Pig farm miscarriage attribution diagnosis method based on blue ear disease risk propagation knowledge graph

Abstract

The application discloses a pig farm abortion attribution diagnosis method based on a blue-ear disease risk propagation knowledge graph, relates to the technical field of artificial intelligence, and comprises the following steps: constructing a pig farm blue-ear disease attribution knowledge graph, predefining a risk propagation mode according to a blue-ear disease risk propagation mechanism, searching for a path conforming to the risk propagation mode through graph mode matching, and integrating the path into a risk subgraph; adopting a graph attention network fused with the blue-ear disease risk propagation knowledge to perform representation learning on the risk subgraph, combining context representation learning and a time difference attenuation mechanism to quantize the contribution degrees of the risk subgraphs to the abortion rate of the pig farm; based on attention score decomposition of the contribution proportions of the risk events in the risk subgraph, generating a quantitative attribution result, and outputting a diagnosis result including risk event identification, a risk propagation link and a quantitative attribution contribution degree. The application realizes risk attribution of the blue-ear disease of the pig farm and quantizes the contribution of specific risk points to the abortion rate of the pig farm.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method for attributing and diagnosing abortion in pig farms based on a knowledge graph of the risk transmission of porcine reproductive and respiratory syndrome (PRRS). Background Technology

[0002] In the pig farming industry, pig farms face numerous biosecurity challenges in production management, with porcine reproductive and respiratory syndrome (PRRS) being one of the major challenges. PRRS spreads, mutates, and proliferates through the introduction and integration of breeding pigs into the herd. This disease can cause abortions in sows, increase piglet mortality and culling rates, and reduce piglet performance, thereby increasing the costs for pig farming enterprises and farmers, and also posing a threat to food safety for consumers.

[0003] On the other hand, while current market offerings include risk warning and attribution diagnostic solutions for major swine diseases, applicable to most diseases such as African swine fever and porcine reproductive and respiratory syndrome (PRRS), these solutions are based on general biosecurity systems for swine farms and do not specifically address PRRS risk diagnosis. Due to PRRS's long transmission chain, numerous production stages involved, and the virus's susceptibility to mutation, and as a domesticable virus, it presents unique key risk points in its control. Therefore, there is an urgent need for a solution that integrates domain knowledge about PRRS from swine farms, constructs a PRRS knowledge graph, enables automatic attribution reasoning for PRRS, and establishes a quantitative model for PRRS to quantify local risks and risk points. Summary of the Invention

[0004] To address the aforementioned technical issues, this invention proposes a method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission. This method enables risk attribution for PRRS in pig farms and quantifies the contribution of specific risk points to the abortion rate in pig farms.

[0005] The first aspect of this invention provides a method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission, comprising the following steps:

[0006] Based on detection data, immunization data, weather data, herd entry data, domestication data, and biosecurity data, a knowledge graph of PRRS attribution in pig farms is constructed, and the entities and relationships in the knowledge graph are stored in a graph database.

[0007] Based on the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS), a risk transmission pattern is predefined. Paths that conform to the risk transmission pattern are searched through graph pattern matching, integrated into a risk subgraph, and an attribution link table is output.

[0008] A graph attention network that integrates knowledge of the risk transmission of porcine reproductive and respiratory syndrome (PRRS) is used to perform representation learning on the risk subgraphs. Combined with contextual representation learning and time difference decay mechanism, the contribution of each risk subgraph to the abortion rate of pig farms is quantified.

[0009] Based on the contribution ratio and the attention score decomposition of the graph attention network, the contribution ratio of each risk event in the risk subgraph is decomposed to generate a quantitative attribution result. The final output includes a diagnostic result including risk event identification, risk propagation link and quantitative attribution contribution.

[0010] This solution constructs a knowledge graph of PRRS (Porcine Reproductive and Respiratory Syndrome) etiology in swine farms based on detection data, immunization data, weather data, herd entry data, domestication data, and biosecurity data, including:

[0011] A production event dataset for the observation period was constructed based on detection data, immunization data, weather data, population entry data, domestication data, and biosafety data. The production event dataset was then cleaned and standardized, and event attributes were completed.

[0012] Entities and relationships are automatically extracted using predefined node and relationship types, and the preprocessed production event dataset is transformed into nodes and edges of a knowledge graph for the attribution of porcine reproductive and respiratory syndrome (PRRS) in pig farms and stored in a graph database.

[0013] Safety risk instances are generated using predefined risk rules and associated with the corresponding nodes in the swine farm porcine reproductive and respiratory syndrome (PRRS) attribution knowledge graph.

[0014] In this scheme, based on the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS), a risk transmission pattern is predefined. Paths conforming to the risk transmission pattern are searched through graph pattern matching, integrated into a risk subgraph, and an attribution link table is output, including:

[0015] Based on domain knowledge, the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS) is obtained. Based on the PRRS risk transmission mechanism, key risk links are extracted by combining veterinary pathology and pig farm management processes. A preset number of transmission patterns are predefined. Each type of transmission pattern includes a triggering event, a transmission path, and inhibition conditions.

[0016] The propagation pattern is converted into a computable graph search rule, which specifies the types of nodes that must be included in the propagation pattern and the legal relationships between nodes, and specifies the timing constraints according to the preset observation window and causal timing.

[0017] For each type of transmission pattern, the path that conforms to the predefined pattern is retrieved by graph query in the constructed knowledge graph of PRRS attribution in pig farms. Different risk paths triggered by the same event in the same pig farm are obtained, aggregated by risk event type, and the most complete transmission link is retained.

[0018] Using pig farms as the root node and abnormal abortion rates as leaf nodes, all associated nodes and edges are merged to form a risk subgraph. The risk subgraph is then converted into an attribution link table for structured output.

