An egg nutrition component regulation method and system using a knowledge graph

Abstract

The application discloses an egg nutrition component regulation method and system applying a knowledge graph, and belongs to the field of agricultural information intelligent control. The method comprises the following steps: multi-dimensional feature space construction, collecting environmental and poultry physiological characteristic data through a sensor array, forming a heterogeneous feature space through alignment and noise reduction; knowledge graph correlation mapping and nutrition gap quantification, inputting real-time data into a pre-constructed environmental physiological nutrition knowledge graph, activating stress nodes by using sub-graph matching, quantifying interference intensity along a causal path, and calculating an initial nutrition compensation vector; closed-loop precise regulation, controlling the system to drive a precise feeding device to add nutritional substances according to the compensation vector, and performing secondary correction by taking the body surface temperature of the poultry as a feedback index, so as to realize dynamic matching. The application can analyze the complex influence of environmental stress on poultry metabolism in real time, accurately quantify the nutrition gap and perform closed-loop compensation, and significantly improve the consistency of egg quality and the breeding benefit.

Description

Technical Field

[0001] This application belongs to the technical field of agricultural information intelligent control, specifically relating to a method and system for regulating the nutritional components of poultry eggs using knowledge graphs. Background Technology

[0002] With the continuous development of precision animal husbandry, improving poultry egg quality and breeding efficiency through dynamic regulation of diet nutrition has become an important research direction in the industry. Under high-density intensive farming models, instantaneous fluctuations in the poultry house environment, such as sudden changes in temperature and humidity and accumulation of harmful gases, can easily induce physiological stress in poultry, thereby interfering with the body's metabolic pathways and affecting the deposition efficiency of functional nutrients in eggs, such as vitamins, trace elements, and unsaturated fatty acids.

[0003] Currently, nutritional regulation techniques for environmental fluctuations typically rely on environmental sensors to monitor single or a few environmental parameters within poultry houses, followed by post-event adjustments using standardized static diet formulations. However, these methods have significant limitations: firstly, there is a complex nonlinear relationship between environmental stimuli and poultry physiological metabolism, making it difficult for existing methods to capture the cumulative effects of multi-pathway metabolic shifts under stress in real time; secondly, traditional regulation strategies lack a closed-loop integration of environmental perception, physiological feedback, and nutritional execution, resulting in nutritional compensation decisions lagging behind environmental changes and failing to achieve precise and dynamic nutritional supply based on the immediate physiological state of poultry.

[0004] Therefore, how to construct a quantitative mapping path that can analyze environmental fluctuations in real time to the nutritional metabolic gap, and form a self-correcting closed-loop precise regulation mechanism, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for regulating the nutritional components of poultry eggs using knowledge graphs, which can effectively solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for regulating the nutritional components of poultry eggs using knowledge graphs includes the following specific steps: Multidimensional feature space construction: Environmental feature data and physiological feature data including poultry body surface temperature and activity sound spectrum are collected in real time by a sensor array deployed in the poultry house. The above data with different sampling frequencies are time-axis aligned and noise-reduced to construct a multidimensional heterogeneous feature space. Knowledge graph association mapping and nutritional gap quantification: Real-time data from the multidimensional heterogeneous feature space is input into a pre-constructed environmental physiological nutrition knowledge graph that integrates metabolic pathway data. The mapping position of the current production state in the knowledge graph is identified by the subgraph matching algorithm to activate the corresponding stress node. Based on the activated stress node, the interference intensity of environmental fluctuations on each metabolic pathway is quantified along the causal path of the knowledge graph to calculate the initial nutritional compensation vector. Closed-loop precision control: The control system generates control commands based on the initial nutrient compensation vector, driving the precision feeding device to add functional nutrients to the basal diet in real time; after feeding is completed, the poultry's body surface temperature is continuously monitored as a feedback indicator, and the initial nutrient compensation vector is corrected a second time based on the magnitude and rate of change of the body surface temperature deviation, generating a corrected nutrient compensation vector and driving the precision feeding device again for compensation until the body surface temperature returns to normal, realizing dynamic closed-loop matching between nutrient supply and instantaneous environmental stress state.

[0007] Furthermore, the environmental physiological nutrition knowledge graph is a deep semantic network containing five types of entity nodes and the logical relationship edges between them. Its construction process specifically includes: The knowledge graph is constructed offline in advance, and the five types of entity nodes are defined as follows: environmental triggering nodes, corresponding to measurable abnormal environmental states in poultry houses; physiological regulation nodes, corresponding to the physiological responses of poultry organisms under environmental stimuli; enzyme activity nodes, including carbonic anhydrase nodes involved in calcium metabolism regulation and alkaline phosphatase nodes involved in phosphorus metabolism regulation; protein transport nodes, including the calcium-binding protein D28k node responsible for intestinal calcium absorption and the vitellogenin transporter node responsible for vitellogenin transport and deposition; and nutrient deposition nodes, corresponding to the final poultry egg quality indicators. The initial weights of the logical relationship edges are calculated by weighted average of the statistical correlation of historical breeding records, the enzymatic reaction kinetic constants of the metabolic pathway data, and the scores of the poultry nutrition expert knowledge base, and are continuously updated through a self-learning mechanism after the system is running.

[0008] Furthermore, the physiological characteristic data specifically includes the distribution of avian body surface temperature and the acoustic spectrum characteristics of activity; the process of constructing the multidimensional heterogeneous feature space specifically includes: An infrared thermal imaging sensor array is installed on the top of the poultry house at a density of one monitoring area per 100 square meters. The array continuously collects images of the body surface temperature distribution of poultry clusters and extracts the average temperature of three heat dissipation sensitive areas: the comb, eye socket, and feet. The arithmetic mean of the three temperature averages is used as the average body surface temperature output value at the current moment. An acoustic sensor cluster is used, with one sensor installed on the top of the poultry house at a density of one sensor per 300 square meters, to continuously collect the feeding sounds, calls, and activity spectrum signals of poultry. The acquired audio signals were analyzed with a frame length of 256 milliseconds and a frame shift of 128 milliseconds. After applying a Hanning window, a 7-layer discrete wavelet decomposition was performed. The wavelet coefficients of the 4th, 5th and 6th layers were selected, and the normalized energy values ​​of each layer were calculated. The coefficients were then spliced ​​together in order of frequency band from low to high to form a 3D acoustic spectrum energy feature vector to reflect the energy change characteristics of changes in the feeding rhythm of poultry and abnormal stress screams.

