Large-scale sow breeding management system and method based on abstract digital representation

Abstract

The application discloses a large-scale sow breeding management system and method based on abstract digital representation, and belongs to the technical field of livestock breeding digitalization. The method comprises the following steps: through atomization disassembly and standardization processing of multi-source heterogeneous characteristic data, a unified sow breeding atomization attribute set is constructed; through atomization disassembly and encapsulation of the whole life cycle business logic, a standardized sow breeding atomization method set is constructed; based on the attribute set and the method set, a digital twin instance is dynamically created for each sow, and the instance state is automatically transferred through the combination of real-time data driving; in the instance running, the sow health and the reproduction stage evolution are dynamically evaluated and decided through multi-modal fusion calculation. The corresponding system comprises modules for realizing the above four steps. The application aims to solve the problems of data island and system barrier from the bottom, significantly improves the health monitoring sensitivity and the reproduction management accuracy through the standardized atomization model and intelligent calculation, and provides a unified technical base for the digital upgrading of large-scale breeding.

Description

Technical Field

[0001] This invention belongs to the field of digital technology for livestock farming, and more specifically, relates to a large-scale sow farming management system and method based on abstract digital representation. Background Technology

[0002] As the pig farming industry rapidly develops towards large-scale, intensive, and intelligent operations, breeding sows, as the core production unit of the pig farming industry chain, directly determine the core production performance and economic benefits of pig farms through their breeding efficiency, health level, and management precision. Currently, large-scale pig farms in China have gradually adopted various digital farming management systems, and the collection and application of farming data has become a core trend in industry development. However, in the field of digital farming, existing technologies still face many insurmountable industry pain points: First, existing systems are mostly custom-developed for specific scenarios (such as feeding and health monitoring) or hardware devices (such as specific vision sensors and feeders), lacking a global, standardized data definition and business logic architecture. This results in data incompatibility between different systems and different pig farms, making it difficult to reuse business logic and severely restricting the large-scale promotion of digital collaboration and management experience.

[0003] Second, existing digital models are often deeply tied to front-end sensing hardware. For example, systems based on specific biometric recognition algorithms rely heavily on dedicated cameras and algorithms, making them difficult to adapt to other hardware or scenarios and resulting in poor business logic reusability. On the other hand, systems focused on production process management often limit their digital representation to a macroscopic level of "sow ID - status tag," lacking a refined and dynamic representation of the real-time, microscopic state of individual sows in terms of physiology, behavior, and environment. This leads to insufficient management flexibility and makes it difficult to support real-time, adaptive intelligent decision-making.

[0004] Third, most technical solutions only model isolated stages of the sow's life cycle (such as reproductive management and health monitoring), failing to achieve continuous and seamless digital representation of the entire process from replacement sow entry, mating, pregnancy, farrowing, lactation to culling. Inconsistent data and status transition rules between stages prevent accurate tracking of individual status and cross-stage collaborative optimization.

[0005] Fourth, existing health assessment or status determination models mostly use fixed threshold judgments or simple linear weighting methods. For example, health scores are merely the sum of fixed weights of several indicators, failing to fully consider the complex nonlinear interactions between environmental stress factors and physiological signs, resulting in low sensitivity to identifying hidden health problems. Reproductive stage management often relies on fixed schedules, failing to dynamically determine outcomes based on the sow's real-time nutritional status (e.g., backfat), behavioral activity, and parity decline patterns. This leads to inaccurate stage segmentation, potentially causing reproductive efficiency losses and an increase in non-productive days.

[0006] A search revealed a Chinese patent publication number, CN120077966A, which discloses a biometric-based pig individual identification and health monitoring system. This system achieves high-precision identification of individual pigs through multi-device collaboration and a Transformer architecture, and constructs a digital twin health monitoring module for early warning. However, this system is essentially a front-end perception-driven application architecture, and its digital model is highly dependent on specific biometric recognition algorithms and visual perception hardware. The business logic of this technology is deeply tied to the hardware scenario, making it unable to support the reuse of the underlying digital foundation across systems, devices, and pig farms. Furthermore, the health assessment of sows is merely a simple linear weighted sum, failing to comprehensively consider the impact of environmental factors and sow physiological characteristics on sow health, thus requiring improvement in assessment accuracy.

[0007] Japanese Patent Publication No. JP2021122253A discloses a livestock information management system. This system aims to achieve group management and reproductive prediction of sows by storing sow identifiers, statuses, and reproductive histories, extracting sows with the same status, and assigning them production group identifiers. However, this system primarily focuses on macro-level production planning and process management; its digital representation remains at the form-based association stage of "sow ID - status tag," lacking refined management at the micro-level of individual sows. This management model leads to a heavy reliance on manual input or fixed schedules for sow management, resulting in insufficient flexibility in handling micro-level, real-time intelligent breeding command scheduling. Furthermore, it cannot achieve deep decoupling between business processing logic and underlying perception scenarios, making it difficult to support the dynamic instantiation of high-precision digital twin instances.

[0008] Therefore, in order to achieve a standardized, reusable, traceable, and highly robust digital representation of the entire life cycle of sows, to overcome the core pain points of severe hardware binding and inaccurate macroscopic representation in existing technologies, and to promote the digital and intelligent upgrading of the pig farming industry, it is urgent to develop an abstract digital representation method for large-scale sow farming. Summary of the Invention

[0009] 1. The problem to be solved One of the objectives of this invention is to provide a large-scale sow farming management method based on abstract digital representation, which aims to solve the problems of low standardization of digital representation data throughout the sow's life cycle, deep coupling between business logic and hardware scenarios, insufficient abstraction level and weak scalability of digital models, and low robustness of data processing and accuracy of business judgment.

[0010] Another objective of this invention is to provide a large-scale sow breeding management system based on abstract digital representation, so as to implement the above-mentioned management method.