[0019] In this scheme, a graph attention network integrating knowledge of porcine reproductive and respiratory syndrome (PRRS) risk transmission is used to learn the representation of the risk subgraph, including:

[0020] A risk subgraph representation learning network based on graph attention is constructed. Node features are extracted from the risk subgraph, and edge structure is encoded according to the relation type. Different weights are assigned to different relation types, and the risk subgraph is transformed into an input form suitable for graph attention network processing.

[0021] Knowledge of PRRS transmission is integrated into the graph attention representation learning network. An independent copy of the concept node in each subgraph is created through a concept duplication mechanism. The initial embeddings of the independent copies are the same, but they participate in attention calculation independently.

[0022] The attention weights of event nodes and their directly associated concept nodes are calculated through a multi-hop attention mechanism to propagate local risks, aggregate multi-level concept nodes for global risk integration, introduce domain rules as bias terms for attention scores, guide the model to focus on critical paths, and obtain attention scores to update node feature representations.

[0023] In this scheme, contextual representation learning and time difference decay mechanisms are combined to quantify the contribution of each risk subgraph to the abortion rate of pig farms, including:

[0024] The context representation learning network module encodes the geographical location of the pig farm and the seasonality of the observation period into a context vector, which is then concatenated with the node features to generate a context embedding.

[0025] The feature representations of all nodes in the risk subgraph are aggregated by graph pooling to obtain the subgraph representation. The subgraph representation and context embedding are fused together, and the fused features are imported into the subgraph contribution estimation network module. A learnable attenuation coefficient is introduced through the time difference attenuation estimation network module to attenuate the feature representation of long-term events before a preset time.

[0026] The contribution of the risk subgraph to the abortion rate in pig farms can be obtained by understanding the linear and activation layers.

[0027] In this scheme, based on the contribution rate combined with the attention score decomposition of the graph attention network, the contribution ratio of each risk event in the risk subgraph is decomposed to generate a quantitative attribution result, including:

[0028] The contribution of the risk subgraph output by the risk subgraph representation learning network to the abortion rate of the pig farm is obtained. The attention scores of risk event nodes in the risk subgraph representation learning network are backtracked. The attention scores of all risk event nodes in the same risk subgraph are summed. The original score of each risk event node is normalized to obtain the normalized attention score.

[0029] The normalized attention score is used to proportionally allocate the contribution of the risk subgraph to the abortion rate of the pig farm to each risk event node, and the contribution ratio of each risk event to the abortion rate in the risk subgraph is obtained, expressed as:

[0030] ,

[0031] in Indicates the first The first risk subgraph The contribution of each risk event to the miscarriage rate. Indicates the first Within the risk subgraph, the first Attention score for each risk event node This represents the sum of attention scores. Indicates the first The contribution of each risk subplot to the abortion rate in pig farms;

[0032] Based on the contribution percentage of each risk event to the miscarriage rate in the risk subplot, quantitative attribution results are generated.

[0033] The final output of this solution includes diagnostic results that identify risk events, assess risk propagation pathways, and quantify attribution contributions, including:

[0034] Obtain the event type and key attributes of the risk event as the risk event identification result, and obtain the complete causal chain from the source event to the miscarriage result as the risk propagation link;

[0035] The contribution of risk subplots to the abortion rate of pig farms and the contribution of risk events to the abortion rate of pig farms are obtained as quantitative attribution contribution. Risk event identification, risk propagation links and quantitative attribution contribution are integrated as diagnostic results.

[0036] Based on the quantitative attribution contribution, early warnings are triggered according to the early warning grading standard, and early warning information is generated and pushed out. Changes in the abortion rate of pig farms are continuously monitored.

[0037] The second aspect of this invention provides a pig farm abortion attribution diagnosis system based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission. The system includes: an automated graph construction module, a risk attribution reasoning module, and a risk quantification module.

[0038] The automated graph construction module constructs a dataset of production events within a preset observation period based on immune data, weather data, herd entry data, domestication data, and biosafety data. According to the predefined risk event criteria, it automatically identifies and extracts risk events from the structured pig farm event data, converts the extracted entities and relationships into knowledge graph nodes and edges, and constructs a knowledge graph of PRRS attribution in pig farms, which is then stored in a graph database.

[0039] The risk attribution reasoning module is based on the transmission mechanism of porcine reproductive and respiratory syndrome (PRRS). It predefines risk transmission patterns and uses graph pattern matching to search for paths that match the preset risk transmission patterns in the PRRS attribution knowledge graph of pig farms. It integrates multiple risk paths, constructs a risk subgraph, and forms a complete PRRS attribution link table for pig farms.

[0040] The risk quantification module performs representation learning on the risk subgraphs, and combines context representation learning and time difference decay mechanism to quantify the contribution of each risk subgraph to the pig farm abortion rate. Based on the contribution, the module decomposes the contribution ratio of each risk event in the risk subgraphs using the attention score of the graph attention network, and generates quantitative attribution results.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0042] This invention enables risk attribution for porcine reproductive and respiratory syndrome (PRRS) in pig farms, quantifying the contribution of specific risk points to the abortion rate. It proposes an attribution knowledge graph for PRRS, whose ontology design and attribution reasoning both incorporate the risk propagation mechanism unique to PRRS. The attribution method is suitable for deep, targeted diagnosis of PRRS in pig farms. Based on PRRS knowledge attribution, and considering the characteristics of PRRS disease, a quantitative attribution model for PRRS is proposed, enabling the quantification of the attribution of local risks and risk points to the abortion rate in pig farms.

