An egg chicken breeding precise supply adjustment system based on big data cloud edge cooperation

Abstract

The present application relates to the technical field of laying hen breeding, and discloses a precise feed supply adjustment system for laying hen breeding based on big data cloud edge collaboration. The system comprises an environment data acquisition module, a cloud edge collaborative analysis module, a multi-dimensional difference evaluation module, an abnormal area positioning module, a feed supply strategy adjustment module and an encrypted communication link module. The environment data acquisition module acquires multi-source data streams of the chicken population through various sensors and generates standard chicken population data after processing; the cloud edge collaborative analysis module constructs a dynamic feed supply model based on the data and outputs the theoretical feed demand; the multi-dimensional difference evaluation module compares the theoretical value with the actual consumption value in multiple dimensions to generate a feed difference coefficient matrix; the abnormal area positioning module generates an abnormal feed probability distribution map in combination with the topological information of the breeding farm; the feed supply strategy adjustment module generates an adaptive feed control signal based on the map; and the encrypted communication link module transmits the control signal to the actuator through a quantum encryption channel. The system realizes precise adjustment and safe control of feed supply for laying hen breeding.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of laying hen breeding, in particular to a precise feed supply adjustment system for laying hen breeding based on big data cloud edge collaboration. BACKGROUND

[0002] In the industry of large-scale breeding of laying hens, the precision of feed supply is one of the key factors that determine the success of breeding, which not only relates to the economic benefits of breeding, but also closely relates to the overall health of the chicken population, egg quality and sustainable development. Under the traditional mode of laying hen breeding, the feed supply adjustment work has long relied on the personal experience of the breeding personnel. Breeders need to observe the activity state of the chicken population, the shape of the excrement, the luster of the feathers, and regularly check the amount of remaining feed in the trough, etc. to adjust the amount and frequency of feed supply by subjective judgment. This artificial method has many insurmountable drawbacks. On the one hand, the frequency and degree of detail of manual observation are limited, and it is difficult to capture the real-time subtle changes of the chicken population. When the chicken population experiences fluctuations in feed demand due to environmental changes or health problems, it is often impossible to make timely adjustments, which may lead to some chickens growing slowly due to insufficient feed or causing digestive system diseases due to excessive feed. On the other hand, large-scale breeding farms are usually divided into multiple breeding areas, and the environmental factors such as temperature, humidity, light intensity and ventilation conditions are different in different areas. At the same time, the age, breed and health status of the chicken population are also different in different areas. These factors together cause significant differences in feed demand in different areas, and the traditional unified feed supply mode cannot meet the individual needs, often resulting in feed accumulation and waste in some areas and feed shortage in other areas.

[0003] With the application and popularization of information technology in the agricultural field, the laying hen breeding industry has gradually transformed towards intelligence. Some breeding farms have begun to introduce Internet of Things technology, deploying temperature sensors, humidity sensors, light sensors, feed weight sensors and chicken activity monitoring equipment to collect breeding environment parameters and chicken-related data. However, the current data processing methods still have obvious limitations. Most breeding farms use local computing terminals for data processing, which are limited by the computing power and storage capacity of the terminals and can only perform simple statistical analysis of the data. They cannot deeply mine the underlying correlation rules of the data, such as the dynamic relationship between environmental temperature and humidity changes and chicken feed consumption, and the potential link between chicken health index abnormalities and feed demand changes, etc. This results in a large amount of valuable data being idle and difficult to be converted into effective feed supply adjustment basis. At the same time, some breeding farms that attempt to interface with cloud platforms lack a perfect cloud edge collaboration mechanism, resulting in delays in data transmission between the cloud and the edge, data format incompatibility and other issues, which makes the big data analysis results of the cloud unable to be fed back to the feed control equipment at the edge in a timely manner, affecting the timeliness and accuracy of the feed supply adjustment.

