the daily feeding times to three times a day: 6:00, 19:00, and 22:00, with proportions of 20%, 30%, and 50% respectively." For situations where growth rates are lower than expected, an adjustment recommendation is made: "Current protein intake is insufficient; it is recommended to increase the feed protein content from 38% to 42%, and supplement with appropriate amounts of vitamin C to enhance stress resistance." For potential typhoon weather, an emergency plan is prepared: "12 hours before the typhoon arrives, reinforce aeration equipment, increase the water level by 30cm, and prepare emergency generators to ensure the continued operation of critical equipment during power outages." Simulations show that adopting the system's recommendations can increase yield by 15%, reduce feed costs by 8%, and reduce disease incidence by 23% compared to traditional management methods.

[0155] like Figure 8 As shown, the adaptive optimization module is used to compare the consistency between user-implemented operations and system suggestions, and optimize the model parameters of the context-aware inference engine based on the consistency results; it is also used to adaptively adjust the context-aware inference engine according to the region and season; and it is used to cache the optimization and adjustment schemes locally to ensure offline operation.

[0156] The adaptive optimization module includes:

[0157] When comparing the consistency between user actions and system recommendations, the core aquaculture parameters (including feeding amount, feeding frequency, water quality adjustment parameters, disease prevention parameters, etc.) output by the personalized recommendation generation module are automatically adjusted using a two-dimensional matrix of region and season. The adjustment range is determined based on statistical analysis of historical data. At the same time, the early warning threshold of the predictive aquaculture management module and the judgment threshold of the context-aware inference engine are automatically updated according to the latest weather forecast data.

[0158] The key parameters (water quality management threshold, feeding control parameters, disease early warning parameters), core models (graph neural network parameters of the context-aware reasoning engine, reinforcement learning model weights of the personalized suggestion generation module, and time series prediction model parameters of the predictive aquaculture management module), and knowledge base updates for each optimization are cached locally. The adaptive optimization module, predictive aquaculture management module, and continuous learning and knowledge iteration module all have local caching functions and automatically update data through a differentiated synchronization mechanism.

[0159] The adaptive optimization module can continuously optimize model parameters and knowledge base by comparing the differences between the user's actual operation results and the system's suggestions, thereby achieving continuous evolution of the system's capabilities.

[0160] User operation data is acquired through three methods: intelligent sensing monitoring, which automatically records actual parameters performed by users through IoT devices deployed in the aquaculture farm (e.g., a smart feeder records a feeding amount of 25 kg per feeding, while the system recommends 20 kg); natural behavior analysis, which uses edge computing cameras to identify the actions of aquaculture workers at the pond edge (e.g., detecting that a user actually performed water quality adjustments but used a dosage of medication 30% higher than recommended); and comparison of aquaculture results, which tracks changes in key indicators (e.g., the system predicts that ammonia nitrogen in a shrimp pond should decrease by 50% within 7 days after adopting the recommended treatment, but it only decreased by 30%, indirectly reflecting that the user may not have fully implemented the bottom sediment improvement measures as recommended). This data helps the system understand the discrepancy between user behavior and system recommendations, providing a basis for model optimization.

[0161] User feedback is obtained through a combination of implicit perception and proactive guidance: the biocomputing interface uses the camera of smart devices to capture micro-expression changes when users view suggestions. For example, if the system detects a user's expression of doubt regarding dissolved oxygen management suggestions, it automatically provides more detailed scientific explanations and operational videos. Operational behavior mirroring records the user's behavioral trajectory after receiving suggestions. For example, if a user spends a long time viewing disease prevention suggestions, returns to view them multiple times, and implements them quickly, it indicates a high acceptance rate for the suggestions. Results are tracked in a closed loop to compare expected and actual results. For example, if the system's suggested water quality control plan is expected to increase transparency by 15cm, and the actual effect is 17cm, a value coefficient of 1.13 is calculated, indicating that the suggestion is more effective than expected.

[0162] The consistency assessment is based on a five-dimensional indicator system, comprehensively evaluating the actual implementation and effectiveness of the system's recommendations: consistency of operational parameters (the user's actual quicklime usage was 45 kg / mu, while the recommended amount was 50 kg / mu, resulting in a consistency rate of 90%); consistency of operational timing (the user performed water quality adjustments 2 hours after the recommended time point, resulting in a timing consistency rate of 80%); consistency of operational methods (the user evenly sprinkled the quicklime along the edge of the pond according to the recommended method, resulting in a method consistency rate of 95%); consistency of effect achievement (the pH value actually increased by 0.8, while the expected increase was 1.0, resulting in an achievement consistency rate of 80%); and consistency of user acceptance (the user adopted 12 out of the system's 15 recommendations, resulting in an acceptance consistency rate of 80%). The system calculated an overall consistency score of 85%, which is higher than the 80% trigger threshold, indicating that the current model parameters are well-suited and do not require immediate updates. However, the system will still record all deviation data for subsequent continuous optimization.