[0043] Strengthening porcine reproductive and respiratory syndrome (PRRS) monitoring in pig farms, implementing risk tracking and problem rectification, reducing abortion rates, and improving piglet quality, thereby lowering the cost per kilogram of market pigs; reducing the efficiency of traceability diagnosis from several weeks to minutes; and solving the previous difficulty in tracing PRRS. Based on automatic diagnosis rules, the system automatically triggers alarms for pig farms with high PRRS risk and automatically pushes attribution diagnosis results. Attached Figure Description

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

[0045] Figure 1A flowchart of a diagnostic method for attribution of abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission is shown.

[0046] Figure 2 A schematic diagram of the knowledge graph of PRRS attribution in swine farms is shown;

[0047] Figure 3 A flowchart illustrating the construction of a knowledge graph for the attribution of porcine reproductive and respiratory syndrome (PRRS) in pig farms is shown.

[0048] Figure 4 A flowchart of the production risk sub-diagram is shown;

[0049] Figure 5 The flowchart illustrates a risk subgraph representation learning network based on graph attention performing quantitative attribution.

[0050] Figure 6 A block diagram of a pig farm abortion attribution diagnostic system based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission is shown. Detailed Implementation

[0051] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0052] 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 therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0053] Figure 1 A flowchart is shown for a method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission.

[0054] like Figure 1 As shown, this embodiment provides a method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission, including:

[0055] S102, Based on detection data, immunization data, weather data, herd entry data, domestication data and biosecurity data, construct a knowledge graph of PRRS attribution in pig farms, and store the entities and relationships in the knowledge graph of PRRS attribution in pig farms to a graph database;

[0056] S104. Based on the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS), a risk transmission pattern is predefined. Paths that conform to the risk transmission pattern are searched through graph pattern matching, integrated into a risk subgraph, and an attribution link table is output.

[0057] S106, A graph attention network integrating knowledge of PRRS risk transmission is used to perform representation learning on the risk subgraphs. Combined with context representation learning and time difference decay mechanism, the contribution of each risk subgraph to the abortion rate of the pig farm is quantified.

[0058] S108, based on the contribution rate combined with the attention score decomposition of the graph attention network, the contribution ratio of each risk event in the risk subgraph is decomposed to generate a quantitative attribution result, and finally outputs a diagnostic result including risk event identification, risk propagation link and quantitative attribution contribution rate.

[0059] It should be noted that, based on the transmission pathology mechanism of porcine reproductive and respiratory syndrome (PRRS), a PRRS risk attribution knowledge graph for pig farm environments was constructed. The constructed PRRS attribution knowledge graph for pig farms is as follows: Figure 2 As shown, the PRRS risk attribution knowledge graph is used for systematic attribution analysis of PRRS risk in pig farms. The attribution graph elucidates the transmission mechanism of PRRS, analyzes the impact path and transmission effect of the disease in various production stages of pig farms, and integrates and synergistically applies the PRRS biosafety knowledge system with the pig farm production management knowledge system. The attribution links constructed by the graph strictly follow the PRRS disease incidence and transmission patterns. This graph focuses on clarifying the transmission mechanism of PRRS in key production stages such as introduction, herd entry, and mating. Furthermore, by associating the analysis of diverse heterogeneous event data from pig farms with the risk system structure built based on PRRS transmission paths, the PRRS risk attribution knowledge graph establishes a structured PRRS knowledge system, achieving the scientific organization and structured presentation of disease prevention and control knowledge.

[0060] like Figure 3 As shown, the steps for constructing the knowledge graph for the attribution of porcine reproductive and respiratory syndrome (PRRS) in pig farms are as follows: A production event dataset for the observation period is constructed based on detection data, immunization data, weather data, herd entry data, domestication data, and biosecurity data. The production event dataset is then cleaned and standardized, and event attributes are completed. Entities and relationships are automatically extracted using predefined node and relationship types. The preprocessed production event dataset is then transformed into nodes and edges of the PRRS attribution knowledge graph and stored in a graph database. Safety risk instances are generated using predefined risk rules and associated with the corresponding nodes in the PRRS attribution knowledge graph.

[0061] The PRRS attribution knowledge graph for pig farms defines 5 thematic categories and 12 node types. The thematic categories include: pig farm status indicators, external environment, production management, safety risks, and concepts. The pig farm status indicators category is further divided into pig farm status nodes and pig farm abortion status nodes. Pig farm status nodes describe the status information of the pig farm to be attributed, including static information such as the pig farm's affiliated organization, name, latitude and longitude, province, and city, as well as dynamic status information such as the attribution date and the corresponding number of pregnant sows, the number of breeding sows, and PRRS positive test results. Pig farm abortion status nodes aim to classify and assess the abortion rate and abortion status of the target pig farm, and determine whether the farm has PRRS infection and its severity. The risk status of the breeding sow population will be directly reflected in changes in the abortion rate index, thereby identifying whether there are abnormal or potential abnormal abortion rates. The production management category includes daily production activities in pig farms, such as mating, introduction of new stock, herd transfer, farrowing, and immunization procedures, all of which can potentially become points of entry and spread for the PRRSV pathogen. The safety risk category includes risk events and safety events. Risk events are non-compliant events or activities in the pig farm's production process that increase the risk of PRRSV, such as quarantine upon entry into the herd. Safety events are activities that suppress the risk of PRRSV, preventing its spread. For example, a PRRSV immunization event indirectly reduces the risk of disease in sows by enhancing the herd's immunity. The external environment category includes weather conditions such as temperature and wind speed, which significantly affect the spread or suppression of the virus. For example, farrowing house management events and weather events indirectly lead to risks in sows through the risk of virus spread. A summary of the specific topics and types of PRRSV attribution knowledge graph nodes in pig farms is shown in Table 1 below:

[0062] Table 1: Overview of Node Topics and Types in the Knowledge Graph of Porcine Reproductive Technology (PRRT) Attribution in Pig Farms

[0063]

[0064] The PRRS attribution knowledge graph for pig farms defines the relationship types of graph nodes, including occurrence, existence, isA, by, experience, lead to, effects, inhibits, exhibit, and harbor. For example, by associating pig farm ID and date, edges are constructed between the pig farm and other production events such as introduction, herd entry, mating, and immunization, with the relationship type being occurrence, indicating that a certain event occurred at the pig farm; based on the specific circumstances of the production event (such as exceeding the age limit for introduction or a positive test result), one or more specific risk instances will be triggered, and edges are constructed between the pig farm, introduction, herd entry, mating, and immunization events and the corresponding risk events, with the relationship type being exist, indicating that a specific risk event exists among these events; based on the causal link layer of the knowledge graph and the pig farm ID and time, edges are constructed between production events and concepts, with the relationship type being by or lead to; clearly defining which abstract, defined risk type each specific risk instance embodies. This helps in the classification, statistics, and understanding of risks. By associating risk instances with attributes such as their stages and event descriptions, an edge is constructed between risk instance nodes and concept nodes in the risk knowledge system, with the relationship type being isA.

[0065] For daily-level data of pig farms within [TN, T-1] days, events are automatically extracted, where T is the running date and N is the observation time interval. Risk events at the pig farm-batch-type granularity and pig farm-type granularity are identified. Based on the knowledge graph and predefined risk propagation rules, the disease risk attribution results of the pig farm are inferred. Based on the constructed pig farm PRRS attribution knowledge graph and the risk propagation rules, the graph path search is completed. Pig farm events, event attributes, and inference links are integrated, and a risk link subgraph is output to realize the attribution task of pig farm diseases.

[0066] Based on domain knowledge, the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS) is obtained. Based on this mechanism, combined with veterinary pathology and pig farm management processes (introduction of breeding stock → isolation → integration into the herd → mating), key risk factors are extracted, such as virus introduction through introduction of breeding stock → insufficient isolation → infection of breeding sows upon integration into the herd → increased abortion rate. Seven transmission modes are predefined, including:

[0067] Propagation Mode 1: Introduction of breeding stock is risky, and entry into the herd is risky. The triplet expression of the risk propagation chain under this case is: (pig farm, occurs, introduction of breeding stock event), (introduction of breeding stock event, exists, introduction of breeding stock risk event), (introduction of breeding stock risk event, leads to risk of replacement gilts), (replacement gilt risk, through, entry into the herd event), (entry into the herd event, exists, entry into the herd risk event), (entry into the herd risk event, leads to risk of sows), (sows risk, affects, abortion event), (abortion event, manifested as, abnormal abortion rate), (pig farm, suffers, abnormal abortion rate).

[0068] Propagation Mode 2: Introduction of breeding stock is risky, but entry into the herd is not. The triplet expression of the risk propagation chain in this case is: (pig farm, occurs, introduction of breeding stock event), (introduction of breeding stock event, exists, introduction of breeding stock risk event), (introduction of breeding stock risk event, leads to risk of replacement gilts), (replacement gilt risk, through, entry into the herd event), (entry into the herd event, is, entry into the herd), (entry into the herd, leads to risk of basal sows), (basal sow risk, affects, abortion rate event), (abortion rate event, manifests as, abortion rate abnormality), (pig farm, suffers, abortion rate abnormality).

[0069] Propagation Mode 3: No risk in introducing new breeds, but risk in herd entry. The triplet expression of the risk propagation chain in this case is: (pig farm, occurrence, herd entry event), (herd entry event, existence, herd entry risk event), (herd entry risk event, leading to risk in basal sows), (basal sow risk, impact, abortion rate event), (abortion rate event, manifested as, abnormal abortion rate), (pig farm, suffers, abnormal abortion rate).

[0070] Transmission Mode 4: Mating is risky. The triplet representation of the risk transmission chain in this case is: (pig farm, occurs, mating event), (mating event, exists, mating risk event), (mating risk event, leads to risk in sows), (risk in sows, affects abortion rate event), (abortion rate event, manifests as abnormal abortion rate), (pig farm, suffers, abnormal abortion rate).

[0071] Transmission Mode 5: Immunization is risky. The triplet expression of the risk transmission chain in this case is: (pig farm, occurs, immunization event), (immunization event, exists, immunization risk event), (immunization risk event, leads to, decreased immunity), (decreased immunity, leads to, risk in basal sows), (basal sow risk, affects, abortion rate event), (abortion rate event, manifests as, abnormal abortion rate), (pig farm, suffers, abnormal abortion rate).

[0072] Transmission Mode 6: There is a risk in farrowing house management. The triplet expression of the risk transmission chain in this case is: (pig farm, occurs, farrowing house management event), (farming house management event, exists, farrowing house management risk event), (farming house management risk event, leads to, virus spread risk), (virus spread risk, leads to, risk in sows), (sow risk, affects, abortion rate event), (abortion rate event, manifests as, abnormal abortion rate), (pig farm, suffers, abnormal abortion rate).

[0073] Transmission Mode 7: Weather is risky. The triplet expression of the risk transmission chain under this situation is: (pig farm, occurs, weather event), (weather event, exists, weather risk event), (weather risk event, leads to, virus spread risk), (virus spread risk, leads to, risk of sows) (sow risk, affects, abortion rate event), (abortion rate event, manifests as, abortion rate abnormality), (pig farm, suffers, abortion rate abnormality).