[0004] In the process of formulating the feeding strategy, the prior art is mostly based on single-dimensional data for judgment, for example, only according to the remaining amount of feed to adjust the next feeding amount, or only according to the change of environmental temperature to modify the feeding parameters, ignoring the complex interaction between environment, chicken flock, feed and other factors, resulting in one-sidedness of the formulated feeding strategy, and large deviation from the actual demand of the chicken flock. At the same time, the safety of breeding data and feeding control signals in the transmission process has not been paid enough attention to, and the traditional encryption method is easy to be cracked, and the data and signals may be illegally intercepted, tampered or interfered, which not only leads to misoperation of the feeding system, causes economic loss, but also may leak the core breeding data of the farm, and brings potential risks to the enterprise. The existence of these problems makes it difficult to improve the feed utilization efficiency of the laying hen breeding, and the growth quality and egg production performance of the chicken flock are restricted, which seriously hinders the transformation of the laying hen breeding industry to the modern mode of high efficiency, precision and sustainability. SUMMARY

[0005] The purpose of the present application is to provide a laying hen breeding precise feeding regulation system based on big data cloud edge cooperation to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present application provides a laying hen breeding precise feeding regulation system based on big data cloud edge cooperation, which comprises:

[0007] An environmental data acquisition module is configured with a chicken flock health sensor, a feed consumption sensor and an environmental monitoring sensor, the chicken flock multi-source data stream acquired by the environmental data acquisition module is initially processed to generate standard chicken flock data;

[0008] A cloud edge cooperative analysis module is configured to construct a dynamic feeding model based on the standard chicken flock data and output a theoretical feed demand;

[0009] A multi-dimensional difference evaluation module is configured to compare and analyze the theoretical feed demand and the actual feed consumption value in multiple dimensions to generate a feed difference coefficient matrix;

[0010] An abnormal area positioning module is configured to generate an abnormal feeding probability distribution map according to the feed difference coefficient matrix combined with the topological information of the farm;

[0011] A feeding strategy regulation module is configured to generate an adaptive feeding control signal based on the abnormal feeding probability distribution map;

[0012] An encrypted communication link module is configured to transmit the adaptive feeding control signal to a feed distribution actuator through a quantum encryption channel.

[0013] Preferably, the environmental data acquisition module comprises:

[0014] Collecting chicken group weight data, feeding frequency data and egg production rate data through the chicken group health sensor;

[0015] Collecting temperature and humidity data and light intensity data through the environment monitoring sensor;

[0016] Noise filtering processing is performed on the collected chicken group multi-source data stream, and abnormal data points are removed according to a preset data quality threshold;

[0017] The filtered chicken group multi-source data stream is time series aligned, and the standard chicken group data is output.

[0018] Preferably, the cloud-edge collaborative analysis module comprises:

[0019] The dynamic feeding model is trained based on historical chicken group operation data, and the dynamic feeding model comprises a time series prediction unit and a feature compensation unit;

[0020] The standard chicken group data is obtained in real time through the edge computing node;

[0021] Incremental data batch processing algorithm is used to process the standard chicken group data in batches, and chicken group state features are extracted;

[0022] The chicken group state features are input into the dynamic feeding model, and the theoretical feed demand is output.

[0023] Preferably, the multi-dimensional difference evaluation module comprises:

[0024] The theoretical feed demand and the actual feed consumption value are calculated in time domain cumulative deviation to generate a time domain deviation vector;

[0025] The theoretical feed demand and the actual feed consumption value are detected in frequency domain energy offset to generate a frequency domain offset vector;

[0026] The sequence similarity of the theoretical feed demand and the actual feed consumption value is evaluated based on a sequence structure matching algorithm to generate a sequence similarity vector;

[0027] The time domain deviation vector, the frequency domain offset vector and the sequence similarity vector are fused to generate the feed difference coefficient matrix.

[0028] Preferably, the abnormal area positioning module comprises:

[0029] A nodal topological network is constructed according to the layout of the farm, and chicken coop location information and feed delivery path impedance parameters are labeled;

[0030] The feed difference coefficient matrix is mapped to the corresponding nodes of the nodal topological network;

[0031] The graph neural network is used to deduce an abnormal feed supply propagation path, and an abnormal attenuation factor is calculated based on a node impedance parameter;

[0032] The abnormal feed supply probability distribution map covering the whole farm is generated, and a high-probability abnormal area is identified.

[0033] Preferably, the feed strategy adjustment module comprises:

[0034] The feed checking parameters are configured according to the abnormal feed supply probability distribution map;

[0035] The high-frequency monitoring mode is enabled for the high-probability abnormal area;

[0036] The feed control rule is adjusted by using a swarm intelligence optimizer;

[0037] The adaptive feed control signal is generated based on the adjusted feed control rule.