[0163] The threshold adjustment mechanism automatically adapts to regional and seasonal changes: For regional adaptability, the system integrates multi-source data to construct a regional feature database. For example, the suitable dissolved oxygen range for subtropical regions in the south is 4-8 mg / L, while for temperate regions in the north it is 5-10 mg / L. For seasonal adaptability, the system establishes a phenological model for aquaculture, dividing the grass carp farming cycle into four stages: overwintering period (water temperature <15℃, dissolved oxygen threshold >6 mg / L), spring transition period (water temperature 15-22℃, dissolved oxygen threshold >5 mg / L), vigorous growth period (water temperature 22-30℃, dissolved oxygen threshold >4 mg / L), and autumn adaptation period (water temperature 15-22℃, dissolved oxygen threshold >5 mg / L). The system automatically adjusts parameter thresholds based on the region-season matrix. For example, the safe dissolved oxygen threshold in summer for the same region can be 0.5-1.0 mg / L lower than in winter, ensuring that the system's recommendations always conform to local conditions.

[0164] The local caching mechanism is based on a layered edge computing architecture, ensuring the system can still function normally in aquaculture sites with unstable networks: At the hardware level, edge computing nodes are deployed at user terminals and farms, equipped with sufficient storage and computing resources; at the data level, a three-level caching strategy is adopted to store critical environmental data, core inference models, and commonly used knowledge bases; at the functional level, core functions such as basic environmental monitoring, common problem diagnosis, and emergency handling are prioritized; and a synchronization mechanism ensures data consistency after network recovery, prioritizing data versions modified locally by users. This design enables the system to adapt to the complex and ever-changing network environment of frontline aquaculture, providing uninterrupted technical support.

[0165] A smart advisory method for aquaculture includes the following steps:

[0166] The multimodal input processing module collects user input and real-time acquired data, and performs analysis and digital processing.

[0167] The context-aware reasoning engine extracts features from the processed data and aquaculture objectives, analyzes the correlation between the environment and aquaculture data, and draws initial reasoning results to provide preliminary basis for aquaculture decision-making.

[0168] The expert collaboration platform incorporates user operations, updates the knowledge graph through continuous learning and knowledge iteration modules, assists in knowledge conflict detection, integration, and value assessment, and ensures the accuracy and timeliness of knowledge.

[0169] The adaptive optimization module compares the consistency between user operations and system suggestions, optimizes the parameters of the context-aware inference engine model, and adjusts the inference engine according to regional and seasonal adaptability. The optimization and adjustment schemes are cached locally to ensure offline operation and improve system adaptability and stability.

[0170] The predictive aquaculture management module constructs a virtual aquaculture environment model, predicts key future aquaculture indicators based on time series analysis, and automatically generates risk prevention and optimization management suggestions to help users address potential problems in advance.

[0171] The personalized suggestion generation module generates implementation suggestions based on real-time collected data, knowledge graph data, and aquaculture goals, using reinforcement learning algorithms to meet users' personalized needs and improve aquaculture efficiency.

[0172] The knowledge graph construction module provides the theoretical foundation for other modules, while the continuous learning and knowledge iteration module acquires expert experience to update the knowledge graph, collaboratively handles new knowledge points and knowledge conflicts, and ensures the continuous improvement and development of the knowledge graph.

