Data knowledge base management method and device for egg industry and medium

Abstract

The invention discloses a data knowledge base management method and device for the egg industry and a medium, and relates to the technical field of intelligent breeding informatization, and the method comprises the steps: inputting a multi-modal fusion data set into a single deep learning framework, carrying out the multi-scale feature coupling of each modal feature through a self-adaptive nonlinear interaction function, and carrying out the multi-scale feature coupling; the feature weight is dynamically adjusted in combination with an enhanced feedback mechanism, and a multi-dimensional health assessment index set is generated by utilizing end-to-end joint feature extraction and nonlinear reasoning; performing entity recognition and relation extraction on the multi-dimensional health assessment index set, constructing an egg industry knowledge graph in combination with the breeding environment and the individual feature information, and performing structure optimization and dynamic evolution to obtain a dynamic evolution knowledge graph; and performing rule reasoning and data analysis on the dynamic evolution knowledge graph to generate a health management decision. Fine modeling and high-quality unified expression of multi-source data are achieved through piecewise nonlinear feature transformation and weighted fusion, and the accuracy of health state analysis input is ensured.

Description

Technical Field

[0001] This invention relates to the field of intelligent aquaculture information technology, and in particular to a data knowledge base management method, equipment and medium for the egg industry. Background Art

[0002] With the large-scale and intensive development of modern livestock and poultry farming, egg-laying hen farming has gradually evolved towards intelligent and data-driven approaches. Current technologies utilize various types of sensors and monitoring equipment deployed within farms to acquire real-time multi-source data, including environmental parameters, feeding information, egg production data, and individual behavioral characteristics, which are then stored and analyzed using an information management platform. Some advanced management methods have begun to explore the use of the Internet of Things, big data, and artificial intelligence to process the collected farming data to assist in health monitoring, production forecasting, and farming decision-making.

[0003] In existing technologies, there is a lack of a unified processing framework for the synchronization and standardization of multi-source data, and insufficient capabilities in multimodal information fusion and deep modeling. This makes it difficult for farm management systems to generate complete and dynamically updated knowledge structures in the face of complex environmental changes and significant individual differences. Especially in the egg industry, data covers multiple aspects such as environment, feed, individual physiology, and egg production quality. Traditional rule-based or single-feature-based analysis methods are insufficient to support holistic health assessments and accurate decision-making. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a data knowledge base management method for the egg industry to solve the problems of insufficient multimodal data fusion and difficulty in dynamically updating the knowledge structure.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a data knowledge base management method for the egg industry, which includes real-time collection of multimodal data in egg-laying hen farms, data synchronization, cleaning, format unification and preliminary labeling operations, to form a multimodal data set;

[0008] The multimodal dataset is normalized, features are extracted, and fusion calculations are performed to generate a multimodal fusion dataset.

[0009] The multimodal fusion dataset is input into a single deep learning framework. Multi-scale feature coupling of each modality feature is performed through an adaptive nonlinear interaction function. The feature weights are dynamically adjusted by combining a reinforcement feedback mechanism. End-to-end joint feature extraction and nonlinear inference are used to generate a multi-dimensional health assessment index set.

[0010] Entity identification and relation extraction were performed on a multi-dimensional health assessment indicator set. Combined with breeding environment and individual characteristic information, an egg industry knowledge graph was constructed, and its structure was optimized and dynamically evolved to obtain a dynamically evolved knowledge graph.

[0011] We perform rule-based reasoning and data analysis on dynamically evolving knowledge graphs to generate health management decisions.

[0012] As a preferred embodiment of the data knowledge base management method for the egg industry described in this invention, the steps of real-time collection of multimodal data from egg-laying hen farms, data synchronization, cleaning, format standardization, and preliminary labeling to form a multimodal data set are as follows.

[0013] Multimodal data from egg-laying hen farms are collected in real time and processed using a weighted dynamic time warping algorithm to generate multimodal data sequences. Anomaly detection and outlier removal are then performed to output a cleaned dataset.

[0014] Deep feature encoding and format unification processing are performed on the cleaned dataset to generate multimodal feature data, and semi-supervised learning annotation is performed to form a multimodal dataset.

[0015] As a preferred embodiment of the data knowledge base management method for the egg industry described in this invention, the specific steps for normalizing, extracting features, and fusing the multimodal data set to generate a multimodal fused dataset are as follows:

[0016] The behavioral features, environmental parameter features, physiological index features, and image visual features of laying hens were extracted from the multimodal dataset and then subjected to nonlinear normalization to obtain normalized feature data.

[0017] A piecewise nonlinear feature transformation operation is performed on the normalized feature data to obtain enhanced feature data, and then weighted fusion calculation is performed to generate a multimodal fusion dataset.

[0018] As a preferred embodiment of the data knowledge base management method for the egg industry described in this invention, the method involves: inputting a multimodal fusion dataset into a single deep learning framework; coupling features of each modality at multiple scales through an adaptive nonlinear interaction function; dynamically adjusting feature weights using a reinforcement feedback mechanism; and generating a multidimensional health assessment index set through end-to-end joint feature extraction and nonlinear inference. The specific steps are as follows:

[0019] The multimodal dataset is input into a single deep learning framework for initial encoding and mapping to generate a standardized feature matrix.