[0009] Furthermore, the process of time axis alignment and noise reduction for data with different sampling frequencies specifically includes: The original signals of each sensor channel of the environmental feature data are denoised in real time using a 5th-order Kalman filter algorithm, and the posterior state estimate is output as the denoised sensor reading. Using a target time interval of 100 milliseconds, a cubic spline interpolation algorithm is used to resample the noise-reduced environmental feature data, the average body surface temperature, and the acoustic spectrum energy feature vector. All channels are unified to a unified time axis with 10 aligned time points per second, ensuring that each time point corresponds to a complete set of multi-channel observation data to form a structured multidimensional heterogeneous feature space.

[0010] Furthermore, the process of quantifying the intensity of environmental fluctuations on each metabolic pathway based on activated stress nodes along the causal path of the knowledge graph to calculate the initial nutrient compensation vector specifically includes: The inference engine starts from the activated stress node and performs a depth-first path search in the knowledge graph, extending downstream along each causal relationship edge until it reaches the nutrient deposition node. The maximum search depth of the path is set to 10 hops. For each activated path retrieved, the negative inhibition coefficient of that path on downstream nutrient deposition nodes is calculated. The negative inhibition coefficient is the product of the weights of all logical relation edges on that path, and its value ranges from 0 to 1. The smaller the value, the stronger the inhibition. When the same nutrient deposition node is pointed to by multiple activation paths at the same time, the minimum value among the negative inhibition coefficients of each path is taken as the comprehensive negative inhibition coefficient of the node. Based on the comprehensive negative inhibition coefficient, the percentage concentration of each type of nutrient that needs to be supplemented under the current stress state is calculated to form the initial nutrient compensation vector.

[0011] Furthermore, the initial nutrient compensation vector is constructed by splicing together the calculated compensation percentage concentrations of various nutrients in a fixed order, specifically a 9-dimensional vector, whose components include the compensation percentage concentrations of vitamin A, vitamin C, vitamin E, methionine, lysine, sodium bicarbonate, potassium chloride, organic selenium, and chelated zinc in sequence. After generating the initial nutrient compensation vector, the system synchronously accesses the aquaculture cost database, calculates the cost increment of the current control scheme based on the purchase price of various nutrients and the corresponding compensation amount, and if there are multiple combination schemes that can achieve the same nutrient deposition target, the scheme with the lowest total cost increment is selected first, and the optimized compensation amount is updated to the initial nutrient compensation vector and passed to subsequent steps for execution.

[0012] Furthermore, the precision feeding device in the closed-loop precise control step consists of a micro-metering pump driven by a stepper motor and a pneumatic ultrasonic atomizing spray head; the process of the control system generating control commands and adding materials specifically includes: The control system reads the values ​​of each component in the initial nutrient compensation vector, combines the current operating speed of the feed conveyor belt and the unit area mass of the pelleted feed, calculates the target flow rate that each metering pump needs to output in the current feeding cycle, and generates a corresponding pulse width modulation signal. By adjusting the duty cycle, the speed of the stepper motor is precisely controlled so that the error between the actual amount of each nutrient added and the target flow rate does not exceed ±1%. The pneumatic ultrasonic atomizing spray head uses ultrasonic vibration to atomize the functional nutrient solution output by the metering pump into micro-mist particles with an average particle size of 20 micrometers, which are then uniformly sprayed onto the surface of the pelleted feed on the conveyor belt. The eight sets of spray heads are evenly distributed along the width of the feed conveyor belt, and the spray coverage area of ​​adjacent spray heads has an overlap of no less than 10% to ensure that the uniformity deviation of the spraying is kept within ±2%.

[0013] Furthermore, the process of secondarily correcting the initial nutritional compensation vector based on the magnitude and rate of change of body surface temperature deviation specifically includes: After feeding is completed, the control system starts the feedback tracking timer and continuously reads the average body surface temperature of the poultry cluster output by the infrared thermal imaging sensor array within a preset tracking window of 15 to 30 minutes, as a direct physiological feedback indicator to judge the degree of stress relief. When the average body surface temperature remains outside the preset normal threshold range within the tracking window, it is determined that the compensation amount of the current compensation vector is insufficient, and a second-order gain correction process is triggered. The second-order gain correction process executes the following correction algorithm to generate the corrected nutrient compensation vector. : ; in, This is the initial nutrient compensation vector before correction. This is the difference between the measured average body surface temperature of the poultry flock at the current moment and the median of the target temperature threshold range. for The first derivative with respect to time reflects the rate of change of body surface temperature deviation. It is the first-order proportional gain coefficient. The second-order differential gain coefficient; the corrected nutrient compensation vector The original vector is replaced, control commands are regenerated, and the precision feeding device is driven to perform a second spraying after the compensation amount is adjusted, until the average body surface temperature returns to the normal threshold range, thus completing the closed-loop dynamic matching.

[0014] Furthermore, the first-order proportional gain coefficient and second-order differential gain coefficient The initial values ​​are preset based on the poultry breed and current age, and after each complete breeding batch, they are based on the actual stress relief response data accumulated in that batch, i.e., the actual... and The degree of contribution to the compensation effect, to and Automatic calibration and updates are performed to gradually optimize and adapt to the specific physiological response characteristics of different breeds and ages of poultry.

[0015] A knowledge graph-based system for regulating the nutritional composition of poultry eggs includes: The multidimensional feature space construction module is used to perform sensor data acquisition, noise reduction, and time alignment processing to construct a multidimensional heterogeneous feature space. The knowledge graph association mapping and nutrient gap quantification module stores and runs a pre-built environmental, physiological and nutritional knowledge graph containing five types of entity nodes: environmental induction, physiological regulation, enzyme activity, protein transport and nutrient deposition. It activates stress nodes using a subgraph matching algorithm and performs weight propagation calculations along the causal path through the inference engine to quantify the interference intensity and generate an initial nutrient compensation vector. The closed-loop precision control module includes a control system, a precision feeding device, and a feedback correction unit. The control system generates instructions based on the initial nutrient compensation vector and drives the precision feeding device to add functional nutrients in real time in the form of micro-mist spraying. The feedback correction unit is used to continuously monitor the body surface temperature of the poultry cluster through an infrared thermal imaging sensor after feeding. The control system performs second-order gain correction on the initial nutrient compensation vector based on the magnitude and rate of change of the body surface temperature deviation, generates a corrected nutrient compensation vector, and drives the precision feeding device again to form a closed-loop dynamic matching.

[0016] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention, by constructing a multi-dimensional feature space and an environmental, physiological, and nutritional knowledge graph, transforms traditional post-hoc adjustments based on human experience into predictive compensation based on real-time data. The system can, in the early stages of physiological stress in poultry before they exhibit significant declines in production performance, activate and predict metabolic risks through the knowledge graph pathways and immediately initiate nutritional mitigation strategies, minimizing the impact of environmental fluctuations on the organism's homeostasis.