[0173] Therefore, the present invention employs the above-mentioned intelligent aquaculture advisory system and method, which can understand and solve complex aquaculture problems, provide accurate and personalized suggestions, predict and prevent potential risks, realize the intelligent transmission and application of aquaculture knowledge, and provide strong support for improving aquaculture efficiency and success rate.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent advisory system for aquaculture, characterized in that, include The multimodal input processing module is used to collect, analyze, and digitize user-end input and real-time acquired data; user-end input includes aquaculture goals or aquaculture status; The context-aware reasoning engine is used to extract features from the implementation data and aquaculture targets processed by the multimodal input processing module, analyze the correlation between environmental data and aquaculture data, and obtain the initial reasoning results. The expert collaboration platform is used to absorb user actions collected by the multimodal input module and update the knowledge graph construction module through the continuous learning and knowledge iteration module. It is used to assist the continuous learning and knowledge iteration modules in knowledge conflict detection, knowledge fusion and value assessment; The adaptive optimization module is used to compare the consistency between user actions and system suggestions, and optimize the model parameters of relevant core modules of the context-aware inference engine based on the consistency results. Used to adaptively adjust the context-aware reasoning engine according to region and season; used to cache optimization and adjustment schemes locally to ensure offline operation; The predictive aquaculture management module is used to predict key aquaculture indicators in the future by constructing a virtual aquaculture environment model and performing time series analysis, and automatically generate risk prevention and optimization management suggestions based on the prediction results; The personalized suggestion generation module is used to generate implementation suggestions based on real-time collected data, knowledge graph data, and aquaculture goals using reinforcement learning algorithms; The knowledge graph construction module provides a theoretical foundation for the context-aware reasoning engine, the predictive aquaculture management module, and the personalized suggestion generation module. The continuous learning and knowledge iteration module is used to acquire expert experience from the expert collaboration platform and use it to update the knowledge graph construction module. When discovering potential new knowledge points in the system or identifying knowledge conflicts, it collaborates with expert platforms to provide authoritative verification and arbitration.

2. The intelligent advisory system for aquaculture according to claim 1, characterized in that, The ontology-based modeling method used in the knowledge graph construction module includes the following steps: The knowledge in the field of aquaculture is divided into core categories, and subcategories are set up under each core category. Establish core category attribute sets for various entities based on the core category definition; Establish the relationship between core categories and core category attribute sets to obtain the corresponding semantic relationships; By extracting and integrating basic academic knowledge, applied knowledge, and expert experience from expert collaboration platforms, professional knowledge is obtained. Using core categories and core category attributes as knowledge graph nodes, the semantic relationships between core categories and core category attribute sets are analyzed. Based on the acquired professional knowledge, connections between knowledge graph nodes are established to form a first-level knowledge network. On the basis of the first-level knowledge network, connections between sub-class structures are established to form a second-level knowledge network. A hierarchical design is adopted to construct general knowledge and regionally specialized knowledge separately, and automatic fusion is achieved through association rules to balance the generality and specificity of knowledge. Together with the first-level knowledge network and the second-level knowledge network, a knowledge graph is formed.

3. The intelligent advisory system for aquaculture according to claim 1, characterized in that, The multimodal input processing module consists of the following units: Set up a data acquisition unit, including user input on the user terminal and setting up acquisition devices for real-time data acquisition; An image processing unit is set up to process images from user input and real-time acquired data; A video analysis unit is set up to process video streams from user input and real-time data acquisition, dynamically analyze the group behavior patterns of farmed organisms, and maintain video analysis performance under insufficient lighting conditions through an adaptive brightness enhancement algorithm. An audio processing unit is set up to process audio data from user input and real-time data acquisition, extract acoustic features, and analyze the operating status of aquaculture equipment and the sounds of abnormal aquatic animal activities. A text processing unit is set up to combine optical character recognition and natural language processing technologies to perform digital processing and structured extraction on user input and real-time collected data, and to transform unstructured text into structured data through named entity recognition and relation extraction technologies. The multimodal fusion unit maps feature vectors from different modalities to a unified semantic space, enhancing the complementarity of information between modalities. At the same time, it uses a modality-specific confidence scoring mechanism to dynamically weight the contribution of each modality, outputting a unified semantic representation.

4. The intelligent advisory system for aquaculture according to claim 1, characterized in that, The context-aware reasoning engine includes: The data processed by the multimodal input processing module is transformed into a graph structure and aligned with the knowledge graph nodes; Feature extraction is performed on the graph structure, and the association patterns of graph structure features and knowledge graph nodes are analyzed. Causal chains are constructed for graph structure features and knowledge graph nodes, and weights are calculated. By combining the temporal and spatial information of the causal chain with the structural features of the graph, a comprehensive representation of the current state of the aquaculture environment is constructed. Through reverse reasoning and template filling, targeted solutions and operational suggestions are generated.

5. The intelligent aquaculture advisory system according to claim 1, characterized in that, The personalized suggestion generation module includes: The reinforcement learning algorithm is used to generate implementation suggestions based on user input, real-time data collection and real-time graphs. Each implementation suggestion contains three levels of implementation details: the first level is the setting of core parameters, the second level is the operation method description, and the third level is the expected effect and risk assessment. The reinforcement learning algorithm adopts a policy optimization framework based on deep Q-networks. It uses real-time monitoring data and regional characteristics of the knowledge graph to model a Markov decision process. With user input as the reward function, it selects the best case in the breeding history of the knowledge graph. By learning from the best case, it forms an optimal policy library for different situations. At the same time, it introduces the Monte Carlo tree search method to evaluate the long-term decision path and ensure the long-term effectiveness of the suggestions. The system updates the Q value in real time from user feedback and continuously optimizes the decision model.