[0020] An adaptive nonlinear interaction function is used to couple the modal features in the standardized feature matrix at multiple scales to form a preliminary fusion feature representation.

[0021] A reinforced feedback mechanism is used to dynamically adjust the weights of the initial fusion feature representation to generate a weighted fusion feature representation.

[0022] End-to-end joint feature extraction and nonlinear inference are performed on the weighted fusion feature representation to generate a multi-dimensional health assessment index set.

[0023] As a preferred embodiment of the data knowledge base management method for the egg industry described in this invention, the specific steps for constructing an egg industry knowledge graph by performing entity recognition and relation extraction on a multi-dimensional health assessment indicator set and combining it with breeding environment and individual characteristic information are as follows.

[0024] Entity recognition based on deep neural networks and conditional random fields is performed on a multi-dimensional health assessment indicator set to generate a set of entities related to the health of laying hens;

[0025] Tensor decomposition model is used to extract relations from the set of entities related to the health of laying hens, resulting in a relation scoring matrix. Threshold filtering and normalization are then performed to generate a set of entity relations.

[0026] Normalized feature data is vectorized and weighted to generate a fused feature vector.

[0027] The fused feature vector is weighted and fused with the entity relationship set to generate a fused feature representation;

[0028] A graph neural network is used to embed and update the fused feature representation and perform semantic reasoning to construct a knowledge graph of the egg industry.

[0029] As a preferred embodiment of the data knowledge base management method for the egg industry described in this invention, the steps for performing structural optimization and dynamic evolution to obtain a dynamically evolving knowledge graph are as follows:

[0030] Consistency scoring is performed on the temporal weights of each node and edge in the egg industry knowledge graph, and low-scoring entity nodes and edges are removed to obtain an optimized knowledge graph.

[0031] The optimized knowledge graph is subjected to topological structure analysis and redundant edge removal operations to obtain a structure-optimized knowledge graph.

[0032] By using a dynamic evolution update function, continuous temporal adjustments and node relationship reconstruction operations are performed on the structure-optimized knowledge graph to form a dynamic evolution knowledge graph.

[0033] As a preferred embodiment of the data knowledge base management method for the egg industry described in this invention, the specific steps for generating health management decisions by performing rule-based reasoning and data analysis on a dynamically evolving knowledge graph are as follows:

[0034] The pre-defined health management rule base is matched with the entity relationship nodes in the dynamic evolutionary knowledge graph to obtain a matching rule set.

[0035] Perform rule reasoning and multi-dimensional data fusion on the matching rule set to obtain a health risk score;

[0036] By combining health risk scores with normalized feature data for comprehensive analysis and weighted evaluation, health management decisions can be generated.

[0037] As a preferred embodiment of the data knowledge base management method for the egg industry described in this invention, the specific steps for generating health management decisions by combining health risk scores with normalized feature data for comprehensive analysis and weighted evaluation are as follows.

[0038] Perform nonlinear transformation and filtering smoothing operations on the normalized feature data to generate a smoothing mapping matrix;

[0039] The health risk score is input into a smoothing mapping matrix, and a weighted expression is generated through a weighted interactive fusion operation. Then, nonlinear activation is performed to generate a weighted activation expression.

[0040] A deep feedforward network is used to perform nonlinear mapping and classification prediction operations on weighted activation representations to generate health management decisions.

[0041] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the data knowledge base management method for the egg industry as described in the first aspect of the present invention.

[0042] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the data knowledge base management method for the egg industry as described in the first aspect of the present invention.

[0043] The beneficial effects of this invention are as follows: by using piecewise nonlinear feature transformation and weighted fusion to achieve fine modeling and high-quality unified expression of multi-source data, the accuracy of health status analysis input is ensured; and by using structure optimization and dynamically evolving knowledge graphs to achieve real-time updating and adaptive reasoning of health knowledge, the timeliness and intelligence of health management decisions are guaranteed, forming a high-quality closed loop from data collection to decision generation, which significantly improves the scientificity and reliability of health management in the egg industry. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A flowchart for data knowledge base management methods in the egg industry.

[0046] Figure 2 This is a flowchart of multimodal data acquisition and preprocessing.

[0047] Figure 3 This is a flowchart for multimodal feature fusion and health assessment.

[0048] Figure 4 Generate flowcharts for knowledge graph-driven decision-making. Detailed Implementation

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0050] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0051] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0052] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a data knowledge base management method for the egg industry, including the following steps:

[0053] S1. Collect multimodal data in the egg-laying hen farm in real time, and perform data synchronization, cleaning, format unification and preliminary labeling operations to form a multimodal data set.

[0054] S1.1 Real-time acquisition of multimodal data from egg-laying hen farms, time synchronization processing using weighted dynamic time warping algorithm to generate multimodal data sequences, and execution of anomaly detection and anomaly removal to output a cleaned dataset.