[0038] Preferably, the feed strategy adjustment module further comprises:

[0039] The feed quantity adjustment value is calculated by using a temperature drop search algorithm;

[0040] The feed quantity adjustment value is converted into the adaptive feed control signal;

[0041] Based on the preset feed constraint condition, the feed quantity distribution is optimized by using a gradient projection adjuster.

[0042] Preferably, the gradient projection adjuster comprises:

[0043] A set of feed quantity constraint conditions is set;

[0044] The target function sub-gradient is calculated according to the current feed scheme;

[0045] The feed scheme is updated by a projection operation to meet all constraint conditions;

[0046] The optimized feed scheme is output to the adaptive feed control signal.

[0047] Preferably, the system further comprises:

[0048] The feedback detection module monitors the adjusted feed consumption data by using a recursive state estimator;

[0049] The monitoring data is fed back to the cloud-edge collaborative analysis module and the feed strategy adjustment module;

[0050] The feed strategy is updated based on the feedback data.

[0051] Preferably, the feedback detection module comprises:

[0052] Collecting chicken group response data after feed distribution;

[0053] Processing the chicken group response data through a recursive state estimator to generate state estimation results;

[0054] Inputting the state estimation results into the multi-dimensional difference evaluation module for difference analysis iteration.

[0055] Compared with the prior art, the beneficial effects of the present application are:

[0056] Through the environmental data acquisition module, multiple sensors are configured to perform initial processing on the multi-source data stream of the chicken group and generate standard chicken group data, realizing comprehensive perception and standardized processing of the growth environment, health status and feed consumption of the chicken group. The originally scattered and chaotic data is converted into effective information with unified standards, providing reliable basic data for subsequent analysis.

[0057] The cloud-edge collaborative analysis module constructs a dynamic feed supply model based on the standard chicken group data and outputs the theoretical feed demand, combining the real-time performance of edge computing and the big data processing capability of cloud computing. The feed supply model can be dynamically adjusted according to real-time data and historical big data of the chicken group, and the output theoretical feed demand is more in line with the actual demand changes of the chicken group, breaking the limitations of traditional local computing or single cloud analysis in timeliness and data processing depth.

[0058] The multi-dimensional difference evaluation module performs multi-dimensional comparative analysis on the theoretical feed demand and the actual feed consumption value and generates a feed difference coefficient matrix, considering the influence of environment, chicken health and other factors on feed consumption. Through multi-dimensional comparative analysis, the differences between theory and actual consumption and the potential associations causing the differences are clearly presented, avoiding one-sided conclusions that may be caused by single-dimensional analysis.

[0059] The abnormal area positioning module generates an abnormal feed supply probability distribution map based on the feed difference coefficient matrix combined with the topological information of the breeding farm, which can accurately lock the area with abnormal feed supply in the breeding farm, allowing staff to intuitively understand the distribution range and probability size of abnormal conditions, facilitating targeted measures to reduce the waste of manpower and time caused by blind investigation.

[0060] The feed supply strategy adjustment module generates adaptive feed control signals based on the abnormal feed supply probability distribution map, enabling feed adjustment to dynamically adapt to the specific conditions of abnormal areas, realizing individualization and precision of feed supply strategy, and avoiding the problem of excessive or insufficient feed supply in some areas under the unified feed supply mode.

[0061] The encrypted communication link module transmits the adaptive feed control signal to the feed distribution actuator by using a quantum encryption channel, and by using the high security of quantum encryption technology, ensures that the control signal cannot be disturbed, stolen or tampered with during transmission, ensures the safe and reliable transmission of the feed system control instructions, and maintains the stable operation of the entire feed regulation system. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A timing diagram of the laying hen breeding precise feed regulation system based on big data cloud edge collaboration according to the present application;

[0063] Figure 2 A flowchart of data processing of the environment data acquisition module;

[0064] Figure 3 A flowchart of modeling and prediction of the cloud edge collaborative analysis module;

[0065] Figure 4 A flowchart of abnormal area positioning of the abnormal area positioning module;

[0066] Figure 5 A flowchart of optimization of the gradient projection adjuster. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0068] Please refer to Figure 1 The present application provides a laying hen breeding precise feed regulation system based on big data cloud edge collaboration, which comprises an environment data acquisition module, a cloud edge collaborative analysis module, a multi-dimensional difference evaluation module, an abnormal area positioning module, a feed strategy regulation module and an encrypted communication link module.