6. The intelligent advisory system for aquaculture according to claim 1, characterized in that, The continuous learning and knowledge iteration module includes: The system processes user interaction records, expert knowledge supplements from the expert collaboration platform, knowledge graphs, and real-time collected data using a multi-stage pipeline architecture to form a knowledge base. The multi-stage pipeline architecture processes data cleaning and standardization using a multimodal input processing module, feature extraction and semantic annotation using a context-aware reasoning engine, and knowledge conflict detection, knowledge fusion and value assessment using an expert collaboration platform. Meanwhile, the knowledge base also has a local device-based hierarchical storage architecture to store data processed by the multi-stage pipeline architecture locally. The knowledge base also includes a central server that processes data from a multi-stage pipeline architecture and updates it in real time, supporting rollback functionality based on time series or specific event sequences.

7. The intelligent aquaculture advisory system according to claim 1, characterized in that, The predictive aquaculture management module includes: Spatial scanning of the aquaculture farm generates a three-dimensional terrain model. Data collected in real time by the multimodal input processing module is used to obtain water quality parameters, meteorological parameters, and biological characteristic parameters of the aquaculture farm. By utilizing a three-dimensional terrain model and water quality parameters, meteorological parameters, and biological characteristic parameters, a coupled model of hydrodynamics, water quality change, biological growth, and meteorological influence is constructed, namely a virtual aquaculture environment model, to achieve real-time mapping between the virtual and real environments; The virtual aquaculture environment model uses a multi-scale fusion method for prediction over time series data; The virtual aquaculture environment model is optimized through a multi-objective decision-making algorithm, taking the aquaculture objectives input by the user as the optimization objective and considering the constraints set by the user.

8. The intelligent advisory system for aquaculture according to claim 1, characterized in that, The adaptive optimization module includes: When comparing the consistency between user actions and system suggestions, the core aquaculture parameters output by the personalized suggestion generation module are automatically adjusted using a two-dimensional matrix of region and season. The adjustment range is determined based on statistical analysis of historical data. At the same time, the early warning threshold of the predictive aquaculture management module and the judgment threshold of the context-aware inference engine are automatically updated based on the latest weather forecast data. The key parameters of each optimization, the graph neural network parameters of the context-aware inference engine, the reinforcement learning model weights of the personalized suggestion generation module, the time series prediction model parameters of the predictive aquaculture management module, and the updated content of the knowledge base are all cached locally. The adaptive optimization module, the predictive aquaculture management module, and the continuous learning and knowledge iteration module all have local caching functions and automatically update data through a differentiated synchronization mechanism.

9. A smart advisory method for aquaculture, characterized in that, Includes the following steps: The multimodal input processing module collects user input and real-time acquired data, and performs analysis and digital processing. The context-aware reasoning engine extracts features from the processed data and aquaculture objectives, analyzes the correlation between the environment and aquaculture data, and draws initial reasoning results to provide preliminary basis for aquaculture decision-making. The expert collaboration platform incorporates user operations, updates the knowledge graph through continuous learning and knowledge iteration modules, assists in knowledge conflict detection, integration, and value assessment, and ensures the accuracy and timeliness of knowledge. The adaptive optimization module compares the consistency between user operations and system suggestions, optimizes the parameters of the context-aware inference engine model, and adjusts the inference engine according to regional and seasonal adaptability. The optimization and adjustment schemes are cached locally to ensure offline operation and improve system adaptability and stability. The predictive aquaculture management module constructs a virtual aquaculture environment model, predicts key future aquaculture indicators based on time series analysis, and automatically generates risk prevention and optimization management suggestions to help users address potential problems in advance. The personalized suggestion generation module generates implementation suggestions based on real-time collected data, knowledge graph data, and aquaculture goals, using reinforcement learning algorithms to meet users' personalized needs and improve aquaculture efficiency. The knowledge graph construction module provides the theoretical foundation for other modules, while the continuous learning and knowledge iteration module acquires expert experience to update the knowledge graph, collaboratively handles new knowledge points and knowledge conflicts, and ensures the continuous improvement and development of the knowledge graph.

Images