[0055] Furthermore, multimodal data from egg-laying hen farms are collected in real time, including behavioral monitoring image information, environmental sensor information, and physiological indicator monitoring information. The collected information is then preliminarily sorted according to the collection time to ensure that all types of collected information have a unified time sequence.

[0056] The sorted information is input into a weighted dynamic time warping algorithm. By dynamically adjusting the time alignment path between multi-dimensional time series, different weights are assigned according to the importance of each type of information, so that the behavior monitoring image information, environmental sensor information and physiological indicator monitoring information are strictly synchronized in the time dimension, and a multimodal data sequence is output.

[0057] Anomaly detection is performed on the output multimodal data sequence. Data segments with obvious mutations, missing values ​​or distortions are identified by statistical feature analysis and pattern deviation detection. Multimodal data segments identified as anomalous are removed to ensure that only the collected information with high continuity and stability is retained in the subsequent feature extraction stage, and finally a cleaned dataset is formed.

[0058] S1.2 Perform deep feature encoding and format unification processing on the cleaned dataset to generate multimodal feature data, and perform semi-supervised learning annotation to form a multimodal dataset.

[0059] Furthermore, after forming the purified dataset, the behavior monitoring image information, environmental sensor information, and physiological indicator monitoring information in the purified dataset are respectively input into the deep feature encoding method. Through a multi-layer feature extraction structure, the original collected information is transformed into encoded features with high-dimensional expressive power, so that the behavior monitoring image information, environmental sensor information, and physiological indicator monitoring information are fully represented in the vector space.

[0060] Information processed through deep feature encoding is formatted uniformly according to a unified data standard, ensuring consistency in dimensional definition, index order, and numerical precision of encoded features, resulting in multimodal feature data. After obtaining the multimodal feature data, a semi-supervised learning annotation method is used, inputting both manually annotated and unannotated samples into the annotation process. By analyzing the feature distribution and similarity of a small number of annotated samples, the system automatically generates identification information for unannotated samples, ultimately forming a multimodal dataset that provides consistent input for subsequent feature extraction and fusion calculations.

[0061] S2. Normalize, extract features, and perform fusion calculations on the multimodal data set to generate a multimodal fusion dataset.

[0062] S2.1 Extract laying hen behavioral features, environmental parameter features, physiological index features, and image visual features from the multimodal dataset, and perform nonlinear normalization processing to obtain normalized feature data.

[0063] Furthermore, after forming a multimodal dataset, the behavioral monitoring image information, environmental sensor information, and physiological indicator monitoring information in the multimodal dataset are subjected to feature extraction according to their corresponding collection sources. This yields behavioral features of laying hens that reflect changes in their status, environmental parameter features that reflect the dynamics of the breeding environment, physiological indicator features that reflect their health status, and visual image features that reflect their appearance.

[0064] The extracted hen behavioral features, environmental parameter features, physiological index features, and image visual features were processed using a nonlinear normalization method. By mapping different feature numerical ranges, the features were made comparable on a uniform scale, and the influence of dimensional differences on subsequent fusion calculations was eliminated. Finally, normalized feature data was obtained, providing a consistent input format for multimodal feature fusion.

[0065] S2.2 Perform piecewise nonlinear feature transformation on the normalized feature data to obtain enhanced feature data, and perform weighted fusion calculation to generate a multimodal fusion dataset.

[0066] Furthermore, after obtaining the normalized feature data, the normalized feature data is divided into segments according to the characteristics of laying hen behavior, environmental parameters, physiological indicators, and image visual features. Nonlinear feature transformation is performed on each feature segment. The normalized feature data is enhanced by existing nonlinear mapping methods, which amplifies the variation information of laying hen behavior, environmental parameters, physiological indicators, and image visual features in local intervals, while suppressing insignificant fluctuations and noise interference, thus forming enhanced feature data.

[0067] After obtaining the enhanced feature data, weighting coefficients are assigned to the overall health assessment based on the importance of laying hen behavioral characteristics, environmental parameter characteristics, physiological indicator characteristics, and image visual characteristics. The enhanced feature data is then weighted and fused, and different types of features are combined in a unified vector space to output a multimodal fusion dataset.

[0068] S3. Input the multimodal fusion dataset into a single deep learning framework, perform multi-scale feature coupling on each modality feature through an adaptive nonlinear interaction function, and dynamically adjust the feature weights by combining a reinforcement feedback mechanism. Utilize end-to-end joint feature extraction and nonlinear inference to generate a multi-dimensional health assessment index set.

[0069] S3.1 Input the multimodal dataset into a single deep learning framework, perform preliminary encoding and mapping, and generate a standardized feature matrix.

[0070] Furthermore, after obtaining the multimodal dataset, behavioral monitoring information, environmental sensor information, physiological indicator monitoring information, and image visual information are used as inputs and fed into a single deep learning framework for preliminary encoding and mapping. Through multi-layer feature extraction and activation function processing, the information of each modality is transformed into a representation of a unified dimension. Normalization and standardization operations are then performed to generate a standardized feature matrix.

[0071] S3.2. Use an adaptive nonlinear interaction function to couple the modal features in the standardized feature matrix at multiple scales to form a preliminary fusion feature representation.