[0017] 2. This invention can deeply analyze the complex nonlinear relationships between environmental stimuli, physiological phenotypes, and nutrient metabolism. By quantifying the specific impact pathways of the environment on the deposition of specific nutrients through weighted edge quantification, nutrient compensation is no longer limited to empirical adjustments but can accurately calculate compensation vectors for specific amino acids or trace elements. Experimental data show that this method can significantly reduce the amount of nutrient additives used, lowering production costs while avoiding resource waste and burden on poultry bodies caused by excessive additives.

[0018] 3. This invention can achieve real-time dynamic compensation of feed nutrient levels in response to sudden environmental challenges such as high temperature and cold stress. This precise matching mechanism ensures that poultry can maintain high metabolic efficiency under different environmental pressures, resulting in a high degree of consistency in the core nutritional indicators of the produced eggs, effectively solving the problem of significant fluctuations in egg quality with seasons and environments in existing technologies.

[0019] 4. This invention combines deterministic knowledge graph logic with self-learning reinforcement learning algorithms, enabling the system not only to handle known stress patterns but also to continuously optimize the weight parameters in the knowledge graph through long-term operation, adapting to the physiological differences of different breeds and ages of poultry. The fusion analysis of multi-source heterogeneous data eliminates the risk of misjudgment caused by the failure of a single sensor, ensuring that the system can continuously provide stable and reliable control decision support in complex intensive farming environments. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall technical solution for regulating the nutritional components of poultry eggs using knowledge graphs. Figure 2 This is a schematic diagram illustrating the core principles of association mapping and reasoning based on an environmental physiological nutrition knowledge graph. Figure 3 It is a logical flowchart for constructing a multidimensional heterogeneous feature space and aligning it with multi-source data; Figure 4 It is a schematic diagram of the multi-level interaction relationship and data flow between environmental characteristics, physiological characteristics and nutrient metabolism nodes; Figure 5 It is a logical flowchart of a closed-loop precision regulation and feedback adjustment mechanism based on real-time nutrient compensation vectors. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 5 The present invention will be further described in detail below with reference to specific embodiments.

[0022] The first aspect is the method for regulating the nutritional components of poultry eggs using knowledge graphs disclosed in this application, which is implemented according to the following steps: First, in step S1, the multidimensional feature space is constructed: environmental feature data and poultry physiological feature data are collected in real time through a sensor array deployed in the poultry house. The environmental feature data includes temperature, humidity, ammonia concentration, and hydrogen sulfide concentration, while the physiological feature data includes the poultry body surface temperature distribution and activity acoustic spectrum characteristics. Then, the two types of data are aligned uniformly on the time axis using a clock synchronization mechanism, and after noise reduction and multi-frequency normalization processing, a complete multidimensional heterogeneous feature space is constructed, specifically through the following implementation steps.

[0023] Step S101, deployment of three-dimensional sensor array: Environmental sensors include temperature sensors, humidity sensors, ammonia concentration sensors and hydrogen sulfide concentration sensors, which are arranged in three layers according to the longitudinal height of the poultry house, namely the bottom layer 0.5 meters above the ground, the middle layer 1.5 meters above the ground and the top layer 3 meters above the ground, with one set of monitoring points set up for every 200 square meters.

[0024] The temperature sensor has a measurement accuracy of 0.1 degrees Celsius, the humidity sensor's measurement error does not exceed ±3% of relative humidity, and the detection limits for ammonia and hydrogen sulfide sensors are uniformly set to 0.1 ppm. The sampling period for all the above environmental sensors is uniformly set to 1 second. The collected data is uploaded to the control system via RS-485 bus to complete one cycle of environmental characteristic data acquisition and recording.

[0025] Step S102, Physiological characteristic collection of poultry: Physiological characteristic collection employs two types of sensors working in tandem. The first type is an infrared thermal imaging sensor array with a thermal sensitivity of no more than 0.05 degrees Celsius and an image resolution of no less than 320×240 pixels. It is installed on the top of the poultry house at a density of one monitoring area per 100 square meters, with the lens pointing vertically downwards, continuously collecting images of the body surface temperature distribution of the poultry cluster at a frame rate of 25 frames per second.

[0026] After non-uniformity correction, each frame of image outputs the calibrated temperature value of each pixel. The non-uniformity correction adopts a two-point correction method, that is, using two blackbody reference sources at high temperature and low temperature to linearly calibrate the response of each pixel of the detector to eliminate the sensitivity difference between pixels.

[0027] After calibration, the average temperature of three heat dissipation sensitive areas—comb, eye socket, and foot—is extracted, and the arithmetic mean of the three temperature averages is taken as the average body surface temperature output value of that area at the current moment.

[0028] The second type is an acoustic sensor cluster with a frequency response range of 20Hz to 20kHz and a signal-to-noise ratio of not less than 70dB. It is installed on the top of the poultry house at a density of one sensor per 300 square meters, and continuously collects the feeding sounds, calls and activity spectrum signals of poultry at a sampling frequency of 44.1kHz and a quantization accuracy of 16 bits.

[0029] Step S103, Acoustic Spectrum Feature Extraction: For the audio signal collected by the acoustic sensor in step S102, with an analysis frame length of 256 milliseconds and a frame shift of 128 milliseconds, a Hanning window is applied to each frame of signal, and then 7-level discrete wavelet decomposition is performed using the db4 wavelet basis function.

[0030] At a sampling frequency of 44.1kHz, the frequency bands corresponding to the coefficients of each layer after the 7-layer decomposition narrow from the higher to the lower layers. The 6th layer corresponds to approximately 344Hz to 688Hz, the 5th layer corresponds to approximately 688Hz to 1375Hz, and the 4th layer corresponds to approximately 1375Hz to 2750Hz. After merging the three layers of frequency bands, they cover approximately 344Hz to 2750Hz, which approximately corresponds to the target extraction range of 500Hz to 3000Hz.

[0031] The system selects wavelet coefficients from layers 4, 5, and 6, and calculates the normalized energy value for each layer. This is done by summing the squares of all wavelet coefficients in that layer and dividing by the total number of coefficients. The normalized energy values ​​from layers 6, 5, and 4 are then concatenated in ascending order of frequency band to form a 3D acoustic spectrum energy feature vector. This vector reflects the energy changes in feeding rhythm and abnormal stress screams in poultry, and together with the average body surface temperature data output in step S102, it constitutes a poultry physiological characteristic dataset.

[0032] Step S104, Sensor signal noise reduction processing: For the raw signals of each environmental sensor channel acquired in step S101, a 5th-order Kalman filter algorithm is used for real-time noise reduction. The filter constructs a linear discrete state-space model with the actual physical quantities of the sensor as state variables. At each sampling time, the prediction step and the update step are executed alternately to gradually suppress random interference noise, and the posterior state estimate of the current time is output as the denoised sensor reading.