[0011] 2. Technical Solution To solve the above problems, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for large-scale sow farming management based on abstract digital representation, executed by a computer device, aiming to construct an underlying data architecture supporting digital twin instances in computer memory, comprising the following steps: S1. Constructing an atomized attribute set for sow farming: Through data parsing algorithms, multi-source heterogeneous feature data in sow farming is atomized, deduplicated, standardized, verified, and normalized to generate and manage a standardized atomized attribute set for sow farming. The core purpose of this step is to construct a universal, unambiguous, and non-redundant standard system for all-dimensional feature data of sows through atomized decomposition and standardized definition. S2. Constructing an atomized method set for sow farming: By atomizing and standardizing the business processing logic throughout the entire lifecycle of sow farming, a standardized atomized method set for sow farming is generated and managed. The core purpose of this step is to construct a set of business logic standards that are decoupled from the underlying hardware, reusable, programmable, and scalable through the atomized decomposition and standardized encapsulation of the business processing logic throughout the entire lifecycle of sows. S3. Digital Twin Instantiation and Dynamic Management: Based on the atomized attribute set of sow breeding generated in step S1 and the atomized method set of sow breeding generated in step S2, a digital twin instance is dynamically created and maintained for each sow. By combining real-time data and calling the atomized method set of sows in step S2, the instance state is driven to automatically flow. The core purpose of this step is to dynamically generate a digital twin abstract model of each sow based on the standardized atomized attribute set and method set, so as to realize the digital twin representation of the sow's entire life cycle and the traceability of the entire process state.

[0012] S4. Intelligent assessment and decision-making: During the operation of the digital twin instance described in step S3, the health status and reproductive stage evolution of the sow are dynamically assessed through multimodal fusion calculation, and corresponding management decisions are triggered based on the assessment results.

[0013] As one possible implementation, in step S1, the multi-source heterogeneous feature data includes, but is not limited to, environmental data, physiological characteristic data, production performance data, and behavioral data. The environmental data includes, but is not limited to, pig house temperature, relative humidity, ammonia concentration, and wind speed in the pig house. The physiological characteristic data includes, but is not limited to, rectal body temperature, body weight, backfat thickness, and respiratory rate. The production performance data includes, but is not limited to, daily feed intake, litter size, weaning weight of piglets, and estrus cycle. The behavioral data includes, but is not limited to, standing time, lying time, feeding duration, and activity level.

[0014] As one possible implementation, in step S1, the acquisition of multi-source heterogeneous feature data in sow farming is carried out by collecting multi-source heterogeneous sow full-dimensional feature data through IoT sensor arrays, visual acquisition platforms and farming management terminal equipment deployed in the underlying hardware scenarios of large-scale pig farms.

[0015] As one possible implementation, in step S1, the rules for atomizing the multi-source heterogeneous feature data are: single semantic meaning, indivisible, unambiguous, and non-redundant. The multi-source heterogeneous data is parsed using a data parsing algorithm, and the atomized attribute units generated from the decomposition undergo four levels of standardization checks, including semantic uniqueness verification, non-redundancy verification, value validity verification, and business coverage integrity verification, to ensure that all attribute units comply with the atomization rules and the requirements of the aquaculture business scenario. After standardization and normalization, the basic features representing a single minimum physical state or logical flag are extracted as atomized attribute units.

[0016] As one possible implementation, in step S1, the atomic attribute units that have passed the standardization verification are processed using a robust normalization formula with multi-segment coupling to improve the sensitivity of identifying subtle changes in attribute values ​​within the normal range. The normalization formula is as follows:

[0017] in, The original values ​​of the atomic attribute units to be normalized; , The upper and lower limits of the values ​​for this atomic attribute unit in the corresponding standards for the breeding farm are determined by industry experts; This is the normalized result of the atomic attribute unit; An adaptive compression factor is set based on the noise level and service sensitivity of the atomic attribute unit. Different compression factors ≥1 are set according to the different noise levels and service sensitivity of the atomic dynamic temporal attributes. Values ​​range from 1 to 3.

[0018] Using the above technical solution, the normalization formula is improved and optimized based on the existing normalization formula. Specifically, in the normal value region, the product of the arctan function and the coefficientd tanh function is used to form complementary gradients, which balances the gradient distribution across the entire range and improves data stability. By adjusting the gradient sensitivity in the normal range through the compression coefficient k, the sensitivity for identifying subtle changes in the state attributes of different sows is improved by more than 30%. This solves the problem that the existing normalization formula has a constant gradient, which cannot amplify subtle differences within the normal range, and the state signal is easily submerged.

[0019] As one possible implementation, in step S1, the data belonging to time-series attributes in the verified atomic attribute units are normalized. The time-series attribute data includes, but is not limited to, production performance data. Normalizing the time-series data aims to ensure a more balanced contribution of time-series features in subsequent multimodal fusion calculations. Non-time-series data can also be normalized, while still fully preserving the relative size relationships and business semantics of the original data, thus eliminating dimensional and numerical range differences between different attributes.

[0020] As one possible implementation, step S1 also includes defining standardized attribute units: assigning a unique globally unified identifier to each atomic attribute unit in computer memory; and standardizing the definition of each atomic attribute unit through a metadata description language, specifically limiting the memory data type, precision boundary value, lifecycle validity flag, data source traceability identifier, and concurrent read / write lock permission configuration of the attribute unit.

[0021] As one possible implementation, step S1 also includes constructing a multi-dimensional classification system: constructing a tree-like hierarchical multi-dimensional classification system, establishing classification association rules for atomic attribute units through a data structure mapping table, and realizing structured management and fast retrieval of atomic attribute sets through this classification system.

[0022] As one possible implementation, in step S2, the business processing logic of the key nodes of the sow's entire life cycle includes, but is not limited to, the business processing logic of the entire life cycle of the sow from replacement, mating, pregnancy, farrowing, lactation to weaning and culling, forming a node decision logic map covering the entire business process of sows.

[0023] As one possible implementation, step S2 includes: using logical topology analysis to map the business processing logic of key nodes throughout the sow's life cycle into a directed acyclic graph model, and decomposing it into the smallest independently executable atomic method units that trigger a single target state register write action. These atomic method units are then encapsulated and defined according to the rules of single function, high cohesion and low coupling, standardized input and output, and decoupling from the underlying hardware-aware scenario. The metadata description dictionary of the atomic method unit strictly encapsulates the function operation signature, input parameter serialization structure, output return value deserialization structure, and exception / interrupt handling mount point mapping.