[0072] Furthermore, after obtaining the standardized feature matrix, behavioral monitoring information features, environmental sensor information features, physiological indicator monitoring information features, and image visual information features are input into an adaptive nonlinear interaction function. The modal features are coupled and nonlinearly combined at multiple scales. Through element-wise weighting, mapping, and interaction calculation, the interaction relationship and local feature changes between the modal features are strengthened, forming a preliminary fusion feature expression. This preliminary fusion feature expression can simultaneously reflect the key patterns of laying hen behavior, environment, physiology, and visual information, as well as the correlation between multimodal features.

[0073] S3.3 Utilize the reinforcement feedback mechanism to dynamically adjust the weights of the preliminary fusion feature representations and generate weighted fusion feature representations.

[0074] Furthermore, after obtaining the preliminary fusion feature expression, the preliminary fusion feature expression is input into the reinforcement feedback mechanism. Based on the importance and interaction contribution of multimodal features in the preliminary fusion feature expression, dynamic weights are assigned to each modal feature. Through element-by-element adjustment and weighted fusion operation, the influence of each modal feature in the fusion process is updated in real time to generate a weighted fusion feature expression.

[0075] S3.4 Perform end-to-end joint feature extraction and nonlinear reasoning on the weighted fusion feature representation to generate a multi-dimensional health assessment index set.

[0076] Furthermore, after obtaining the weighted fusion feature expression, the weighted fusion feature expression is input into the end-to-end joint feature extraction and nonlinear reasoning method. Layer-by-layer feature extraction and nonlinear mapping operations are performed on the weighted fusion feature expression to capture feature patterns in laying hen behavioral features, environmental parameter features, physiological indicator features and image visual features. In the feature extraction process, multi-dimensional information is combined to perform interactive reasoning to generate a multi-dimensional health assessment indicator set.

[0077] S4. Entity identification and relation extraction are performed on the multi-dimensional health assessment indicator set. Combined with breeding environment and individual characteristic information, an egg industry knowledge graph is constructed, and its structure is optimized and dynamically evolved to obtain a dynamically evolved knowledge graph.

[0078] S4.1 Perform entity recognition based on deep neural networks and conditional random fields on the multi-dimensional health assessment indicator set to generate a set of entities related to the health of laying hens.

[0079] Furthermore, after obtaining the multi-dimensional health assessment indicator set, the multi-dimensional health assessment indicator set is input into a deep neural network for feature encoding. The core information features that can effectively represent the health status of laying hens are extracted from the multi-dimensional health assessment indicator set, including behavioral features, environmental parameter features, and physiological indicator feature combination patterns. Then, the output of the deep neural network is jointly processed with a conditional random field. The conditional random field captures the temporal and structural dependencies between the multi-dimensional health assessment indicators, realizes the accurate identification of the boundaries of health status entities, and finally generates a set of entities related to the health of laying hens.

[0080] It should also be noted that the training process of a deep neural network first involves preprocessing the input data, such as normalization or standardization; initializing the network's weight parameters, typically using random initialization; performing forward propagation calculations, where the input data passes through each layer of the network to obtain the final prediction result; calculating the difference between the predicted result and the true label using a loss function; and using the backpropagation algorithm to calculate the gradient and update the network's weight parameters; optimization algorithms, such as gradient descent, are used to adjust the weights and minimize the loss function. During training, the model performance is periodically evaluated using a validation set, and hyperparameters are adjusted to prevent overfitting until the network converges. Finally, a test set is used for evaluation to ensure that the deep neural network has good generalization ability.

[0081] S4.2. Use the tensor decomposition model to extract relations from the set of entities related to the health of laying hens, obtain the relation scoring matrix, and perform threshold filtering and normalization to generate a set of entity relations.

[0082] Furthermore, after obtaining the set of entities related to the health of laying hens, the set of entities related to the health of laying hens is input into the tensor decomposition model. By constructing the set of entities related to the health of laying hens into a high-order tensor and performing tensor decomposition, the relationship strength between each pair of entities after the factor matrix is ​​reconstructed is calculated, thereby generating a relationship scoring matrix.

[0083] Each element in the relationship scoring matrix is ​​filtered according to the relationship scoring threshold, and low-scoring relationships are removed to retain key entity relationships. At the same time, the filtered scores are normalized so that the scores of different relationship dimensions can be compared and integrated under a unified scale, and finally an entity relationship set is generated, which can accurately reflect the relationship strength and association pattern between entities related to the health of laying hens.

[0084] It should also be explained that the process of setting the relationship scoring threshold involves statistically analyzing the distribution characteristics of relationship scores between all entities based on the relationship scoring matrix generated by the tensor decomposition model, such as calculating the mean, standard deviation, or quantiles of the scores; combining the importance of different entity relationships in the set of entities related to the health of laying hens, relationships with scores below a specific distribution index are considered low-associations; this score is used as the relationship scoring threshold to filter elements in the relationship scoring matrix, retaining high-scoring relationships and removing low-scoring relationships, thereby forming the final set of entity relationships.