[0033] The process noise covariance matrix is ​​calibrated offline based on the signal variance of each sensor channel in a static environment for one hour of historical data. The observation noise covariance matrix is ​​initialized based on the accuracy specifications of the sensor manufacturer and is automatically updated every 24 hours based on newly acquired data.

[0034] Step S105, Multi-frequency sensor data time alignment: In step S101, the sampling frequency of the environmental sensor is 1Hz. In step S102, the output frame rate of the infrared thermal imaging sensor is 25Hz. In step S103, the output frequency of the acoustic spectrum feature is about 7.8Hz. This frequency is calculated by analyzing the frame shift of 128 milliseconds, that is, one frame of acoustic spectrum feature is output every 128 milliseconds, corresponding to an output frequency of about 7.8Hz, which is close to the target alignment frequency of 10Hz. The interpolation processing can be performed normally.

[0035] The sampling frequencies of the environmental sensor, infrared thermal imaging sensor, and acoustic sensor differ significantly. The system uses a cubic spline interpolation algorithm with a target time interval of 100 milliseconds to resample the data from each channel, unifying all channels to a unified time axis of 10 aligned time points per second. Cubic spline interpolation fits a third-order polynomial with continuous first and second derivatives between adjacent original sampling points, with natural boundary conditions added at the endpoints to ensure the smoothness of the interpolation curve.

[0036] After alignment, each 100-millisecond time point corresponds to a complete set of multi-channel observation data, including timestamp, spatial coordinates, noise-reduced temperature value, humidity value, ammonia concentration value, hydrogen sulfide concentration value, average body surface temperature, and acoustic spectrum energy feature vector. The data of all time points are arranged in chronological order, together forming a structured multidimensional heterogeneous feature space, which can be directly called by the knowledge graph association mapping in step S2.

[0037] In summary, step S1 completes the collection, feature extraction, noise reduction, and time alignment of raw data from two dimensions: poultry house environment and poultry physiology. This ultimately generates a structured, multidimensional, heterogeneous feature space updated every 100 milliseconds. This feature space synchronously records various heterogeneous data at each time point, including environmental temperature and humidity, harmful gas concentrations, poultry body surface temperature, and acoustic spectrum energy distribution. This eliminates interference from random noise from sensors and resolves the time asynchrony issue between sensors with different sampling frequencies. This provides a time-aligned, dimensionally complete, and reliable data foundation for the association mapping and subgraph matching of the environmental, physiological, and nutritional knowledge graph in subsequent step S2.

[0038] The next step is step S2, knowledge graph association mapping: real-time data from the multidimensional heterogeneous feature space constructed in step S1 is input into the pre-constructed environmental physiological nutrition knowledge graph. The mapping position of the current production status in the knowledge graph is identified through the subgraph matching algorithm, the corresponding stress node is activated, and the subsequent path retrieval is started from the activated node, providing the graph location result for the nutrition gap reasoning in step S3. The specific implementation steps are as follows.

[0039] Step S201, Construction of the Environmental Physiology and Nutrition Knowledge Graph: The environmental physiology and nutrition knowledge graph is the core knowledge carrier of this method, and its construction is completed offline before system deployment. The graph as a whole is a deep semantic network containing more than 10,000 entity nodes and 50,000 logical relationship edges, with data sources covering three aspects.

[0040] The first source is historical breeding records, specifically including poultry house environmental monitoring data, daily feed records, poultry physiological characteristic data, and poultry egg quality test reports. After being cleaned, deduplicated, and standardized, these records serve as the statistical basis for the causal relationships between nodes in the graph.

[0041] The second source is the biochemical metabolic pathway database. The system accesses metabolic pathway data related to avian lipid metabolism, amino acid metabolism, mineral transport and antioxidant defense in the KEGG database, and maps the enzymes, substrates, products and regulatory factors in the pathways to corresponding nodes and relationship edges in the map.

[0042] The third source is the Poultry Nutrition Expert Knowledge Base, which is entered by nutrition experts with more than 10 years of breeding experience in a structured question-and-answer format. The content covers the compensation priority and dosage range of various nutrients under different stress types, serving as the source of expert rule edges in the graph.

[0043] Step S202, Definition of map node types: The entity nodes in the map are divided into 5 categories according to the biological hierarchy. The first category is environmentally induced nodes, which correspond to the abnormal environmental conditions that can be directly measured in the poultry house, including high temperature nodes, low temperature nodes, high ammonia nodes, high hydrogen sulfide nodes, and high humidity nodes. The activation threshold of each node is set according to the sensor acquisition parameters in step S101. The activation condition for the high temperature node is that the poultry house temperature exceeds 32 degrees Celsius, the activation condition for the high ammonia node is that the ammonia concentration exceeds 20 ppm, and the activation condition for the temperature fluctuation node is that the temperature changes by more than 2 degrees Celsius within 1 hour.

[0044] The second category is physiological regulation nodes, which correspond to the physiological responses of avian organisms under environmental stimuli. These include heart rate change nodes, body temperature regulation nodes, and respiratory rate nodes. Their states are determined by the average body surface temperature data collected by the infrared thermal imaging sensor in step S102 and the acoustic spectrum energy feature vector in step S103.

[0045] The third category consists of enzyme activity nodes, which correspond to enzymes involved in key steps of nutrient metabolism, including carbonic anhydrase nodes and alkaline phosphatase nodes, which play regulatory roles in calcium metabolism and phosphorus metabolism pathways, respectively.

[0046] The fourth category consists of protein transport nodes, which correspond to functional proteins involved in nutrient transport, including the calcium-binding protein D28k node and the vitellogenin transporter node. The former is responsible for intestinal calcium absorption, while the latter is responsible for the transport and deposition of vitellogenin.

[0047] The fifth category is nutrient deposition nodes, which correspond to the final poultry egg quality indicators. These include eggshell calcification nodes, yolk lecithin deposition nodes, unsaturated fatty acid deposition nodes, and trace element deposition nodes. These are the target endpoints for the nutrient compensation vector calculation in step S3.

[0048] Step S203, assigning weights to logical relationship edges: Each logical relationship edge in the graph represents the strength of the causal influence between two nodes. The weight value is set to a real number between 0 and 1, with a larger weight value indicating a stronger causal influence. The initial weights are determined jointly by the three types of data sources in step S201, as follows.

[0049] For historical aquaculture records, the Spearman rank correlation coefficient is used to calculate the statistical correlation strength between the corresponding indicators of any two nodes. The absolute value of the correlation coefficient is normalized to the interval between 0 and 1 and then used as the initial statistical weight of the relationship edge.