[0069] The environmental data acquisition module is configured with chicken health sensors, feed consumption sensors and environmental monitoring sensors, performs initial processing on the acquired multi-source data stream of the chicken group, and generates standard chicken group data. The cloud-edge collaborative analysis module constructs a dynamic feed supply model based on the standard chicken group data and outputs the theoretical feed demand. The multi-dimensional difference evaluation module compares and analyzes the theoretical feed demand and the actual feed consumption value in multiple dimensions to generate a feed difference coefficient matrix. The abnormal area positioning module generates an abnormal feed supply probability distribution map based on the feed difference coefficient matrix combined with the topological information of the farm. The feed supply strategy adjustment module generates an adaptive feed control signal based on the abnormal feed supply probability distribution map. The encrypted communication link module transmits the adaptive feed control signal to the feed distribution executor through a quantum encryption channel. The system realizes the collaborative operation of data acquisition, analysis, evaluation, positioning, adjustment and communication through modular design, ensuring the accuracy and safety of the feed supply process.

[0070] Embodiment 1: see Figure 2 The implementation of the environmental data acquisition module is realized by deploying various sensor devices in the layer farm. The chicken health sensors adopt a distributed installation mode and are arranged around the periphery of the roost, feeding area and egg laying box in the henhouse. The body weight monitoring uses an embedded weighing platform. When the chicken stands on the platform, the pressure sensor collects the weight signal, and the data is transmitted to the acquisition terminal through ZigBee wireless transmission. The feeding frequency monitoring uses an infrared beam array to form a monitoring area at the entrance of the feeding trough, records the number of times and duration of the chicken's head breaking through the light beam, and each feeding event is marked with a timestamp and associated with an individual RFID identification. The egg laying rate data acquisition is realized by a photoelectric counter at the end of the egg conveying belt. Each egg triggers a counting pulse when passing through the detection area, and the weight sensor distinguishes abnormal eggs.

[0071] The environmental monitoring sensor network covers all functional areas of the henhouse. The wall-mounted temperature and humidity sensors are installed at a height of 1.5 meters from the ground, with one monitoring point per 100 square meters. The sensor probe has an external protective cover to prevent dust pollution. The light intensity sensor is installed on the henhouse roof truss, evenly distributed and avoiding direct lighting areas, collecting the integrated intensity value of natural light and artificial light. All sensor data are collected at a frequency of once per minute and preliminarily aggregated through an industrial Internet of Things gateway.

[0072] The raw data stream is first filtered by a hardware filter to remove power frequency interference, and then a software filtering algorithm is run on the gateway processor. For body weight data, a sliding window median filter is used, with a window width of 10 sampling points to eliminate transient fluctuations caused by chicken movement. The feeding frequency data uses threshold filtering to remove abnormal records with a duration of less than 1 second or more than 300 seconds. The environmental data is smoothed using Kalman filtering to reduce the influence of random interference. The body weight data is set to accept a range of 1.5 kg to 4.5 kg, and data points outside this range are marked as invalid. The reasonable threshold of feeding frequency is dynamically adjusted according to historical data, and abnormal values are excluded using the 3σ principle. The temperature and humidity data verification is based on the local climate model, with a temperature acceptance range of -5°C to 45°C and a humidity range of 30% to 85%. After all invalid data points are removed, the system automatically triggers the data re-sampling mechanism.

[0073] The time series alignment process uses a time synchronization scheme based on the NTP protocol, and each sensor node performs time calibration with the central server every 24 hours, with a maximum allowed deviation of 100 milliseconds. The data alignment algorithm uses the feeding event as the reference time axis to resample the data streams of different sensors onto a unified time grid. The aligned data is organized by minute granularity, forming a standard chicken flock data packet containing timestamp, sensor type, data value, and quality flag.

[0074] The original sampling data is saved in the ring buffer of the edge computing node, retaining the full amount of data for the last 72 hours. The preprocessed data is uploaded to the cloud platform object storage system and stored by date partition. The data index is established based on the combination key of chicken coop number, date, and time period, supporting multi-dimensional fast query. CRC checksum and retransmission mechanism are used during data transmission to ensure data integrity. Daily sensor health check is performed, including circuit self-test, signal strength detection, and data consistency verification. Faulty sensors automatically trigger an alarm and switch to backup nodes, and the system maintenance log records all device state changes. The sensor calibration period is 30 days, using standard calibration equipment for on-site calibration, and the calibration data is entered into the device management database.