[0085] It should also be noted that the training process of the tensor decomposition model first involves preprocessing the input data, including data normalization and missing value imputation, to ensure data quality and integrity. The factor matrices in the tensor decomposition model are initialized, typically using random initialization. An objective function is defined, usually by minimizing the reconstruction error, which measures the difference between the original tensor and the decomposed tensor. The factor matrices are updated using gradient descent or other optimization algorithms until the error converges or a predetermined number of iterations is reached. During training, it may be necessary to periodically use a validation set to tune hyperparameters and prevent overfitting. The performance of the tensor decomposition model is evaluated using a test set to ensure that the model can effectively predict or reconstruct the input data.

[0086] S4.3 Perform vectorization encoding and feature weighted fusion processing on the normalized feature data to generate a fused feature vector.

[0087] Furthermore, after obtaining the normalized feature data, the normalized feature data is vectorized and encoded, mapping the behavioral features of laying hens, environmental parameter features, physiological indicator features, and image visual features into a unified dimension feature vector representation. The various feature vectors are then weighted and fused according to preset weights to integrate the key feature patterns of behavioral, environmental, physiological, and visual information, generating a fused feature vector.

[0088] It should also be noted that the preset weights can be automatically derived through data analysis methods. By calculating the correlation between different features and the target variable, the influence of each feature on the health score is assessed. Based on these assessment results, features with stronger correlations are assigned higher initial weights.

[0089] S4.4. The fused feature vector and the entity relationship set are weighted and fused to generate a fused feature representation.

[0090] Furthermore, after obtaining the fused feature vector and the entity relationship set, the fused feature vector and the entity relationship set are weighted and fused element by element according to the preset weighting coefficient. Through the weighting operation, behavioral features, environmental parameter features, physiological indicator features, image visual features and entity relationship information are integrated into the same feature representation to generate a fused feature expression. This fused feature expression reflects the importance of multimodal features and the strength of the relationship between entities related to the health of laying hens.

[0091] Preset weighting coefficients can be obtained by statistical analysis of historical multimodal feature data and entity relationship data. For example, the correlation or information gain of various features on health management results can be calculated, and then the correlation or information gain can be normalized to generate a weight vector.

[0092] S4.5. Graph neural networks are used to embed and update the fused feature representations and perform semantic reasoning to construct a knowledge graph of the egg industry.

[0093] Furthermore, after obtaining the fused feature representation, the fused feature representation is input into a graph neural network for embedding and updating. Through multi-layer graph convolution operations, the node and edge information in the fused feature representation is propagated and aggregated to update the vector representation of each egg-laying hen health-related entity node. At the same time, semantic reasoning is performed in the graph neural network, and the relationship features between entities are interactively calculated with the features of egg-laying hen health-related entity nodes to generate semantic association weights. Based on the updated node vectors, each egg-laying hen health-related entity node is represented as a node attribute, and related nodes are connected according to the semantic association weights to form weighted edges, thereby constructing a complete egg industry knowledge graph that can simultaneously reflect the multimodal feature information of egg-laying hen health-related entities and the relationship patterns between entities.

[0094] It should also be noted that the fused feature representations and corresponding health-related entity relationship labels are used as training samples to construct the input graph structure of the graph neural network. In this graph, the nodes representing health-related entities of laying hens correspond to health-related entities, and the edges correspond to the relationships between entities. The number of layers, the number of neurons per layer, and the activation function type of the graph neural network are set. The entity node embeddings are calculated through forward propagation, and the neighbor information of entity nodes is weighted and aggregated at each layer. A supervised learning method is employed, utilizing the cross-entropy loss function or existing optimization algorithms, to iteratively update the network weights and biases through gradient descent, minimizing the error between the predicted and actual labels. This enables the graph neural network to accurately capture the structural and semantic associations of the fused feature representations and entity relationship information. During training, validation samples are evaluated and parameters are tuned until the entity node embeddings are stable and can reflect the relationship patterns and multimodal characteristics between health-related entities of laying hens.

[0095] S4.6. Perform consistency scoring on the temporal weights of each node and edge in the egg industry knowledge graph, and remove low-scoring entity nodes and edges to obtain an optimized knowledge graph.

[0096] Furthermore, after obtaining the egg industry knowledge graph, a consistency score is performed on the temporal weights of each entity node and edge in the knowledge graph. By analyzing the frequency of occurrence, weight change trends, and multimodal feature correlation of entity nodes and edges over continuous time periods, the consistency score value of each entity node and edge is calculated, expressed as:

[0097] S i,j =α·f(Y i,j )+β·g(X i,j )+γ·h(C i,j );

[0098] Among them, S i,j Let represent the consistency score of the edge between entity node i and entity node j, where i represents the starting entity node index of the edge, j represents the ending entity node index of the edge, α represents the frequency of occurrence weight coefficient, β represents the weight change trend weight coefficient, γ represents the multimodal feature correlation weight coefficient, and Y represents the frequency of occurrence weight coefficient. i,j X represents the frequency of the edge between entity node i and entity node j within a continuous time period. i,j C represents the trend of edge weight changes between entity node i and entity node j over a continuous time period. i,j The multimodal feature correlation between node i and node j is represented by f(·), and the frequency of opposite edges Y is represented by f(·). i,j The mapping function, g(·), represents the trend X of the edge weight change. i,j The mapping function, h(·), represents the correlation degree C of the multimodal features. i,j The function that performs the mapping;

[0099] Based on the consistency score threshold, entity nodes and edges with low scores are removed, thereby eliminating elements with unstable time sequence or weak correlation, generating an optimized knowledge graph that retains highly consistent and reliable entities and entity relationships related to the health of laying hens.