[0050] For metabolic pathway data in the KEGG database, the relevant relationship edges are weighted according to the ratio of the catalytic efficiency constant to the substrate affinity constant of the enzymatic reaction, with pathways having higher catalytic efficiency corresponding to greater weights. For the expert knowledge base, experts evaluate the influence strength of each relationship edge using integer scores from 0 to 10, and the scores are linearly mapped to the 0-1 interval and used as the expert rule weights.

[0051] The weights from the three sources are combined using a weighted average, and the combination formula is as follows: ; in, For nodes With nodes The overall weight of the edges relating to each other; These are statistical weights calculated based on historical aquaculture data. The biochemical weights are calculated based on metabolic pathway data from the KEGG database. These are the expert weights obtained based on the expert knowledge base scoring mapping; , , The fusion coefficients for the three weight classes respectively satisfy the following conditions: The initial value is set based on the reliability of the data source. , , And after the system is running, it will be continuously updated by the self-learning mechanism in step S2.

[0052] Step S204, Real-time Observation Vector Construction: During system operation, at every 100-millisecond time point, the system reads the complete data frame of the current moment from the multi-dimensional heterogeneous feature space generated in step S1, and concatenates the temperature value, humidity value, ammonia concentration value, hydrogen sulfide concentration value, average body surface temperature, and 3D acoustic spectrum energy feature vector in a fixed order to form a 7-dimensional real-time observation vector, denoted as . ,in Each component corresponds to the current sampled value of the above 7 feature channels. The dimensions are uniformly normalized to the maximum and minimum values ​​before splicing, and the values ​​of each channel are linearly mapped to the interval between 0 and 1 to eliminate the influence of the difference in the dimensions of different physical quantities on the subsequent distance calculation.

[0053] Step S205, Subgraph Matching and Stress Node Activation: The subgraph matching algorithm uses the real-time observation vector constructed in step S204. As input, the system retrieves the predefined feature subgraph that most closely matches the current production state from the knowledge graph. The graph pre-stores feature subgraphs covering various typical stress scenarios. Each feature subgraph consists of several nodes and their relational edges, and corresponds to a reference feature vector with the same dimension as the real-time observation vector. subscript Indicates the first Numbering of each predefined feature submap.

[0054] The matching process calculates the real-time observation vector. With each reference eigenvector Euclidean distance between To perform similarity measurement, the calculation formula is: ; in, For real-time observation vector and the first The smaller the Euclidean distance between each reference feature vector, the higher the degree of matching between the current production state and the feature sub-map. For real-time observation vectors The One component; For the first reference feature vectors The One component; The total dimension of the observation vector in this system. The value is 7; the summation symbol indicates summation over all The squares of the differences between the components in each dimension are summed one by one and the square root is taken.

[0055] The system selects the feature subgraph with the smallest distance as the mapping result for the current production state, that is: ; in, Number the feature sub-image with the smallest distance. Indicates the number of all candidate subgraphs. envoy The smallest number value.

[0056] After determining the optimal matching subgraph, the system activates the graph nodes according to preset node activation rules. These rules are divided into two categories: Category 1 is a threshold activation rule, where the corresponding node is directly activated when the value of a specific component in the real-time observation vector exceeds a preset threshold. For example, an ammonia poisoning warning node is activated when the ammonia concentration exceeds 20 ppm, and a temperature fluctuation stress node is activated when the temperature component changes by more than 2 degrees Celsius cumulatively within one hour. Category 2 is a subgraph matching activation rule, where the optimal matching subgraph... corresponding distance Below the preset matching threshold At that time, all environment-induced nodes corresponding to that subgraph are activated. The initial value is set to 0.15, and it is automatically calibrated quarterly as system operating data accumulates.

[0057] The activated node serves as the starting point for subsequent path retrieval and is passed to the inference engine in step S3 to calculate the interference intensity of environmental fluctuations on the nutrient metabolism pathway. The real-time observation vector constructed in step S205, after activating nodes through subgraph matching, inherits its data logic from the subsequent step S3. The output vector finally generated in step S3 is the initial nutrient compensation vector described in claim 1.

[0058] In summary, step S2 completes the entire process from knowledge graph construction, node type definition, relation edge weight assignment, to real-time observation vector generation and subgraph matching activation. The multi-dimensional sensor data collected in step S1 is transformed into a set of activated nodes with clear semantics in the graph, enabling the inference engine in the subsequent step S3 to perform directional weight propagation calculations along the activation path, thereby outputting a real-time compensation scheme for specific nutrient deposition targets.

[0059] Then, in step S3, nutrient gap reasoning: the reasoning engine takes the stress node activated in step S2 as the starting point, performs weight propagation calculation along the causal path in the knowledge graph, quantifies the interference intensity of environmental fluctuations on various metabolic pathways of poultry, and, combined with the preset nutrient deposition target, outputs a real-time nutrient compensation vector containing the compensation amount of multiple types of nutrients, which is then directly executed by the precision feeding device in step S4.

[0060] Step S301, Depth-first search of activation path: The inference engine takes the set of activation nodes determined in step S205 as the starting point and performs depth-first path search in the knowledge graph, extending downstream along each causal relationship edge until it reaches the nutrient deposition node. The maximum search depth of the path is set to 10 hops.

[0061] For different types of nutrient deposition targets, the engine dynamically adjusts its search strategy. When the target is the deposition of macronutrients, such as increasing the content of unsaturated fatty acids or lecithin, the engine prioritizes expanding the main pathways involving protein metabolism nodes and lipid transport nodes. When the target is the deposition of trace components, such as improving the deposition efficiency of organic selenium or chelated zinc, the engine prioritizes expanding the micropaths involving enzyme activity nodes and ion channel regulation nodes.

[0062] Taking a high-temperature stress scenario as an example, the typical activation path is as follows: the high-temperature node points to the respiratory rate node, the respiratory rate node points to the respiratory alkalosis node, the respiratory alkalosis node points to the plasma bicarbonate reduction node, the plasma bicarbonate reduction node points to the calcium transporter activity decrease node, and finally reaches the eggshell calcification node. This path passes through a total of 5 relation edges.

[0063] Step S302, Negative Inhibition Coefficient Calculation: For each activated path retrieved in step S301, the inference engine calculates the negative inhibition coefficient of that path on downstream nutrient deposition nodes. This coefficient reflects the overall inhibition strength of environmental stress on the target nutrient deposition process after transmission through that path. The calculation formula is as follows: ; in, This is the negative inhibition coefficient for the activation path, a real number ranging from 0 to 1. The smaller the value, the stronger the inhibition. For the first on this path The weight values ​​of each relation edge are determined by the comprehensive weight in step S203. Sure; This represents the total number of relation edges traversed by the path. Indicates all The weights of the edges in the relation are multiplied together.