[0024] As one possible implementation, step S2 further includes: constructing a multi-level method call system: the basic data transformation layer is the lowest-level atomic method unit, implementing only basic functions such as single data calculation and state update; the cross-business collaborative scheduling layer is a higher-level method unit, only allowed to call the lower-level single-function atomic method units through standardized application programming to achieve the orchestration and control routing of complex business logic processes. Through this multi-level call system, pig farm managers can flexibly arrange the calling order and combination logic of atomic method units according to the actual business needs of the pig farm, improving the scalability and flexibility of the system.

[0025] As one possible implementation, in step S3, the instantiation of the digital twin involves: encapsulating and integrating the standardized sow breeding attribute set and the standardized sow breeding atomization method set to construct a hierarchical finite state machine-driven digital twin abstract model of the entire life cycle of sows; dynamically creating a digital twin instance for each physical sow in computer memory and assigning a globally unique identifier to the instance; according to the initial physiological stage of the sow, retrieving and loading the corresponding atomized attribute units from the "standardized sow attribute set" as needed through mapping rules; loading the corresponding atomized attribute units from the sow atomized attribute set as needed and assigning initial values ​​to them, thus completing the instantiation.

[0026] As one possible implementation, in step S3, the automatic state transition of the driving instance is achieved through a built-in hierarchical finite state machine. This hierarchical finite state machine uses the macroscopic life cycle stages of the sow as the top-level master state and the microscopic states representing its physiological, behavioral, and environmental characteristics as the bottom-level sub-states. The digital twin instance receives data streams from the underlying IoT sensor array in real time and continuously calculates key indicators by calling the sow atomization method set. When preset conditions are met, it automatically triggers state transitions in the hierarchical finite state machine. That is, when a specific threshold or logical condition is met, the twin's state node transitions are automatically triggered, achieving millisecond-level state synchronization between the physical and digital spaces.

[0027] As one possible implementation, step S4, which involves dynamically assessing the sow's health status, includes calculating a dynamic health stress index. The dynamic health stress index The following formula, which integrates environmental stress and multidimensional physiological deviations, is used to characterize the real-time comprehensive health status of individual sows:

[0028] in, For a moment tTemperature-Humidity Index (THI) extracted from environmental sensors; To correspond to the optimal temperature and humidity thresholds for sows at their current reproductive stage, values ​​are typically less than 72; α is the environmental sensitivity attenuation coefficient. This is used to characterize the physical phenomenon that when the environment deviates from the optimal range, the stress response deteriorates rapidly in a Weibull distribution, that is, the further the environment deviates from the optimal range, the faster the rate of increase in health stress. Calculate the degree of environmental deviation, if (Environment is better than optimal), deviation is 0; if (The environment is worse than optimal), and the deviation is a positive value; For the first data collected from multi-source heterogeneous devices Atomized attribute values ​​of physiological signs; and The first Historical baseline mean and standard deviation of atomized attribute values ​​of physiological signs. For the first Feature weight coefficients of physiological signs and These are the fusion weights for the environmental dimension and the physiological characteristics dimension, respectively. , And satisfy ; This is the physiological regulatory coefficient. Used to control the overall intensity of the impact of physiological deviations on health stress; through The function maps the Euclidean distance of physiological deviations and converges to The interval is used to avoid system state calculation collapse caused by extreme single outliers; m is a weighted sum of m physiological signs, reflecting the overall deviation of physiological signs.

[0029] When the above technical solution is adopted, the dynamic health stress index Based on the original physiological characteristic attributes, this invention introduces the influence of environmental factors on sows. Through multimodal data fusion, it incorporates two major influencing factors: environmental stress and physiological signs. Combined with the physiological laws of the sow's reproductive cycle, a calculation model is constructed. Compared with existing health assessment methods based on fixed thresholds, the dynamic health stress index of this invention can accurately reflect the real-time health status changes of sows, improve the sensitivity of identifying hidden health problems by more than 40%, and significantly improve the reliability and stability of health status assessment.

[0030] As one possible implementation, step S4, which involves dynamically evaluating the reproductive stage evolution of the sow, includes: when the composite execution path makes decisions at the node of state transition, it calls a state evolution evaluation function. To calculate the state transition confidence of the current digital twin instance to the next lifecycle stage, that is, to calculate the state evolution confidence at the node where a decision needs to be made to transition to the reproduction stage. The confidence level is calculated using the following formula, which integrates parity decline factor, real-time nutritional status, and changes in behavioral activity:

[0031] in, For a moment Evolution confidence score for unit execution state transition determination in atomization method; first item The parity decline factor is based on the Logistic function. This refers to the sow's current parity. The inflection point of production capacity decline for a specific product type is typically represented by a value of 5. The decay rate constant is This characterizes the decline in the physical function of sows with increasing parity; The real-time backfat thickness of the sows was collected; This represents the optimal backfat target value for the current life cycle stage. The rate varies with the sow's reproductive stage, being higher during gestation and lower during lactation; Tolerance for back fat fluctuations The smaller the value, the stricter the backfat control requirements; the larger the value, the higher the tolerance. Finally, the Gaussian kernel function is used to calculate the penalty for backfat that is too thin or too fat. The further the sow's current backfat thickness deviates from the target value, the more severe the backfat penalty. The variance of activity level calculated from time-series data of behavioral trajectories; Normal baseline activity; For emergency enhancement coefficient, The higher the value, the more sensitive it is to abnormal activity. Using logarithmic functions Extract the mutation features of behavioral stress and perform normalization processing. This is the maximum value for the activity, to prevent the denominator from being 0 during calculation; For cross-modal coupling scheduling weights between nutritional and behavioral states, The value is determined based on the sow's breed and stage of production; for example, body condition nutrition has a higher priority in the mid-to-late stages of pregnancy. Larger breastfeeding mothers have a higher priority for behavioral health and stress management during lactation. Too small.