[0100] It should also be noted that the process of setting the consistency score threshold can be determined by statistical analysis of the historical consistency score distribution of entity nodes and edges in the optimized knowledge graph. This involves collecting the temporal consistency scores of all entity nodes and edges within a continuous time period to form a score set; calculating the mean, variance, and quantiles of the score set, for example, using the median or upper quartile as a reference value; and selecting an appropriate quantile as the consistency score threshold based on the proportion of highly reliable nodes or strongly related edges to be retained, such as the median of the score set plus a certain percentage of the variance.

[0101] S4.7 Perform topological analysis and redundant edge removal on the optimized knowledge graph to obtain a structure-optimized knowledge graph.

[0102] Furthermore, after obtaining the optimized knowledge graph, the importance of each node is evaluated by observing the number of edges directly connected to each entity node. The strength of the direct or indirect connection between nodes is determined by observing the length of the edge chain connecting any two entity nodes in the optimized knowledge graph and the number of intermediate nodes it passes through. The close or loose relationship patterns between nodes are also identified, thereby analyzing the closeness between nodes.

[0103] The overall connectivity of the graph is evaluated by examining the existence of effective paths or the size of connected segments between any two nodes, thereby identifying potentially redundant or weakly connected edges in the topology. At the same time, duplicate relationships or highly similar edge features between nodes are identified and eliminated to form a structure-optimized knowledge graph. This structure-optimized knowledge graph retains key entity nodes and highly correlated edges, ensuring the logical integrity and structural compactness of egg-laying hen health-related entities and relationships between entities in the graph.

[0104] S4.8. Utilize the dynamic evolution update function to perform continuous temporal adjustments and node relationship reconstruction operations on the structure optimization knowledge graph to form a dynamic evolution knowledge graph.

[0105] Furthermore, after obtaining the structure-optimized knowledge graph, a dynamic evolution update function is used to continuously adjust the attributes and weights of each node and edge in the structure-optimized knowledge graph. Based on the feature changes of nodes and edges in different time periods and the correlation of multimodal features, the relationships between nodes and edge weights are dynamically updated. At the same time, newly emerging entities related to the health of laying hens and potential relationships are identified, and the node relationships in the graph are reconstructed to form a dynamic evolution knowledge graph. This dynamic evolution knowledge graph can reflect the temporal evolution of entities related to the health of laying hens and the relationships between entities, while maintaining the continuity and dynamism of the graph structure, providing timeliness and reliability assurance for health management decisions.

[0106] S5. Perform rule-based reasoning and data analysis on the dynamic evolutionary knowledge graph to generate health management decisions.

[0107] S5.1 Match the preset health management rule base with the entity relationship nodes in the dynamic evolution knowledge graph to obtain the matching rule set.

[0108] Furthermore, after obtaining the dynamic evolutionary knowledge graph, each rule in the preset health management rule base is compared one by one with the entity relationship nodes in the dynamic evolutionary knowledge graph. By checking the consistency between the rule conditions and the attributes, states, and multimodal features of the entity relationship nodes, entity relationship combinations that satisfy the rules in the preset health management rule base are identified. These matching rules are then matched with the corresponding entity relationship nodes to generate a matching rule set. This ensures that the matching rule set can reflect the relationship patterns of egg-laying hen health-related entities in the dynamic evolutionary knowledge graph, providing a foundation for subsequent rule reasoning and health management decisions.

[0109] It should also be noted that the pre-set health management rule base is established by analyzing a multi-dimensional health assessment indicator set, normalized feature data, and entity relationship nodes in a dynamic evolutionary knowledge graph. First, health status judgment conditions are formulated based on the behavioral characteristics, environmental parameter characteristics, and physiological indicator characteristics of laying hens. For example, trigger conditions include abnormal laying hen activity, abnormal environmental temperature and humidity, or deviation of physiological indicators from example values. These judgment conditions are converted into rule expressions, clarifying the rule triggering logic, priority, and scope of application. Combined with the multimodal feature correlation and temporal changes of entity relationship nodes in the dynamic evolutionary knowledge graph, a rule triggering relationship mapping is constructed to form a pre-set health management rule base that can be matched with the dynamic evolutionary knowledge graph. This pre-set health management rule base can accurately reflect the correlation between the health status of laying hens and environmental conditions, providing an operational basis for health management decisions.

[0110] S5.2 Perform rule reasoning and multi-dimensional data fusion on the matching rule set to obtain a health risk score.