[0064] When the same nutrient deposition node is simultaneously pointed to by multiple activation paths, the minimum value among the negative inhibition coefficients of each path is taken as the comprehensive negative inhibition coefficient of that node, in order to reflect the dominant effect of the strongest inhibition path when multiple stresses are superimposed.

[0065] Step S303, Calculation of Nutrient Compensation: Based on the comprehensive negative inhibition coefficient of each nutrient deposition node, the inference engine calculates the percentage concentration of each nutrient that needs to be supplemented in addition to the basal diet. The calculation formula is as follows: ; in, This represents the percentage concentration of this nutrient that needs to be supplemented under the current stress state, calculated with an accuracy of 0.001%. The baseline compensation level for this nutrient under non-stress conditions is determined by the preset benchmark values ​​for different nutrients in the expert knowledge base. The comprehensive negative inhibition coefficient of the nutrient deposition node calculated in step S302; To compensate for the amplification factor, the stronger the suppression, the larger the amplification factor, and the higher the corresponding compensation amount.

[0066] The above calculations cover four major categories: vitamins, amino acids, electrolytes, and trace elements, specifically including vitamin A, vitamin C, vitamin E, methionine, lysine, sodium bicarbonate, potassium chloride, organic selenium, and chelated zinc.

[0067] Step S304, Generation of real-time nutrient compensation vector: The compensation amounts of all nutrients in step S303 are concatenated in a fixed order to form a 9-dimensional real-time nutrient compensation vector. : ; in, , , These are the percentage concentrations of additional vitamin A, vitamin C, and vitamin E, respectively. , These are the additional percentage concentrations of methionine and lysine, respectively. , These are the additional percentage concentrations of sodium bicarbonate and potassium chloride, respectively. , The percentage concentrations are for additional organic selenium and chelated zinc, respectively; all components are in percentage form and the calculation accuracy is 0.001%.

[0068] vector After generation, the system synchronously accesses the aquaculture cost database and calculates the cost increment of this control plan based on the purchase price of each nutrient component and the corresponding compensation amount. If multiple combinations of plans can achieve the same nutrient deposition target, the system prioritizes the plan with the lowest total cost increment and updates the optimized compensation amount to the vector. The corresponding components are used to output the final real-time nutrient compensation vector, which is then passed to step S4 for execution.

[0069] In summary, after step S3, the inference engine transforms the graph activation nodes output in step S2 into a 9-dimensional real-time nutrient compensation vector that can directly guide the feeding execution. This enables a quantitative mapping from environmental stress signals to specific nutrient compensation amounts, while simultaneously optimizing both nutrient deposition effects and economic costs, providing a clear execution basis for the closed-loop precise control of the precision feeding device in step S4.

[0070] Finally, in step S4, closed-loop precise control: the control system uses the real-time nutrient compensation vector output in step S3. The system generates control commands to drive the precision feeding device to add functional nutrients to the basal diet in real time. It also continuously monitors the stress relief status of poultry through a feedback adjustment mechanism, and performs secondary correction on the compensation vector to achieve dynamic matching between nutrient supply and instantaneous environmental stress status.

[0071] Step S401: Control commands for the precision feeding device are generated. The precision feeding device consists of four micro-metering pumps driven by stepper motors and eight pneumatic ultrasonic atomizing spray heads. The four metering pumps correspond to four independent storage tanks for liquid nutrients. Each storage tank has a double-layer stainless steel structure, filled with polyurethane foam to maintain a constant internal temperature, and coated with a light-shielding paint to protect the photosensitive nutrients. A pressure-type liquid level sensor is installed at the bottom of each storage tank. When the remaining amount in the tank falls below 10% of the total capacity, the system automatically sends a replenishment reminder to the management terminal via an IoT protocol.

[0072] The control system reads the real-time nutrient compensation vector output in step S3. The values ​​of each component are combined with the current operating speed of the feed conveyor belt and the unit area mass of the pelleted feed to calculate the target flow rate that each metering pump needs to output in the current feeding cycle, in milliliters per second.

[0073] The control system generates a corresponding pulse width modulation signal based on this, and precisely controls the speed of the stepper motor by adjusting the duty cycle of the pulse width modulation signal, thereby controlling the output flow of the metering pump so that the error between the actual amount of each nutrient added and the target flow does not exceed ±1%.

[0074] Step S402, atomization spraying of functional nutrients: The functional nutrient solution output by the metering pump is delivered to 8 sets of pneumatic ultrasonic atomizing spray heads through the conduit. The atomizing spray heads use ultrasonic vibration to atomize the solution into micro-mist particles with an average particle size of 20 micrometers, which are then evenly sprayed onto the surface of the pellet feed on the conveyor belt.

[0075] Eight sets of spray heads are evenly distributed along the width of the feed conveyor belt, with at least 10% overlap between the spray coverage areas of adjacent spray heads to ensure uniform nutrient adhesion to all parts of the feed surface. The system requires the spray uniformity deviation to be maintained within ±2%, which is verified by sampling and weighing at the end of the conveyor belt every 30 minutes.

[0076] Step S403, continuous tracking of stress relief indicators: After the functional nutrients are sprayed, the control system starts the feedback tracking timer. Within 15 to 30 minutes after the feeding is completed, it continuously reads the average body surface temperature of the poultry cluster output by the infrared thermal imaging sensor array in step S102, and the real-time update value of the acoustic spectrum energy feature vector in step S103, as two types of feedback indicators to judge the degree of stress relief.

[0077] The system presets normal body surface temperature threshold ranges for various stress scenarios. Taking a high-temperature stress scenario as an example, the normal threshold range for the average body surface temperature of poultry flocks is set to 38.5 degrees Celsius to 40.5 degrees Celsius. If the average body surface temperature remains outside the threshold range within the tracking window, the current compensation vector is determined to be outside the threshold range. The compensation amount is insufficient, triggering the second-order gain correction process in step S404.

[0078] Step S404, Second-order gain correction of the compensation vector: When step S403 determines that the compensation amount is insufficient, the control system performs second-order gain correction on the real-time nutrient compensation vector. The correction formula is as follows: ; in, The corrected nutrient compensation vector has the following dimensions: The same, it is a 9-dimensional vector, and the units of each component are percentages; This is the initial real-time nutrient compensation vector output in step S3, i.e., the original vector without correction. This is a first-order proportional gain coefficient, reflecting the contribution weight of the current magnitude of the body surface temperature deviation to the compensation correction. The initial value is preset according to the breed and age, and the typical value range is 0.01 to 0.05, with the unit being percentage per degree Celsius. This is the second-order differential gain coefficient, reflecting the contribution weight of the rate of change of body surface temperature deviation to the compensation correction. The typical initial value ranges from 0.005 to 0.02, and the unit is percentage per degree Celsius per minute. This is the difference between the measured average body surface temperature of the poultry flock at the current moment and the median of the target temperature threshold range, expressed in degrees Celsius. When the measured value is higher than the median... Take a positive value; take a negative value if the value is below the median. for The first derivative with respect to time reflects the rate of change of body surface temperature deviation, measured in degrees Celsius per minute. This is achieved by analyzing three consecutive tracking time points. The slope is obtained by performing a linear fit on the value.