[0032] Using the above technical solution, based on the original formula, the state evolution confidence of this invention... The formula innovatively incorporates the influence of sow parity decline patterns when calculating state transition confidence, and combines it with backfat nutritional status and behavioral activity changes to fit the physiological state evolution pattern of sows. It captures the changes in sow parity and different production stages. Compared with existing technologies that rely solely on fixed time points for reproductive stage transitions, the state evolution evaluation function of this scheme can combine the real-time physiological state of individual sows to accurately determine the optimal timing for stage transitions, improving the stage determination accuracy by more than 50% and significantly reducing the loss of breeding efficiency caused by unreasonable stage division.

[0033] As one possible implementation, step S4 further includes: when the calculated dynamic health stress index... A yellow health warning is triggered when the temperature falls below the first warning threshold; when When the threshold is lower than the second warning threshold, a red disease alert is triggered and a diagnostic and treatment suggestion is generated; the second warning threshold is lower than the first warning threshold. When planning a transition to the reproduction phase, the calculated state evolution confidence is... The digital twin instance's breeding stage status will only be transitioned when the threshold for stage transition is exceeded, and feeding or management instructions will be updated accordingly.

[0034] As one possible implementation, every attribute update instruction, method call execution, and state transition trajectory during the operation of the digital twin instance will generate a log with a high-precision timestamp. After the system calls the atomic method unit to execute business logic and update the memory state data, the state update record is hashed and stored on the blockchain.

[0035] The method of this invention can realize seamless, standardized, reusable and scalable digital representation of individual sows throughout their entire life cycle from replacement to culling. It solves the industry pain points of data silos, poor logic reusability and strong coupling with hardware from the bottom up, and provides a unified and standardized underlying digital foundation for various intelligent breeding applications, promoting the digital upgrade of large-scale pig farms at low cost and high efficiency.

[0036] A second aspect of the present invention provides a large-scale sow farming management system based on abstract digital representation, comprising: The atomic attribute modeling module is used to atomically decompose and standardize multi-source heterogeneous feature data in sow farming, and to construct and manage a standardized set of atomic attributes for sow farming. The atomization method encapsulation module is used to atomically decompose and standardize the business processing logic throughout the entire lifecycle of sow farming, and to build and manage a standardized set of atomized methods for sow farming. The digital twin instantiation module is used to dynamically create and maintain a digital twin instance for each sow based on the atomized attribute set and atomized method set for sow breeding, and to drive the state of the digital twin instance to automatically flow according to real-time data. The multimodal fusion computing module is used to provide dynamic intelligent assessment calculations of the health status and reproductive stage evolution status of the digital twin instance.

[0037] 3. Beneficial effects Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention completely solves the long-standing industry pain points of inconsistent sow entity standards and deep coupling between business logic and hardware scenarios in the breeding system by atomically decomposing and standardizing the sow characteristic indicators and business logic. It realizes cross-system and cross-pig farm data interoperability and business logic reuse, and greatly reduces the development and reconstruction costs of digital systems for large-scale pig farms.

[0038] (2) This invention constructs a calculation model that conforms to the physiological laws of sows by using a multi-segment coupling normalization method, a multi-modal fusion dynamic health stress index and a state evolution evaluation function. Compared with the traditional fixed threshold judgment method, it improves the sensitivity of sow health abnormality identification and the accuracy of reproductive stage judgment by more than 40%, effectively reducing the risk of misjudgment in breeding decisions and improving sow breeding efficiency and health management accuracy.

[0039] (3) The present invention builds a digital twin architecture based on hierarchical finite state machine, realizing millisecond-level synchronization between physical space and digital space, which can support the rapid development and implementation of various intelligent breeding applications, providing reliable technical support for the refined and intelligent management of the entire life cycle of large-scale pig breeding, and effectively promoting the digital upgrade of the pig breeding industry. Attached Figure Description

[0040] Figure 1 This is the overall flowchart of the management method of the present invention.

[0041] Figure 2 This is a flowchart illustrating the construction process of the atomized attribute set for sow breeding in this invention.

[0042] Figure 3 This is a flowchart illustrating the implementation of the atomization method set for sow breeding of the present invention.

[0043] Figure 4 This is a flowchart illustrating the implementation process of instantiating the digital twin in this invention.

[0044] Figure 5 The figures show the results of the dynamic monitoring experiment of sows throughout their entire life cycle in the examples and comparative examples.

[0045] Figure 6This is a comparison chart of the core production performance of sows in the example and the comparative examples. Detailed Implementation

[0046] The present invention will be further described below with reference to specific embodiments.

[0047] Example This embodiment was carried out in a large-scale pig farm, where the breed of pigs is Large White sows. The farm has 1,200 breeding sows and has a complete IoT sensing and digital management foundation. The following sensing devices are uniformly deployed on the ground floor of the experimental pigsty, with the following hardware environment: Temperature and humidity sensor (Kunlun Coast JWSK-6ACW): sampling frequency once every 10 minutes, deployed in the center of the pigsty 1.5m above the ground; Ear tag-type body temperature sensor (Ruichu Technology RHT-01 smart ear tag): sampling frequency once every 30 minutes, one for each sow; Automatic feeding system (Nanshang Agricultural Science NS-2000 intelligent sow feeding station): records the daily feed intake and feeding time of each sow in real time; Backfat measuring instrument (Top Cloud Agriculture TP-BB-1): measures the backfat thickness of sows once a week; AI visual behavior acquisition system (Hikvision DS-2CD3T46WD-I3+ agricultural AI algorithm platform): sampling frequency once every 5 minutes, collecting real-time sow movement trajectory and behavior data; Breeding management terminal: used for manual input of production data such as mating, farrowing, and immunization.