[0111] Furthermore, rule reasoning and multi-dimensional data fusion are performed on the matching rule set. By analyzing the rule conditions and logical relationships in the matching rule set one by one, and comparing the consistency between the rule conditions and the attributes, states, and multimodal features of the entity relationship nodes in the dynamic evolutionary knowledge graph, the matching degree is quantified into a rule trigger score. The triggering degree of the rule trigger score in the current dynamic evolutionary knowledge graph entity relationship nodes is calculated, and weighted fusion processing is performed by combining normalized feature data and multimodal information of fused feature vectors to form the risk contribution value of each laying hen in each health dimension. The risk contribution value of each laying hen in each health dimension is weighted and summed according to preset weights, taking into account the weights and interactions of different behavioral characteristics, environmental parameter characteristics, and physiological indicator characteristics, to generate a health risk score, so that the health risk score can quantify the overall health risk level of laying hens in the current breeding environment.

[0112] S5.3 Perform nonlinear transformation and filtering smoothing operations on the normalized feature data to generate a smoothing mapping matrix.

[0113] Furthermore, nonlinear transformation and filtering smoothing operations are performed on the normalized feature data. By mapping the behavioral features, environmental parameter features, physiological index features, and image visual features of laying hens in the temporal and spatial dimensions using nonlinear functions, the distribution differences between features are enhanced. Then, filtering methods are used to smooth the mapping results to suppress abnormal fluctuations and noise interference, ultimately forming a smooth mapping matrix. The smooth mapping matrix can maintain the continuity and correlation of multimodal features.

[0114] S5.4 Input the health risk score into the smoothing mapping matrix, generate a fused weighted expression through weighted interactive fusion operation, and perform nonlinear activation to generate a weighted activation expression.

[0115] Furthermore, the health risk score is input into the smoothing mapping matrix. A weighted, interactive fusion calculation is performed on the health risk score and each element in the smoothing mapping matrix to generate a fused weighted expression, which is:

[0116] F = σ(W⊙R);

[0117] Where F represents the fusion weighted expression, σ(·) represents the nonlinear activation function, W represents the smoothing mapping matrix, R represents the health risk score, and ⊙ represents the element-wise multiplication operator;

[0118] The influence weight of health risk score on the smooth mapping of multimodal features is considered in the fusion process, and a nonlinear activation function is applied to the fusion result to enhance the response of important features, finally obtaining a weighted activation expression, which can reflect the comprehensive influence of multimodal features and health risk score.

[0119] S5.5. Utilize a deep feedforward network to perform nonlinear mapping and classification prediction operations on the weighted activation expression to generate health management decisions.

[0120] Furthermore, the weighted activation representation is input into a deep feedforward network, where multi-layer nonlinear mapping is performed. Key features are extracted through layer-by-layer weighting and activation operations. In the deep feedforward network, the output of each layer is processed by a nonlinear activation function to form a new feature representation. Subsequently, the feature vector of the final layer is input into the classification prediction mechanism. By matching the feature vector with predefined health category labels, the prediction probability is calculated to determine the affiliation of each laying hen under different health states, thereby generating specific health management decisions. The deep feedforward network enhances the separability of important features in the weighted activation representation through nonlinear mapping and integrates multi-dimensional information during the classification prediction process to achieve the generation of health management decisions.

[0121] It should also be noted that the deep feedforward network is trained with weighted activation expressions and corresponding health management labels. The network weights and biases are adjusted through the backpropagation algorithm to optimize the classification prediction accuracy. For example, cross-entropy loss function and stochastic gradient descent method are used for iterative training until convergence to ensure that the deep feedforward network can effectively perform nonlinear mapping and health management judgment of the weighted activation expressions.

[0122] This embodiment also provides a computer device applicable to the data knowledge base management method for the egg industry, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the data knowledge base management method for the egg industry as proposed in the above embodiment.

[0123] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0124] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the data knowledge base management method for the egg industry as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0125] In summary, this invention achieves refined modeling and high-quality unified expression of multi-source data through piecewise nonlinear feature transformation and weighted fusion, ensuring the accuracy of health status analysis input; and achieves real-time updating and adaptive reasoning of health knowledge through structural optimization and dynamic evolution of knowledge graphs, ensuring the timeliness and intelligence of health management decisions. The whole process forms a high-quality closed loop from data collection to decision generation, significantly improving the scientificity and reliability of health management in the egg industry.

[0126] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. 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 be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data knowledge base management method for the egg industry, characterized in that: include, Real-time collection of multimodal data from egg-laying hen farms, followed by data synchronization, cleaning, format standardization, and preliminary labeling to form a multimodal data set; The multimodal dataset is normalized, features are extracted, and fusion calculations are performed to generate a multimodal fusion dataset. The multimodal fusion dataset is input into a single deep learning framework. Multi-scale feature coupling of each modality feature is performed through an adaptive nonlinear interaction function. The feature weights are dynamically adjusted by combining a reinforcement feedback mechanism. End-to-end joint feature extraction and nonlinear inference are used to generate a multi-dimensional health assessment index set. Entity identification and relation extraction were performed on a multi-dimensional health assessment indicator set. Combined with breeding environment and individual characteristic information, an egg industry knowledge graph was constructed, and its structure was optimized and dynamically evolved to obtain a dynamically evolved knowledge graph. We perform rule-based reasoning and data analysis on dynamically evolving knowledge graphs to generate health management decisions.