[0079] Corrected vector Replace the original vector Then, step S401 is re-entered to generate a new round of control commands, driving the precision feeding device to perform a second spraying after the compensation amount adjustment, until the average body surface temperature of the poultry cluster returns to the normal threshold range.

[0080] and The value is automatically calibrated and updated based on the actual stress relief response data of each breeding batch after the end of each breeding batch, gradually adapting to the differences in physiological response characteristics of different breeds and different ages of poultry.

[0081] In summary, step S4 achieves a complete closed-loop control process, from compensation vector generation to physical spraying, stress monitoring, and secondary correction. This closed-loop mechanism enables the system to autonomously and accurately respond to the specific physiological feedback of individual poultry flocks without human intervention. Together with steps S1 to S3, it constitutes an end-to-end automatic control scheme from environmental perception to nutritional execution.

[0082] On the other hand, the poultry egg nutritional composition regulation system using knowledge graphs disclosed in this application includes: The multidimensional feature space construction module is used to perform sensor data acquisition, noise reduction, and time alignment processing to construct a multidimensional heterogeneous feature space. The knowledge graph association mapping and nutrient gap quantification module stores and runs a pre-built environmental, physiological and nutritional knowledge graph containing five types of entity nodes: environmental induction, physiological regulation, enzyme activity, protein transport and nutrient deposition. It activates stress nodes using a subgraph matching algorithm and performs weight propagation calculations along the causal path through the inference engine to quantify the interference intensity and generate an initial nutrient compensation vector. The closed-loop precision control module includes a control system, a precision feeding device, and a feedback correction unit. The control system generates instructions based on the initial nutrient compensation vector, driving the precision feeding device to add functional nutrients in real time in the form of micro-mist spraying. The feedback correction unit is used to continuously monitor the body surface temperature of the poultry cluster through an infrared thermal imaging sensor after feeding. The control system performs second-order gain correction on the initial nutrient compensation vector based on the magnitude and rate of change of the body surface temperature deviation, generates a corrected nutrient compensation vector, and drives the precision feeding device again to form a closed-loop dynamic matching.

[0083] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0084] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for regulating the nutritional components of poultry eggs using knowledge graphs, characterized in that, Includes the following steps: Multidimensional feature space construction: Environmental feature data and physiological feature data including poultry body surface temperature and activity sound spectrum are collected in real time by a sensor array deployed in the poultry house. The above data with different sampling frequencies are time-axis aligned and noise-reduced to construct a multidimensional heterogeneous feature space. Knowledge graph association mapping and nutritional gap quantification: Real-time data from the multidimensional heterogeneous feature space is input into a pre-constructed environmental physiological nutrition knowledge graph that integrates metabolic pathway data. The mapping position of the current production state in the knowledge graph is identified by a subgraph matching algorithm to activate the corresponding stress node. Based on the activated stress node, the interference intensity of environmental fluctuations on each metabolic pathway is quantified along the causal path of the knowledge graph to calculate the initial nutritional compensation vector. Closed-loop precise control: The control system generates control commands based on the initial nutrient compensation vector to drive the precision feeding device to add functional nutrients to the basic diet in real time; after feeding is completed, the poultry body surface temperature is continuously monitored as a feedback indicator, and the initial nutrient compensation vector is corrected a second time based on the magnitude and rate of change of the body surface temperature deviation, generating a corrected nutrient compensation vector and driving the precision feeding device again to compensate until the body surface temperature returns to normal, thus realizing dynamic closed-loop matching between nutrient supply and instantaneous environmental stress state; Specifically, the process of quantifying the intensity of environmental fluctuations on each metabolic pathway based on activated stress nodes along the causal path of the knowledge graph to calculate the initial nutrient compensation vector includes: The inference engine starts from the activated stress node and performs a depth-first path search in the knowledge graph, extending downstream along each causal relationship edge until it reaches the nutrient deposition node. The maximum search depth of the path is set to 10 hops. For each activated path retrieved, the negative inhibition coefficient of that path on downstream nutrient deposition nodes is calculated. The negative inhibition coefficient is the product of the weights of all logical relation edges on that path, and its value ranges from 0 to 1. The smaller the value, the stronger the inhibition. When the same nutrient deposition node is pointed to by multiple activation paths at the same time, the minimum value among the negative inhibition coefficients of each path is taken as the comprehensive negative inhibition coefficient of that node. Based on the comprehensive negative inhibition coefficient, the percentage concentration of each type of nutrient that needs to be supplemented under the current stress state is calculated to form the initial nutrient compensation vector.

2. The method for regulating the nutritional components of poultry eggs using knowledge graphs according to claim 1, characterized in that, The environmental physiological nutrition knowledge graph is a deep semantic network containing five types of entity nodes and the logical relationship edges between them. Its construction process specifically includes: The knowledge graph is constructed offline in advance, and the five types of entity nodes are defined as follows: environmental triggering nodes, corresponding to measurable abnormal environmental states in poultry houses; physiological regulation nodes, corresponding to the physiological responses of poultry organisms under environmental stimuli; enzyme activity nodes, including carbonic anhydrase nodes involved in calcium metabolism regulation and alkaline phosphatase nodes involved in phosphorus metabolism regulation; protein transport nodes, including the calcium-binding protein D28k node responsible for intestinal calcium absorption and the vitellogenin transporter node responsible for vitellogenin transport and deposition; and nutrient deposition nodes, corresponding to the final poultry egg quality indicators. The initial weights of the logical relationship edges are calculated by weighted average of the statistical correlation of historical breeding records, the enzymatic reaction kinetic constants of the metabolic pathway data, and the scores of the poultry nutrition expert knowledge base, and are continuously updated through a self-learning mechanism after the system is running.