[0048] This embodiment focuses on 100 sows (50 gilts and 50 multiparous sows, parity 1-4) and employs the large-scale sow breeding management system and method based on abstract digital representation described in this invention for full life-cycle management. The experimental period for gilts is 90 days (covering the gilt rearing to mating stage); the experimental period for multiparous sows is 150 days (covering the complete reproductive cycle from mating, gestation, farrowing, lactation to weaning). This embodiment describes a large-scale sow breeding management method based on abstract digital representation. Figure 1 The overall process shown is executed, and the specific steps are as follows: S1. Constructing the atomized attribute set for sow farming: like Figure 2 As shown, the multi-source heterogeneous feature data collected in sow farming is atomically decomposed, deduplicated, standardized, and normalized using data parsing algorithms to generate and manage a standardized set of atomic attributes for sow farming. The core objective of this step is to construct a universal, unambiguous, and non-redundant standard system for all-dimensional feature data of sows through atomic decomposition and standardized definition. The specific steps are as follows: S11. Extraction of multi-source heterogeneous feature data Through the aforementioned deployed sensing devices, namely the IoT sensor array (temperature and humidity sensor, ear tag-type body temperature sensor, automatic feeding system, backfat measuring instrument), visual acquisition platform (AI visual behavior acquisition system), and breeding management terminal, multi-source heterogeneous sow full-dimensional characteristic data are collected, including: Environmental data: pigsty temperature, relative humidity, ammonia concentration, and wind speed in the pigsty; Physiological data: rectal temperature, weight, backfat thickness, respiratory rate; Production performance data: daily feed intake, litter size, weaning weight of piglets, estrus cycle; Behavioral data: standing time, lying time, feeding duration, and motor activity.

[0049] S12, Atomization and Deduplication The multi-source heterogeneous feature data extracted in step S11 is atomized and decomposed. The core objective is to construct a universal, unambiguous, and non-redundant standard system for all-dimensional feature data of sows. Following the atomization rules of single semantics, indivisibility, unambiguity, and non-redundancy, the collected multi-source heterogeneous feature data is decomposed using data parsing algorithms, and the smallest indivisible units after decomposition are deduplicated. For example, daily feed intake is decomposed into the smallest indivisible units such as cumulative daily feed intake, average feed intake per feeding, number of feedings per day, and deviation of feed intake from historical benchmarks. This embodiment ultimately extracts 87 atomized attribute units.

[0050] S13, Level 4 Standardization Verification Perform a four-level normalization check on all atomic attribute units disassembled in step S12. Units that fail the check are either re-disassembled or removed. Semantic uniqueness verification: Ensure that each attribute unit corresponds to only one physical meaning, such as body temperature only referring to rectal temperature, eliminating ambiguity of body surface temperature.

[0051] Non-redundancy check: Attribute units with a similarity greater than 95% are removed by calculating cosine similarity.

[0052] Value validity verification: The valid value range of each attribute is set according to the "Technical Specifications for Large-scale Pig Production" (NY / T 2664-2014), such as the normal body temperature range of sows being 38.0℃~39.5℃.

[0053] Business coverage integrity verification: Ensure that all attribute units cover the business needs of sows throughout their entire life cycle from replacement to culling.

[0054] S14, Normalization Processing The atomic attribute units that pass the verification in step S13 are normalized using the following multi-segment coupling robust normalization formula.

[0055]

[0056] in, The original values ​​of the atomic attribute units to be normalized; , The upper and lower limits of the values ​​for this atomic attribute unit in the corresponding standards for the breeding farm are determined by industry experts; This is the normalized result of the atomic attribute unit; An adaptive compression coefficient, set based on the noise level and business sensitivity of the atomic attribute unit, is used. Different compression coefficients ≥1 are set according to the different noise levels and business sensitivity of the atomic dynamic temporal attributes. The adaptive compression coefficient k is set according to the principle that the lower the noise level and the higher the business sensitivity, the larger the k value, with the value strictly controlled between 1.0 and 3.0. See Table 1 below for details. The standard value ranges of all indicators are formulated based on the Ministry of Agriculture and Rural Affairs' "Data Elements for Pig Farming" (NY / T3774-2020) and "Technical Specifications for Large-Scale Pig Farming" (NY / T2664-2014).

[0057] Table 1. Values ​​of k when normalizing the verified atomized attribute units in this embodiment.

[0058] S2. Constructing an atomized method set for sow farming: according to... Figure 3 The execution process of the standardized encapsulation of the sow atomization method unit, as shown, involves the atomization and standardized encapsulation of the business processing logic throughout the entire lifecycle of sow farming, generating and managing a standardized set of atomized methods for sow farming. The core objective of this step is to construct a standardized business logic system that is decoupled from the underlying hardware, reusable, programmable, and scalable through the atomization and standardized encapsulation of the business processing logic throughout the sow lifecycle. The specific steps are as follows: S21. Full extraction of business logic throughout the entire life cycle: Extract the processing logic of sows from the entry of replacement sows, estrus detection, mating, pregnancy diagnosis, pregnancy management, farrowing, lactation, weaning to culling, and construct a decision logic graph containing 127 business nodes.

[0059] S22. Atomized Decomposition of Business Logic: Using logical topology analysis, complex business logic is mapped to a directed acyclic graph and forcibly decomposed into atomic method units that trigger a single target state register write action. For example, pregnancy stage management is decomposed into 19 atomic method units, such as updating backfat attributes, calculating state evolution confidence, triggering mid-pregnancy transition, and generating feeding adjustment instructions, ultimately resulting in 73 independently executable atomic method units.

[0060] S23. Standardized encapsulation of atomic method units: Encapsulation is carried out according to the rules of single function, high cohesion and low coupling, and decoupling from hardware. The metadata dictionary of each method unit includes: function operation signature, input parameter serialization structure, output return value deserialization structure, and exception interrupt handling mount point.

[0061] S24. Construct a multi-level method call system: Construct a two-layer call system: Basic data transformation layer: Contains all atomic method units, implementing only a single basic function; Cross-business collaborative scheduling layer: Calls basic layer methods through standardized APIs to orchestrate complex business processes. For example, the mating confirmation process is implemented by calling four atomic method units: querying estrus status, updating mating records, initializing pregnancy attributes, and triggering pregnancy status transition.