2. The data knowledge base management method for the egg industry as described in claim 1, characterized in that: The process involves real-time collection of multimodal data from egg-laying hen farms, followed by data synchronization, cleaning, format standardization, and preliminary labeling to form a multimodal dataset. The specific steps are as follows: Multimodal data from egg-laying hen farms are collected in real time, and time synchronization is performed using a weighted dynamic time warping algorithm to generate multimodal data sequences. Anomaly detection and anomaly removal are then performed to output a cleaned dataset. Deep feature encoding and format unification processing are performed on the cleaned dataset to generate multimodal feature data, and semi-supervised learning annotation is performed to form a multimodal dataset.

3. The data knowledge base management method for the egg industry as described in claim 2, characterized in that: The process of normalizing, extracting features, and fusing the multimodal dataset to generate a multimodal fusion dataset involves the following steps: The behavioral features, environmental parameter features, physiological index features, and image visual features of laying hens were extracted from the multimodal dataset and then subjected to nonlinear normalization to obtain normalized feature data. A piecewise nonlinear feature transformation operation is performed on the normalized feature data to obtain enhanced feature data, and then weighted fusion calculation is performed to generate a multimodal fusion dataset.

4. The data knowledge base management method for the egg industry as described in claim 3, characterized in that: The process involves inputting a multimodal fusion dataset into a single deep learning framework, coupling features of each modality at multiple scales using an adaptive nonlinear interaction function, dynamically adjusting feature weights using a reinforcement feedback mechanism, and generating a multidimensional health assessment index set through end-to-end joint feature extraction and nonlinear inference. The specific steps are as follows: The multimodal dataset is input into a single deep learning framework for initial encoding and mapping to generate a standardized feature matrix. An adaptive nonlinear interaction function is used to couple the modal features in the standardized feature matrix at multiple scales to form a preliminary fusion feature representation. A reinforced feedback mechanism is used to dynamically adjust the weights of the initial fusion feature representation to generate a weighted fusion feature representation. End-to-end joint feature extraction and nonlinear inference are performed on the weighted fusion feature representation to generate a multi-dimensional health assessment index set.

5. The data knowledge base management method for the egg industry as described in claim 4, characterized in that: The steps for constructing an egg industry knowledge graph by performing entity recognition and relation extraction on a multi-dimensional health assessment indicator set, combined with information on the breeding environment and individual characteristics, are as follows: Entity recognition based on deep neural networks and conditional random fields is performed on a multi-dimensional health assessment indicator set to generate a set of entities related to the health of laying hens; Tensor decomposition model is used to extract relations from the set of entities related to the health of laying hens, resulting in a relation scoring matrix. Threshold filtering and normalization are then performed to generate a set of entity relations. Normalized feature data is vectorized and weighted to generate a fused feature vector. The fused feature vector is weighted and fused with the entity relationship set to generate a fused feature representation; A graph neural network is used to embed and update the fused feature representation and perform semantic reasoning to construct a knowledge graph of the egg industry.

6. The data knowledge base management method for the egg industry as described in claim 5, characterized in that: The process involves structural optimization and dynamic evolution to obtain a dynamically evolving knowledge graph. The specific steps are as follows. Consistency scoring is performed on the temporal weights of each node and edge in the egg industry knowledge graph, and low-scoring entity nodes and edges are removed to obtain an optimized knowledge graph. The optimized knowledge graph is subjected to topological structure analysis and redundant edge removal operations to obtain a structure-optimized knowledge graph. By using a dynamic evolution update function, continuous temporal adjustments and node relationship reconstruction operations are performed on the structure-optimized knowledge graph to form a dynamic evolution knowledge graph.

7. The data knowledge base management method for the egg industry as described in claim 6, characterized in that: The specific steps for performing rule-based reasoning and data analysis on the dynamically evolving knowledge graph to generate health management decisions are as follows: The preset health management rule base is matched with the entity relationship nodes in the dynamic evolution knowledge graph to obtain the matching rule set; Perform rule reasoning and multi-dimensional data fusion on the matching rule set to obtain a health risk score; By combining health risk scores with normalized feature data for comprehensive analysis and weighted evaluation, health management decisions can be generated.

8. The data knowledge base management method for the egg industry as described in claim 7, characterized in that: The process of combining health risk scores with normalized feature data for comprehensive analysis and weighted evaluation to generate health management decisions involves the following steps: Perform nonlinear transformation and filtering smoothing operations on the normalized feature data to generate a smoothing mapping matrix; The health risk score is input into a smoothing mapping matrix, and a weighted expression is generated through a weighted interactive fusion operation. Then, nonlinear activation is performed to generate a weighted activation expression. A deep feedforward network is used to perform nonlinear mapping and classification prediction operations on weighted activation representations to generate health management decisions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the data knowledge base management method for the egg industry as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the data knowledge base management method for the egg industry as described in any one of claims 1 to 8.

Images