3. The method for regulating the nutritional components of poultry eggs using knowledge graphs according to claim 1, characterized in that, Physiological data specifically include the distribution of body surface temperature and the acoustic spectrum of activity in poultry; The process of constructing the multidimensional heterogeneous feature space specifically includes: An infrared thermal imaging sensor array is installed on the top of the poultry house at a density of one monitoring area per 100 square meters. The array continuously collects images of the body surface temperature distribution of poultry clusters and extracts the average temperature of three heat dissipation sensitive areas: the comb, eye socket, and feet. The arithmetic mean of the three temperature averages is used as the average body surface temperature output value at the current moment. An acoustic sensor cluster is used, with one sensor installed on the top of the poultry house at a density of one sensor per 300 square meters, to continuously collect the feeding sounds, calls, and activity spectrum signals of poultry. The acquired audio signals were analyzed with a frame length of 256 milliseconds and a frame shift of 128 milliseconds. After applying a Hanning window, a 7-layer discrete wavelet decomposition was performed. The wavelet coefficients of the 4th, 5th and 6th layers were selected, and the normalized energy values ​​of each layer were calculated. The coefficients were then spliced ​​together in order of frequency band from low to high to form a 3D acoustic spectrum energy feature vector to reflect the energy change characteristics of changes in the feeding rhythm of poultry and abnormal stress screams.

4. The method for regulating the nutritional components of poultry eggs using knowledge graphs according to claim 3, characterized in that, The process of time axis alignment and noise reduction for data with different sampling frequencies specifically includes: The original signals of each sensor channel of the environmental feature data are denoised in real time using a 5th-order Kalman filter algorithm, and the posterior state estimate is output as the denoised sensor reading. Using a target time interval of 100 milliseconds, a cubic spline interpolation algorithm is used to resample the noise-reduced environmental feature data, the average body surface temperature, and the acoustic spectrum energy feature vector. All channels are unified to a unified time axis with 10 aligned time points per second, ensuring that each time point corresponds to a complete set of multi-channel observation data to form a structured multidimensional heterogeneous feature space.

5. The method for regulating the nutritional components of poultry eggs using knowledge graphs according to claim 1, characterized in that, The initial nutrient compensation vector is constructed by splicing the calculated compensation percentage concentrations of various nutrients in a fixed order. Specifically, it is a 9-dimensional vector, whose components include the compensation percentage concentrations of vitamin A, vitamin C, vitamin E, methionine, lysine, sodium bicarbonate, potassium chloride, organic selenium, and chelated zinc in that order. After generating the initial nutrient compensation vector, the system synchronously accesses the aquaculture cost database, calculates the cost increment of the current control scheme based on the purchase price of various nutrients and the corresponding compensation amount, and if there are multiple combination schemes that can achieve the same nutrient deposition target, the scheme with the lowest total cost increment is selected first, and the optimized compensation amount is updated to the initial nutrient compensation vector and passed to subsequent steps for execution.

6. The method for regulating the nutritional components of poultry eggs using knowledge graphs according to claim 1, characterized in that, The precision feeding device in the closed-loop precise control step consists of a micro-metering pump driven by a stepper motor and a pneumatic ultrasonic atomizing spray head; the process of the control system generating control commands and adding materials specifically includes: The control system reads the values ​​of each component in the initial nutrient compensation vector, combines the current operating speed of the feed conveyor belt and the unit area mass of the pelleted feed, calculates the target flow rate that each metering pump needs to output in the current feeding cycle, and generates a corresponding pulse width modulation signal. By adjusting the duty cycle, the speed of the stepper motor is precisely controlled so that the error between the actual amount of each nutrient added and the target flow rate does not exceed ±1%. The pneumatic ultrasonic atomizing spray head uses ultrasonic vibration to atomize the functional nutrient solution output by the metering pump into micro-mist particles with an average particle size of 20 micrometers, which are then uniformly sprayed onto the surface of the pelleted feed on the conveyor belt. The eight sets of spray heads are evenly distributed along the width of the feed conveyor belt, and the spray coverage area of ​​adjacent spray heads has an overlap of no less than 10% to ensure that the uniformity deviation of the spraying is kept within ±2%.

7. The method for regulating the nutritional components of poultry eggs using knowledge graphs according to claim 6, characterized in that, The process of making a secondary correction to the initial nutritional compensation vector based on the magnitude and rate of change of body surface temperature deviation specifically includes: After feeding is completed, the control system starts the feedback tracking timer and continuously reads the average body surface temperature of the poultry cluster output by the infrared thermal imaging sensor array within a preset tracking window of 15 to 30 minutes, as a direct physiological feedback indicator to judge the degree of stress relief. When the average body surface temperature remains outside the preset normal threshold range within the tracking window, it is determined that the compensation amount of the current compensation vector is insufficient, and a second-order gain correction process is triggered. The second-order gain correction process executes the following correction algorithm to generate the corrected nutrient compensation vector; Wherein, is the initial nutrient compensation vector before correction, is the difference between the measured value of the average body surface temperature of the poultry cluster at the current moment and the median of the target temperature threshold range, is the first derivative with respect to time, reflecting the rate of change of the body surface temperature deviation, is the first-order proportional gain coefficient, and is the second-order differential gain coefficient; the corrected nutrient compensation vector replaces the original vector, and the control command is regenerated and the precision feeding device is driven to perform the re-spraying after the compensation amount adjustment, until the average body surface temperature returns to the normal threshold range, completing the closed-loop dynamic matching.

8. The method for regulating the nutritional components of poultry eggs using knowledge graphs according to claim 7, characterized in that, The initial values ​​of the first-order proportional gain coefficient and the second-order differential gain coefficient are preset based on the breed of poultry and the current age. After each complete breeding batch, based on the actual stress relief response data accumulated in that batch, i.e. the actual value and the degree of contribution to the compensation effect, the values ​​of the first-order proportional gain coefficient and the second-order differential gain coefficient are automatically calibrated and updated to gradually optimize and adapt to the specific physiological response characteristics of poultry of different breeds and ages.

9. A knowledge graph-based system for regulating the nutritional components of poultry eggs, characterized in that, The system, employing the method of any one of claims 1 to 8, comprises: The multidimensional feature space construction module is used to perform sensor data acquisition, noise reduction, and time alignment processing to construct a multidimensional heterogeneous feature space. The knowledge graph association mapping and nutrient gap quantification module stores and runs a pre-built environmental, physiological and nutritional knowledge graph containing five types of entity nodes: environmental induction, physiological regulation, enzyme activity, protein transport and nutrient deposition. It activates stress nodes using a subgraph matching algorithm and performs weight propagation calculations along the causal path through the inference engine to quantify the interference intensity and generate an initial nutrient compensation vector. The closed-loop precision control module includes a control system, a precision feeding device, and a feedback correction unit. The control system generates instructions based on the initial nutrient compensation vector and drives the precision feeding device to add functional nutrients in real time in the form of micro-mist spraying. The feedback correction unit is used to continuously monitor the body surface temperature of the poultry cluster through an infrared thermal imaging sensor after feeding. The control system performs second-order gain correction on the initial nutrient compensation vector based on the magnitude and rate of change of the body surface temperature deviation, generates a corrected nutrient compensation vector, and drives the precision feeding device again to form a closed-loop dynamic matching.

Images