[0062] S3. Digital Twin Instantiation and Dynamic Management: Based on the atomized attribute set for sow breeding generated in step S1 and the atomized method set for sow breeding generated in step S2, digital twin instances are dynamically created and maintained for individual sows. By combining real-time data with calls to the atomized method set for sows in step S2, the instance state is automatically transitioned. The core objective of this step is to dynamically generate a digital twin abstract model of an individual sow based on a standardized atomized attribute set and method set, achieving digital twin representation and full-process state traceability of the sow throughout its entire lifecycle. Specifically, according to... Figure 4 The following is the execution process for the dynamic instantiation of the sow digital twin model: S31. Dynamic Creation and Initialization of Digital Twin Instances: The standardized sow breeding attribute set and the standardized sow breeding atomization method set are encapsulated and integrated to construct a hierarchical finite state machine-driven digital twin abstract model of the entire life cycle of sows. A digital twin instance is dynamically created in computer memory for each physical sow, and a globally unique identifier is assigned to the instance. According to the initial physiological stage of the sow, the corresponding atomized attribute units are retrieved and loaded from the "standardized sow attribute set" on demand through mapping rules. The corresponding atomized attribute units are loaded from the sow atomized attribute set on demand and assigned initial values ​​to complete the instantiation.

[0063] S32. Hierarchical Finite State Machine Configuration and Automatic State Transition: In the digital twin architecture, a hierarchical finite state machine is constructed, with the macro life cycle stages of sows (replacement → mating → pregnancy → farrowing → lactation → empty pregnancy → culling) as the top-level master state, and the micro-physiological, behavioral, and environmental characteristics of each stage as the bottom-level sub-states.

[0064] The twin instance receives streaming data from the underlying IoT sensor array in real time and continuously calculates key metrics using a multi-level method call system. When specific thresholds or logical conditions are met, the twin's state node transitions are automatically triggered, achieving millisecond-level state synchronization between the physical and digital spaces.

[0065] S4. Intelligent Assessment and Decision-Making: During the operation of the digital twin instance described in step S3, the health status and reproductive stage evolution of the sow are dynamically assessed through multimodal fusion calculation, and corresponding management decisions are triggered based on the assessment results. Specifically: (1) Calculate the dynamic health stress index When the calculated dynamic health stress index A yellow health warning is triggered when the temperature falls below the first warning threshold; when When the threshold is lower than the second warning threshold, a red disease alert is triggered and a diagnostic and treatment suggestion is generated; the second warning threshold is lower than the first warning threshold. The following formula, which integrates environmental stress and multidimensional physiological deviations, is used to characterize the real-time comprehensive health status of individual sows:

[0066] in, For a moment t Temperature-Humidity Index (THI) extracted from environmental sensors; To correspond to the optimal temperature and humidity thresholds for sows at their current reproductive stage, values ​​are typically less than 72; α is the environmental sensitivity attenuation coefficient. This is used to characterize the physical phenomenon that when the environment deviates from the optimal range, the stress response deteriorates rapidly in a Weibull distribution, that is, the further the environment deviates from the optimal range, the faster the rate of increase in health stress. Calculate the degree of environmental deviation, if (Environment is better than optimal), deviation is 0; if (The environment is worse than optimal), and the deviation is a positive value; For the first data collected from multi-source heterogeneous devices Atomized attribute values ​​of physiological signs; and The first Historical baseline mean and standard deviation of atomized attribute values ​​of physiological signs. For the first Feature weight coefficients of physiological signs and These are the fusion weights for the environmental dimension and the physiological characteristics dimension, respectively. , And satisfy ; This is the physiological regulatory coefficient. Used to control the overall intensity of the impact of physiological deviations on health stress; through The function maps the Euclidean distance of physiological deviations and converges to The interval is used to avoid system state calculation collapse caused by extreme single outliers; m is a weighted sum of m physiological signs, reflecting the overall deviation of physiological signs.

[0067] (2) Calculate the confidence level of state evolution When planning a transition to the reproduction phase, the calculated state evolution confidence level... The digital twin instance's breeding stage status will only be transitioned when the threshold for stage transition is exceeded, and feeding or management instructions will be updated accordingly.

[0068] When making node decisions during state transitions, the composite execution path calls a state evolution evaluation function. To calculate the state transition confidence of the current digital twin instance to the next lifecycle stage, that is, to calculate the state evolution confidence at the node where a decision needs to be made to transition to the reproduction stage. The confidence level is calculated using the following formula, which integrates parity decline factor, real-time nutritional status, and changes in behavioral activity:

[0069] in, For a moment Evolution confidence score for unit execution state transition determination in atomization method; first item The parity decline factor is based on the Logistic function. This refers to the sow's current parity. The inflection point of production capacity decline for a specific product type is typically represented by a value of 5. The decay rate constant is This characterizes the decline in the physical function of sows with increasing parity; The real-time backfat thickness of the sows was collected; This represents the optimal backfat target value for the current life cycle stage. The rate varies with the sow's reproductive stage, being higher during gestation and lower during lactation; Tolerance for back fat fluctuations The smaller the value, the stricter the backfat control requirements; the larger the value, the higher the tolerance. Finally, the Gaussian kernel function is used to calculate the penalty for backfat that is too thin or too fat. The further the sow's current backfat thickness deviates from the target value, the more severe the backfat penalty. The variance of activity level calculated from time-series data of behavioral trajectories; Normal baseline activity; For emergency enhancement coefficient, The higher the value, the more sensitive it is to abnormal activity. Using logarithmic functions Extract the mutation features of behavioral stress and perform normalization processing. This is the maximum value for the activity, to prevent the denominator from being 0 during calculation; For cross-modal coupling scheduling weights between nutritional and behavioral states, The value is determined based on the sow's breed and stage of production; for example, body condition nutrition has a higher priority in the mid-to-late stages of pregnancy. Larger breastfeeding mothers have a higher priority for behavioral health and stress management during lactation. Too small.

[0070] This embodiment presents a large-scale sow farming management system based on abstract digital representation, including modules that implement the above four steps: The atomic attribute modeling module is used to atomically decompose and standardize multi-source heterogeneous feature data in sow farming, and to construct and manage a standardized set of atomic attributes for sow farming. The atomization method encapsulation module is used to atomically decompose and standardize the business processing logic throughout the entire lifecycle of sow farming, and to build and manage a standardized set of atomized methods for sow farming. The digital twin instantiation module is used to dynamically create and maintain a digital twin instance for each sow based on the atomized attribute set and atomized method set for sow breeding, and to drive the state of the digital twin instance to automatically flow according to real-time data. The multimodal fusion computing module is used to provide dynamic intelligent assessment calculations of the health status and reproductive stage evolution status of the digital twin instance.