[0075] RESTful API interface is provided upward for cloud-edge collaborative analysis module to call, supporting combined query by time range, sensor type, and spatial location. Downward communication with the sensor network through Modbus protocol supports real-time parameter configuration and data acquisition control. All data exchanges use JSON format encapsulation, including metadata description information.

[0076] The system is deployed considering the actual environmental conditions of the farm, the sensor selection meets the IP67 protection level, and is suitable for high dust and high humidity environment. Wireless transmission uses multi-band adaptive technology to avoid industrial equipment interference. The power supply system uses a solar auxiliary power supply scheme to ensure continuous operation in rainy weather. The device installation position avoids direct contact with the chicken area, and uses anti-pecking design.

[0077] The data processing algorithm is optimized to adapt to the edge computing resource limit, the filtering algorithm uses fixed-point number operation to reduce the calculation overhead, and the data compression uses lightweight LZ77 algorithm to reduce the storage space. Memory management uses pooling technology to avoid frequent memory allocation, and real-time performance monitoring tracks processing delay and resource usage. When the processing delay exceeds the threshold, the system automatically downgrades to a simplified processing mode. The data quality control establishes a multi-level supervision mechanism, the edge node performs real-time data verification, the cloud platform performs batch data quality analysis, and generates data quality reports regularly. The abnormal data root cause analysis tool helps to locate sensor faults or environmental interference problems, and maintenance personnel adjusts the device according to the system prompt. Historical data quality indicators are included in the device performance evaluation system. Columnar storage is used to improve compression efficiency, a special index structure is supported for time range queries, and data partitioning strategy is based on time sliding window automatic management. Data caching mechanism retains hot data in memory to reduce query response time. Data archiving strategy moves data more than a year to low-cost storage systems.

[0078] The sensor node supports hot addition, and the new node is automatically registered to the system after access and starts data collection. The processing capacity supports horizontal expansion, and when the breeding scale expands, the processing load can be shared by increasing edge computing nodes. All configuration information is centrally managed, supporting batch deployment and unified configuration update. Real-time display of the working state and data quality indicators of each sensor, abnormal state highlighted and processing suggestion provided. Historical data trend chart supports multi-parameter superimposed display, helping to analyze the correlation between environmental parameters and chicken behavior. Data export function supports standard format output, convenient for third-party analysis tools. Data collection period avoids water mist disinfection operation period to prevent sensor reading distortion. Device maintenance is arranged synchronously with the empty period of the farm, minimizing the impact on production activities. The wireless transmission power is optimized according to the layout of the chicken coop to ensure signal coverage while reducing energy consumption. All device housings use corrosion-resistant materials to adapt to the use of cleaning agents in the breeding environment.

[0079] Example 2: see Figure 3The implementation of the cloud-edge collaborative analysis module is based on the cooperative working architecture of the edge computing nodes deployed locally in the farm and the cloud server cluster. The edge computing nodes use industrial-grade embedded systems, equipped with multi-core processors and dedicated neural network accelerator cards, and are installed in the control cabinet of each chicken coop. These nodes are connected to the sensor network through gigabit Ethernet and receive real-time standard flock data transmitted by the environmental data acquisition module. The cloud server cluster is deployed on a cloud computing platform and uses a distributed architecture, including model training nodes and inference service nodes.

[0080] The training process of the dynamic feeding model uses historical flock operation data covering a complete production cycle of the past twelve months. The training data set includes parameters such as flock body weight, feeding frequency, egg production rate, temperature and humidity, and light intensity sampled every minute, as well as actual feed feeding records. Model training first performs feature engineering to extract meaningful features from raw data, including time series features such as moving average, trend slope, and statistical features such as variance, peak value, etc. k-fold cross-validation is used during training to prevent overfitting, and model parameters are optimized through the backpropagation algorithm.

[0081] The time series prediction unit uses a long short-term memory network structure, which includes an input layer, three hidden layers, and an output layer. The input layer receives standardized time series data, the hidden layer uses LSTM units to capture long-term dependencies, and the output layer produces a 24-hour feed demand prediction. The network training uses the Adam optimizer with an exponential decay learning rate strategy. The feature compensation unit integrates a random forest algorithm to handle the nonlinear relationship between environmental factors and feed demand. This unit receives real-time environmental data and outputs compensation coefficients to adjust the baseline prediction value.