[0074] Each propagation pattern includes a triggering event (e.g., positive test result for introduced breed, expired immunization), a propagation path (e.g., introduced breed risk → replacement gilt risk → sow risk → abortion), and a suppression condition (e.g., adequate quarantine upon entry can block propagation). The propagation patterns are converted into computable graph search rules, specifying the node types that must be included in the propagation pattern (e.g., introduced breed event nodes, abortion rate abnormality concept nodes) and the legitimate relationships between nodes (e.g., introduced breed event → exist → risk event → leadto → replacement gilt risk), and temporal constraints are specified based on a preset observation window and causal time sequence. The risk subgraph generation process is as follows: Figure 4 As shown, for each type of transmission pattern, the path conforming to the predefined pattern is retrieved using graph query in the constructed knowledge graph of PRRS attribution in pig farms. Different risk paths triggered by the same event in the same pig farm are obtained, aggregated by risk event type, and the most complete transmission link is retained. With the pig farm as the root node and abnormal abortion rate as the leaf node, all related nodes and edges are merged to form a risk subgraph. The risk subgraph is then converted into an attribution link table for structured output.

[0075] According to an embodiment of the present invention, a graph attention network incorporating knowledge of porcine reproductive and respiratory syndrome (PRRS) risk transmission is used to learn the representation of the risk subgraph. For example... Figure 5As shown, a risk subgraph representation learning network based on graph attention is constructed, including a graph attention representation learning network module, a context representation learning network module, a subgraph contribution estimation network module, and a time difference decay estimation network module. Node features are extracted from the risk subgraph, and edge structure is encoded according to relation types, assigning different weights to different relation types. This transforms the risk subgraph into an input form suitable for graph attention network processing. The attribution path of each pig farm is treated as an independent subgraph, preserving its internal node and edge topology. Knowledge of PRRS transmission is integrated into the graph attention representation learning network to enhance the semantic representation of risk paths. Since shared concept nodes (such as basic sow risk) in traditional graph attention networks can lead to feature confusion in different subgraphs, a concept replication mechanism is used to create independent copies of concept nodes in each subgraph. These independent copies have the same initial embedding but participate independently in attention calculations. The concept replication operation is introduced to expand virtual concept nodes. Virtual concept nodes share a consistent input embedding representation, but perform different convolution operations in each subgraph, thus effectively preserving the independence of the subgraphs. The attention weights of event nodes and their directly related concept nodes are calculated through a multi-hop attention mechanism to propagate local risks. The weights are determined by the similarity of node features and edge type. Multi-level concept nodes are aggregated to integrate global risks and capture long-distance dependencies. Domain rules (such as insufficient group isolation contributing more to the abortion rate) are introduced as bias terms for attention scores to guide the model to focus on critical paths and obtain attention scores to update node feature representations.

[0076] The context representation learning network module encodes the geographical location of the pig farm and the season of the observation period into context vectors, which are then concatenated with the node features to generate a context embedding. It learns the additional impact of different geographical regions and seasonal factors on risk transmission, such as the significant inhibitory effect of high temperatures on porcine reproductive and respiratory syndrome (PRRS). The feature representations of all nodes in the risk subgraph are aggregated through graph pooling to obtain a subgraph representation. This subgraph representation and the context embedding are then fused, and the fused features are imported into the subgraph contribution estimation network module. A learnable attenuation coefficient is introduced through the time difference attenuation estimation network module to enhance the weight allocation of recent risk subgraphs and attenuate the feature representations of long-term events before a preset time. If the event occurrence time is... The current observation date is The attenuation factor is The contribution of the risk subgraph to the abortion rate in pig farms was obtained using the understanding of linear and activation layers. A graph attention network incorporating knowledge of PRRS risk transmission was trained based on MSE loss, with the loss formula as follows:

[0077] ,

[0078] Wherein represents the sample index corresponding to the attribution performance period. Indicates pig farm sample The A risk subgraph, Indicates pig farm sample The miscarriage rate corresponding to the performance period, Indicates pig farm sample The contribution of the risk subplot to the abortion rate in pig farms reflects the impact of the risk subplot on the sample. Quantitative attribution results for miscarriage rate.

[0079] It should be noted that obtaining the risk subgraph representation network's output contribution to the pig farm's abortion rate involves backtracking the attention scores of risk event nodes in the risk subgraph representation network, summing the attention scores of all risk event nodes within the same risk subgraph, and normalizing the original score of each risk event node to obtain a normalized attention score. This normalized attention score is then used to proportionally allocate the contribution of the risk subgraph to the pig farm's abortion rate to each risk event node, thus obtaining the contribution ratio of each risk event in the risk subgraph to the abortion rate, expressed as:

[0080] ,

[0081] in Indicates the first The first risk subgraph The contribution of each risk event to the miscarriage rate. Indicates the first Within the risk subgraph, the first Attention score for each risk event node This represents the sum of attention scores. Indicates the first The contribution of each risk subplot to the abortion rate in pig farms.

[0082] Based on the contribution percentage of each risk event to the miscarriage rate in the risk subplot, quantitative attribution results are generated, including risk event type, contribution percentage, key attributes, and improvement suggestions. High-risk events are checked to ensure they align with domain knowledge, and the quantitative attribution results are validated and calibrated by comparing them with the actual impact of similar events in historical data. These quantitative results directly guide the prioritization of prevention and control measures.

[0083] The preferred approach is to integrate risk events across subgraphs, grouping similar risk events from different risk subgraphs into a single group, and then calculating the contribution value of each group of events. Summation yields the global contribution of this type of risk event, merging the contributions of the same type of risk event from multiple subgraphs within the same pig farm. Additionally, counterfactual contribution analysis assesses the potential impact of improving specific risk events on abortion rates, assuming the elimination of a certain event (e.g., "insufficient quarantine time"), and its contribution value. This will be deducted from the total miscarriage rate. For example, if the current miscarriage rate is 30% and a certain event contributes 10%, the improved miscarriage rate is expected to drop to 20%.