[0071] Comparative Example The comparative example and embodiment are basically the same as a large-scale sow farming management method and system based on abstract digital representation, except that: 1. Data normalization uses the traditional linear normalization method, which has a constant gradient and cannot amplify subtle differences within the normal range. The normalization formula is:

[0072] 2. The health assessment uses a fixed threshold method, and an alert is only triggered when a certain indicator in Table 2 exceeds the preset range. Table 2. Situations triggering early warnings in the comparative examples.

[0073] 3. The management of the reproductive stage adopts a fixed timetable, such as automatically determining the farrowing period 114 days after mating, without taking into account the real-time physiological state of the sow and the parity decline pattern.

[0074] Analysis of results from examples and comparative examples: 1. Comparison of normalization results for all indicators: By measuring the original values ​​of different indicators, comparative experiments were conducted using a multi-segment coupled robust normalization method and a traditional normalization method. The results are shown in Table 3 below: Table 3. Normalization results of each atomized property unit in the Examples and Comparative Examples

[0075] In Table 3, the difference amplification rate = (normalized difference of the example - normalized difference of the comparison example) / normalized difference of the comparison example. It can be seen that the multi-segment coupling robust normalization method of the present invention has a weighted average amplification rate of 33.5% for subtle differences within the normal range. It can effectively capture early health abnormality signals of sows and provide a high-quality data foundation for subsequent intelligent assessment and decision-making. 2. Calculation of Dynamic Health Stress Index. Taking the health assessment of a sow as an example, the dynamic stress index of its entire production cycle is calculated. According to the system settings of this invention: First warning threshold (yellow warning): This indicates that the sow has a mild health abnormality or stress response; the second warning threshold (red warning): This indicates a significant health risk in the sow, requiring immediate intervention. For example... Figure 5 The following is a graph showing the results of the dynamic monitoring experiment on the entire life cycle of this sow. The comparison table of the full-cycle monitoring data is shown in Table 4 below: Table 4. Calculation results of dynamic health stress index for both the examples and comparisons.

[0076] The multimodal fusion dynamic health stress index calculation method proposed in this invention is significantly superior to the traditional fixed-threshold health assessment method used in the control group. This index simultaneously integrates environmental stress and multidimensional physiological sign deviations, enabling it to accurately capture subtle health changes throughout the sow's life cycle. It provides an average early warning time of 26 hours for health problems such as subclinical mastitis and early heat stress, and its recognition sensitivity is improved by 43.5%. Figure 5 As shown, this method triggers an early warning of red health abnormalities as early as day 185 of the second trimester. After timely intervention, the sows did not develop serious health problems. In contrast, the control group only issued an alarm when the sows showed obvious clinical symptoms such as fever and a sudden drop in feed intake. Furthermore, it could not distinguish between normal physiological stress such as parturition and lactation and pathological abnormalities, resulting in a false alarm rate that was more than three times higher than that of this method. This fully verifies the accuracy and early warning of the health assessment model of this invention.

[0077] 3. State evolution confidence calculation: At the transition node in the pregnancy stage, the state evolution evaluation function is called. Calculate the jump confidence when Automatic stage transition is triggered at certain times, where: Based on the characteristics of the Large White sow breed, reproductive performance begins to decline significantly after the 5th parity, and is taken as 5; The decay rate constant was obtained by fitting the reproductive data of 5,000 sows in the pig farm from 2019 to 2024. The measurement is dynamically adjusted according to the stage of pregnancy: 16mm in the first trimester (1-30 days), 18mm in the second trimester (31-90 days), and 20mm in the third trimester (91-114 days). The tolerance for backfat fluctuations is in line with the actual management precision of large-scale pig farms; , The baseline and maximum activity levels were taken from healthy pregnant sows.

[0078] Depend on Figure 5 It can be seen that, compared with the control group, the experimental group had a 96.2% accuracy rate in determining the pregnancy stage, while the control group had a 72.5% accuracy rate, representing an improvement of 50.6%.

[0079] 4. Comparison results of core reproductive performance of sows: From Figure 6 It can be seen that all core production performance indicators of the experimental group were significantly better than those of the control group. Average number of healthy piglets per litter: 12.8 in the experimental group, 11.6 in the control group, an increase of 10.3%; Number of weaned piglets per sow per year (PSY): 27.2 in the experimental group, 24.8 in the control group, an increase of 9.7%; Lactation survival rate: 96.5% in the experimental group, 90.2% in the control group, an increase of 6.3 percentage points; Average number of non-productive days: 46.3 days in the experimental group, 48.7 days in the control group, a decrease of 2.4 days; Feed consumption during gestation: an average of 286 kg per sow in the experimental group, 294.2 kg per sow in the control group, a decrease of 8.2 kg.

[0080] Experimental results show that this invention significantly improves the accuracy and intelligence of sow breeding management through techniques such as atomized modeling, multi-segment coupling robust normalization, hierarchical finite state machine, and multimodal fusion evaluation. It completely solves the core pain points of existing technologies, such as data silos, hardware binding, coarse representation, and simple evaluation, and has broad prospects for industrial application.