[0084] It should be noted that the event type and key attributes of risk events are obtained as the risk event identification result. The event types include introduction risk, herd entry risk, immunization risk, environmental risk, etc., and the key attributes include time, batch, test results, compliance, etc. The complete causal chain from the source event to the abortion result is obtained as the risk propagation link, for example, positive wild-type virus in introduction → risk in replacement gilts → insufficient isolation upon entry into the herd → risk in foundation sows → abnormal abortion rate. The contribution of the risk subgraph to the abortion rate of the pig farm and the contribution value of the risk event to the abortion rate of the pig farm are obtained as the quantitative attribution contribution. The risk event identification, risk propagation link, and quantitative attribution contribution are integrated as the diagnostic result. The warning is triggered based on the quantitative attribution contribution according to the warning classification standard. The warning classification standard includes red warning: single event contribution ≥ 20% or risk subgraph contribution ≥ 40%; yellow warning: 10% ≤ single event contribution < 20% or risk subgraph contribution 20%~40%; blue warning: single event contribution or risk subgraph contribution < 10%. The system generates and pushes early warning information containing event summaries, obtains suggested measures based on domain knowledge, continuously monitors changes in pig farm abortion rates to track effects, and achieves closed-loop management of early warning.

[0085] In a preferred embodiment of the present invention, intelligent disease attribution is performed on the increased abortion rate caused by porcine reproductive and respiratory syndrome (PRRS) in pig farms, aiming to achieve the following tasks:

[0086] Task 1: Risk identification and risk attribution. On day T, collect multidimensional heterogeneous data on pig farm production activities (introduction of pigs, introduction of pigs into herds, mating), porcine reproductive and respiratory syndrome (PRRS) biosecurity activities (immunization, testing), and external conditions (such as weather) over the past 90 days.

[0087] (1) Identify the risk events that occurred in the pig farm within the previous 90 days;

[0088] (2) Give the attribution results of the abortion rate in the first 7 days of the pig farm for each risk event (risk propagation causation).

[0089] Example: On [May 1, 2025], complete data recorded on the production line of a pig farm for the previous 90 days [January 31, 2025 to April 30, 2025] are given. According to Task 1 (risk identification and risk attribution), identify all risk events that occurred in the pig farm during the observation period. Based on the direction of risk propagation, connect these risk events into a complete risk chain as the attribution chain for the abortion rate.

[0090] Task 2: Risk Attribution Quantification. On day T, quantify the attribution of risk events and risk propagation chains obtained from Task 1 (risk identification and risk attribution).

[0091] (1) Quantify the risk of the risk event on the link, that is, obtain the contribution of the event to the miscarriage rate / death rate;

[0092] (2) Support counterfactual reasoning, that is, improve the risk event and quantify the degree of optimization of the current risk indicators.

[0093] Example: On [2025-05-01], the attribution chain of the abortion rate / mortality rate of a certain pig farm for the previous 90 days [2025-01-31~2025-04-30] is given, and the contribution percentage of each event to the abortion rate is calculated. For example, the specific time of a certain introduction of a certain breed accounts for 19%, and the specific event of isolation accounts for 31%.

[0094] Figure 6 A block diagram of a pig farm abortion attribution diagnostic system based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission is shown.

[0095] The second embodiment of the present invention provides a pig farm abortion attribution diagnosis system based on a knowledge graph of PRRS risk transmission. The system includes: an automated graph construction module, a risk attribution reasoning module, and a risk quantification module.

[0096] The automated knowledge graph construction module constructs a dataset of production events within a preset observation period based on immunization data, weather data, herd entry data, domestication data, and biosecurity data. According to predefined risk event criteria, it automatically identifies and extracts risk events from structured pig farm event data, converting the extracted entities and relationships into knowledge graph nodes and edges. This constructs a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) attribution and stores it in a graph database. The module defines graph node types, including pig farm status indicators, external environment, production management, safety risks, and concepts. It also defines graph node relationship types, including occur, exist, isA, by, experience, lead to, affect, inhibit, exhibit, and harbor.

[0097] The risk attribution reasoning module is based on the transmission mechanism of porcine reproductive and respiratory syndrome (PRRS) and predefines risk transmission patterns, including: risky introduction of new breeds, risky entry into the herd, risky introduction of new breeds with no risk of entry, risky introduction of new breeds with no risk of entry, risky mating, risky immunization, risky farrowing house management, and risky weather. Graph pattern matching is used to search for paths matching the predefined risk transmission patterns in the PRRS attribution knowledge graph of the pig farm. Multiple risk paths are integrated to construct a risk subgraph, forming a complete PRRS attribution link table for the pig farm.

[0098] The risk quantification module performs representation learning on the risk subgraphs, and combines context representation learning and time difference decay mechanism to quantify the contribution of each risk subgraph to the pig farm abortion rate. Based on the contribution, the module decomposes the contribution ratio of each risk event in the risk subgraphs using the attention score of the graph attention network, and generates quantitative attribution results.

[0099] Preferably, the system also includes a visualization and decision support module that outputs diagnostic results including risk event identification, risk propagation links, and quantitative attribution contribution, provides an interactive graph visualization interface, displays risk propagation links and structured reports, and triggers early warnings based on the diagnostic results.