[0081] The invention and its embodiments have been described above illustratively. This description is not restrictive, and the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. The accompanying drawings are only one embodiment of the invention, and the actual structure is not limited thereto. No reference numerals in the claims should limit the scope of the claims. Therefore, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the invention, such design should fall within the scope of protection of this patent. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Multiple elements stated in the product claims may also be implemented by a single element through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for large-scale sow farming management based on abstract digital representation, characterized in that: Performed by a computer device, including the following steps: S1. Constructing an atomized attribute set for sow farming: By performing atomized decomposition, deduplication, standardization verification, and normalization on multi-source heterogeneous feature data in sow farming, a standardized atomized attribute set for sow farming is generated and managed. S2. Construct an atomized method set for sow farming: By atomizing and standardizing the business processing logic throughout the entire lifecycle of sow farming, a standardized atomized method set for sow farming is generated and managed. S3. Digital Twin Instantiation and Dynamic Management: Based on the atomized attribute set of sow breeding generated in step S1 and the atomized method set of sow breeding generated in step S2, digital twin instances are dynamically created and maintained for individual sows. By combining real-time data and calling the atomized method set of sow breeding in step S2, the instance state is driven to automatically change. S4. Intelligent assessment and decision-making: During the operation of the digital twin instance described in step S3, the health status and reproductive stage evolution of the sow are dynamically assessed through multimodal fusion calculation, and corresponding management decisions are triggered based on the assessment results.

2. The method for large-scale sow farming management based on abstract digital representation according to claim 1, characterized in that: In step S1, the rules for atomically decomposing the multi-source heterogeneous feature data are single semantic, indivisible, unambiguous, and non-redundant; and the atomic attribute units generated by the decomposition are subjected to four levels of standardized verification, including semantic uniqueness verification, non-redundancy verification, value validity verification, and business coverage integrity verification.

3. The method for large-scale sow farming management based on abstract digital representation according to claim 2, characterized in that: In step S1, the atomic attribute units that have passed the standardization verification are processed using a robust normalization formula with multi-segment coupling to improve the sensitivity of identifying subtle changes in attribute values ​​within the normal range. The normalization formula is as follows: ； in, The original values ​​of the atomic attribute units to be normalized; , The upper and lower limits of the value of this atomized attribute unit in the corresponding standard of the breeding farm; This is the normalized result of the atomic attribute unit; An adaptive compression factor set based on the noise level and business sensitivity of the atomic attribute unit.

4. The method for large-scale sow farming management based on abstract digital representation according to claim 1, characterized in that: Step S2 includes: using logical topology analysis to map the business processing logic of the entire life cycle of the sow into a directed acyclic graph model, and decomposing it into the smallest independently executable atomic method unit that triggers a single target state register write action, and encapsulating and defining it according to the rules of single function, high cohesion and low coupling.

5. The method for large-scale sow farming management based on abstract digital representation according to claim 1, characterized in that: In step S3, the instantiation of the digital twin is as follows: a digital twin instance is dynamically created in the computer memory for each physical sow, and a globally unique identifier is assigned to the instance; according to the initial physiological stage of the sow, the corresponding atomic attribute units are loaded from the sow atomic attribute set as needed and assigned initial values ​​to complete the instantiation.

6. The method for large-scale sow farming management based on abstract digital representation according to claim 5, characterized in that: In step S3, the automatic state transition of the driving instance is achieved through a built-in hierarchical finite state machine. The hierarchical finite state machine takes the macroscopic life cycle stage of the sow as the top-level master state and the microscopic state representing its physiological, behavioral, and environmental characteristics as the bottom-level sub-state. The digital twin instance receives data streams in real time and calls the sow atomization method set for calculation. When preset conditions are met, the state transition in the hierarchical finite state machine is automatically triggered.

7. The method for large-scale sow farming management based on abstract digital representation according to claim 1, characterized in that: Step S4 involves dynamically assessing the sow's health status, including calculating the dynamic health stress index. The dynamic health stress index The following formula was used to calculate the results, which integrates environmental stress and multidimensional physiological sign deviations: ； in, For a moment t Temperature and humidity index; α represents the optimal temperature and humidity thresholds for the current breeding stage; α is the environmental sensitivity attenuation coefficient, α∈[0,1]; For the first Atomized attribute values ​​of physiological signs, and The first Historical baseline mean and standard deviation of atomized attribute values ​​of physiological signs. For the first The characteristic weighting coefficient of each physiological sign and These are the fusion weights for the environmental dimension and the physiological characteristics dimension, respectively. , And satisfy ; This is the physiological regulatory coefficient. .

8. The method for large-scale sow farming management based on abstract digital representation according to claim 7, characterized in that: In step S4, the dynamic evaluation of the sow's reproductive stage evolution includes: calculating the state evolution confidence at nodes where a reproductive stage transition decision is required. The confidence level is calculated using the following formula, which integrates parity decline factor, real-time nutritional status, and changes in behavioral activity: ； in, This refers to the sow's current parity. The inflection point of declining production capacity for product standards; The decay rate constant; Real-time back fat thickness; This represents the optimal backfat target value for the current life cycle stage. Tolerance for backfat fluctuations; The variance of activity level calculated from time-series data of behavioral trajectories; Normal baseline activity; This is the emergency enhancement coefficient; The scheduling weights are used for cross-modal coupling of nutritional and behavioral states.

9. The method for large-scale sow farming management based on abstract digital representation according to claim 8, characterized in that: Step S4 further includes: when the calculated dynamic health stress index is obtained A yellow health warning is triggered when the temperature falls below the first warning threshold; when When the threshold is lower than the second warning threshold, a red disease alert is triggered and a diagnostic and treatment suggestion is generated; the second warning threshold is lower than the first warning threshold. When planning to jump to the breeding phase, the calculated state evolution confidence is... The digital twin instance's breeding stage status will only be transitioned when the threshold for stage transition is exceeded, and feeding or management instructions will be updated accordingly.

10. A large-scale sow farming management system based on abstract digital representation, characterized in that: include The atomic attribute modeling module is used to atomically decompose and standardize multi-source heterogeneous feature data in sow farming, and to construct and manage a standardized set of atomic attributes for sow farming. The atomization method encapsulation module is used to atomically decompose and standardize the business processing logic throughout the entire lifecycle of sow farming, and to build and manage a standardized set of atomized methods for sow farming. The digital twin instantiation module is used to dynamically create and maintain a digital twin instance for each sow based on the atomized attribute set and atomized method set for sow breeding, and to drive the state of the digital twin instance to automatically flow according to real-time data. The multimodal fusion computing module is used to provide dynamic intelligent assessment calculations of the health status and reproductive stage evolution status of the digital twin instance.

Images