[0100] The third embodiment of this application also provides a computer-readable storage medium storing computer program code, which, when executed by the processor, causes the electronic device to perform the relevant method steps in the above method embodiments.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, and can be electrical, mechanical, or other forms. Furthermore, in the various embodiments of the present invention, all functional units can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0102] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0103] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission, characterized in that, Includes the following steps: Based on detection data, immunization data, weather data, herd entry data, domestication data, and biosecurity data, a knowledge graph of PRRS attribution in pig farms is constructed, and the entities and relationships in the knowledge graph are stored in a graph database. Based on the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS), a risk transmission pattern is predefined. Paths that conform to the risk transmission pattern are searched through graph pattern matching, integrated into a risk subgraph, and an attribution link table is output. A graph attention network that integrates knowledge of the risk transmission of porcine reproductive and respiratory syndrome (PRRS) is used to perform representation learning on the risk subgraphs. Combined with contextual representation learning and time difference decay mechanism, the contribution of each risk subgraph to the abortion rate of pig farms is quantified. Based on the contribution ratio and the attention score decomposition of the graph attention network, the contribution ratio of each risk event in the risk subgraph is decomposed, and a quantitative attribution result is generated. The final output includes the diagnostic results of risk event identification, risk propagation link and quantitative attribution contribution. Based on the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS), a predefined risk transmission pattern is used. Paths matching the risk transmission pattern are searched and integrated into a risk subgraph, and an attribution link table is output, including: Based on domain knowledge, the risk transmission mechanism of porcine reproductive and respiratory syndrome (PRRS) is obtained. Based on the PRRS risk transmission mechanism, key risk links are extracted by combining veterinary pathology and pig farm management processes. A preset number of transmission patterns are predefined. Each type of transmission pattern includes a triggering event, a transmission path, and inhibition conditions. The propagation pattern is converted into a computable graph search rule, which specifies the types of nodes that must be included in the propagation pattern and the legal relationships between nodes, and specifies the timing constraints according to the preset observation window and causal timing. For each type of transmission pattern, the path that conforms to the predefined pattern is retrieved by graph query in the constructed knowledge graph of PRRS attribution in pig farms. Different risk paths triggered by the same event in the same pig farm are obtained, aggregated by risk event type, and the most complete transmission link is retained. Using the pig farm as the root node and abnormal abortion rate as the leaf node, all associated nodes and edges are merged to form a risk subgraph. The risk subgraph is then converted into an attribution link table and output in a structured manner. A graph attention network incorporating knowledge of porcine reproductive and respiratory syndrome (PRRS) risk transmission is used to learn the representation of the risk subgraph, including: A risk subgraph representation learning network based on graph attention is constructed. Node features are extracted from the risk subgraph, and edge structure is encoded according to the relation type. Different weights are assigned to different relation types, and the risk subgraph is transformed into an input form suitable for graph attention network processing. Knowledge of PRRS transmission is integrated into the graph attention representation learning network. An independent copy of the concept node in each subgraph is created through a concept duplication mechanism. The initial embeddings of the independent copies are the same, but they participate in attention calculation independently. The attention weights of event nodes and their directly related concept nodes are calculated through a multi-hop attention mechanism to propagate local risks, aggregate multi-level concept nodes for global risk integration, introduce domain rules as bias terms for attention scores, guide the model to focus on critical paths, and obtain attention scores to update node feature representations. By combining contextual representation learning and time-difference decay mechanisms, the contribution of each risk subgraph to the abortion rate in pig farms is quantified, including: The context representation learning network module encodes the geographical location of the pig farm and the seasonality of the observation period into a context vector, which is then concatenated with the node features to generate a context embedding. The feature representations of all nodes in the risk subgraph are aggregated by graph pooling to obtain the subgraph representation. The subgraph representation and context embedding are fused together, and the fused features are imported into the subgraph contribution estimation network module. A learnable attenuation coefficient is introduced through the time difference attenuation estimation network module to attenuate the feature representation of long-term events before a preset time. The contribution of the risk subgraph to the abortion rate in pig farms can be obtained by understanding the linear and activation layers.

2. The method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission, as described in claim 1, is characterized in that... Based on detection data, immunization data, weather data, herd entry data, domestication data, and biosecurity data, a knowledge graph attributing porcine reproductive and respiratory syndrome (PRRS) to swine farms was constructed, including: A production event dataset for the observation period was constructed based on detection data, immunization data, weather data, population entry data, domestication data, and biosafety data. The production event dataset was then cleaned and standardized, and event attributes were completed. Entities and relationships are automatically extracted using predefined node and relationship types, and the preprocessed production event dataset is transformed into nodes and edges of a knowledge graph for the attribution of porcine reproductive and respiratory syndrome (PRRS) in pig farms and stored in a graph database. Safety risk instances are generated using predefined risk rules and associated with the corresponding nodes in the swine farm porcine reproductive and respiratory syndrome (PRRS) attribution knowledge graph.

3. The method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission, as described in claim 1, is characterized in that... Predefined propagation patterns of a preset number, including: Transmission Mode 1: Introduction of breeding stock carries risks, and herd entry carries risks. The risk transmission chain in this scenario is represented by the following triplet: Pig farm, occurrence of introduction of breeding stock event; introduction of breeding stock event, existence of introduction of breeding stock risk event; introduction of breeding stock risk event, leading to risk in replacement gilts; replacement gilt risk, through herd entry event; herd entry event, existence of herd entry risk event; herd entry risk event, leading to risk in breeding sows; breeding sow risk, affecting abortion events; abortion events, manifested as abnormal abortion rate; pig farm, suffering from abnormal abortion rate. Transmission Mode 2: Introduction of breeding stock carries risk, but entry into the herd is not. The risk propagation chain in this scenario is represented by the following triplets: Pig farm, occurrence of introduction of breeding stock event; introduction of breeding stock event, existence of introduction risk event; introduction risk event, leading to gilt risk; gilt risk, through entry into the herd event; entry into the herd event, is entry into the herd; entry into the herd, leading to basal sow risk; basal sow risk, affecting abortion rate event; abortion rate event, manifested as abnormal abortion rate; pig farm, suffering from abnormal abortion rate. Transmission Mode 3: No risk in introducing new breeds, but risk in introducing them into the herd. The risk transmission chain in this case can be represented by the following triplet: Pig farm, occurrence, herd introduction event; herd introduction event, existence, herd introduction risk event; herd introduction risk event, leading to risk in basal sows; basal sow risk, affecting abortion rate event; abortion rate event, manifested as abnormal abortion rate; pig farm, suffering from abnormal abortion rate. Transmission Mode 4: Mating is risky. The triplet representation of the risk transmission chain in this case is: pig farm, occurrence, mating event; mating event, existence, mating risk event; mating risk event, leading to risk in sows; risk in sows, affecting abortion rate event; abortion rate event, manifested as abnormal abortion rate; pig farm, suffering from abnormal abortion rate. Transmission Mode 5: Immunization is risky. The triplet expression of the risk transmission chain in this case is: pig farm, occurrence, immunization event; immunization event, existence, immunization risk event; immunization risk event, leading to decreased immunity; decreased immunity, leading to risk in basal sows; risk in basal sows, affecting abortion rate events; abortion rate events manifest as abnormal abortion rate; pig farm, suffering from abnormal abortion rate. Transmission Mode 6: Risk in farrowing house management. The risk transmission chain in this case is represented by the following triplet: Pig farm, occurrence, farrowing house management event; farrowing house management event, existence, farrowing house management risk event; farrowing house management risk event, leading to virus spread risk; virus spread risk, leading to risk in sows; sow risk, affecting abortion rate event; abortion rate event, manifested as abnormal abortion rate; pig farm, suffering from abnormal abortion rate. Transmission Mode 7: Weather is risky. The triplet representation of the risk transmission chain in this case is: Pig farm, occurs, weather event; weather event, exists, weather risk event; weather risk event, leads to virus spread risk; virus spread risk, leads to risk in breeding sows; risk in breeding sows, affects abortion rate event; abortion rate event, manifests as abnormal abortion rate; pig farm, suffers from abnormal abortion rate.

4. The method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission as described in claim 1, characterized in that, Based on the contribution ratio and the attention score decomposition of the graph attention network, the contribution proportion of each risk event in the risk subgraph is generated, including: The contribution of the risk subgraph output by the risk subgraph representation learning network to the abortion rate of the pig farm is obtained. The attention scores of risk event nodes in the risk subgraph representation learning network are backtracked. The attention scores of all risk event nodes in the same risk subgraph are summed. The original score of each risk event node is normalized to obtain the normalized attention score. The normalized attention score is used to proportionally allocate the contribution of the risk subgraph to the abortion rate of the pig farm to each risk event node, and the contribution ratio of each risk event to the abortion rate in the risk subgraph is obtained, expressed as: ， in Indicates the first The first risk subgraph The contribution of each risk event to the miscarriage rate. Indicates the first Within the risk subgraph, the first Attention score for each risk event node This represents the sum of attention scores. Indicates the first The contribution of each risk subplot to the abortion rate in pig farms; Based on the contribution percentage of each risk event to the miscarriage rate in the risk subplot, quantitative attribution results are generated.

5. The method for attributing and diagnosing abortion in pig farms based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission, as described in claim 1, is characterized in that... The final output includes diagnostic results for risk event identification, risk propagation pathways, and quantified attribution contributions, including: Obtain the event type and key attributes of the risk event as the risk event identification result, and obtain the complete causal chain from the source event to the miscarriage result as the risk propagation link; The contribution of risk subplots to the abortion rate of pig farms and the contribution of risk events to the abortion rate of pig farms are obtained as quantitative attribution contribution. Risk event identification, risk propagation links and quantitative attribution contribution are integrated as diagnostic results. Based on the quantitative attribution contribution, early warnings are triggered according to the early warning grading standard, and early warning information is generated and pushed out. Changes in the abortion rate of pig farms are continuously monitored.

6. A pig farm abortion attribution diagnostic system based on a knowledge graph of porcine reproductive and respiratory syndrome (PRRS) risk transmission, characterized in that, The system is used to implement the pig farm abortion attribution diagnosis method based on the knowledge graph of PRRS risk transmission as described in any one of claims 1-5. The system includes: an automated graph construction module, a risk attribution reasoning module, and a risk quantification module. The automated graph construction module constructs a dataset of production events within a preset observation period based on immune data, weather data, herd entry data, domestication data, and biosafety data. According to the predefined risk event criteria, it automatically identifies and extracts risk events from the structured pig farm event data, converts the extracted entities and relationships into knowledge graph nodes and edges, and constructs a knowledge graph of PRRS attribution in pig farms, which is then stored in a graph database. The risk attribution reasoning module is based on the transmission mechanism of porcine reproductive and respiratory syndrome (PRRS). It predefines risk transmission patterns and uses graph pattern matching to search for paths that match the preset risk transmission patterns in the PRRS attribution knowledge graph of pig farms. It integrates multiple risk paths, constructs a risk subgraph, and forms a complete PRRS attribution link table for pig farms. The risk quantification module performs representation learning on the risk subgraphs, and combines context representation learning and time difference decay mechanism to quantify the contribution of each risk subgraph to the pig farm abortion rate. Based on the contribution, the module decomposes the contribution ratio of each risk event in the risk subgraphs using the attention score of the graph attention network, and generates quantitative attribution results.

Images