[0082] The edge computing node obtains the latest standard flock data from the local sensor network every five minutes. Data acquisition uses a subscription and publication mode, with the node registered as a data consumer that pushes new data to the edge node immediately upon generation. The incremental data batch processing algorithm uses a sliding window mechanism with a window size of 60 sampling points (corresponding to 1 hour of data) and a window sliding step of 5 minutes. Each data batch is preprocessed, including missing value filling and outlier processing, and then fed into the feature extraction pipeline. Flock state features include behavior features, physiological features, and environmental features. Behavior features extract feeding activity frequency, duration, and regularity; physiological features calculate body weight change rate, egg production trend, and health indicators; environmental features integrate the effects of temperature, humidity, and light parameters. Feature extraction uses an autoencoder for dimensionality reduction to retain the most important feature representations. The extracted feature vectors are standardized and then fed into the model inference engine.

[0083] The time series prediction unit first processes the time series data to generate a baseline feed demand prediction. The feature compensation unit simultaneously analyzes the environmental parameters to calculate the impact of environmental factors on feed demand. The outputs of the two units are combined through a weighted fusion module to produce the final theoretical feed demand. The theoretical value is output in kilograms at five-minute intervals, accompanied by a confidence score.

[0084] The multi-dimensional discrepancy assessment module is deployed on the edge computing node, receiving the theoretical feed demand and actual values from the feed consumption sensors. The actual values are measured by weight sensors and flow meters, recording the actual feeding amount every five minutes. The time-domain cumulative deviation calculation uses a sliding window integration method with a window size of 24 hours to calculate the cumulative deviation between the theoretical value and the actual value. This calculation produces a time-domain deviation vector containing deviation indicators at multiple time scales, from minute-level to hour-level cumulative deviations.

[0085] The frequency domain energy shift detection converts the time-domain signal to the frequency domain through the Fast Fourier Transform. The theoretical demand and actual consumption data are first resampled to the same time resolution. Then the power spectral density of both is calculated to compare the energy distribution difference at different frequency components. Frequency domain analysis can reveal periodic supply-demand deviation patterns that may be masked in time domain analysis. The detection result generates a frequency domain shift vector containing the amplitude and phase difference of the main frequency components.

[0086] The sequence structure matching uses the dynamic time warping algorithm, which can handle nonlinear deformation between time series. The algorithm first constructs the distance matrix of the theoretical sequence and the actual sequence, and then finds the optimal warping path. The matching process considers the shape similarity and trend consistency of the sequences, producing a similarity score. This score is standardized and added to the sequence similarity vector, which also contains local and global similarity indicators. The tensor fusion process combines the time-domain deviation vector, frequency-domain shift vector, and sequence similarity vector into a multi-dimensional array. The fusion algorithm uses an attention mechanism to assign adaptive weights to different dimensional features. The weight coefficients are learned from historical data, reflecting the importance of each dimension feature to the final discrepancy assessment. The fused tensor is normalized to generate a feed discrepancy coefficient matrix. This matrix contains comprehensive discrepancy information, reflecting the inconsistency between theoretical demand and actual consumption from multiple angles.

[0087] The resource allocation of edge computing nodes is optimized to ensure real-time data processing and model inference. The model update mechanism supports online learning, automatically triggering model retraining when pattern changes are detected. The data flow processing uses a backpressure control mechanism to prevent processing delays caused by data congestion. All processing processes have log records and monitoring, facilitating problem troubleshooting and performance optimization. The communication protocol uses a lightweight message queue to ensure efficient data transmission between edge nodes and the cloud. The security mechanism includes data encryption and identity verification to prevent unauthorized access. The system supports smooth expansion, allowing the number of edge nodes to be increased as the scale of the farm expands without affecting the overall architecture. Maintenance tools provide remote monitoring and fault diagnosis functions to reduce operational costs.

[0088] Embodiment 3: Refer to Figure 4 The implementation of the anomaly area positioning module is based on the digital twin model of the physical layout of the farm. The module first divides the farm into several topological nodes, each corresponding to a physical area, usually in the form of a chicken coop, and also includes key control points on the feed delivery path. The node-based topological network is represented in a graph structure, where vertices represent spatial locations and edges represent connection relationships. Each vertex attribute includes information such as chicken coop number, spatial coordinates, and capacity specifications. Each edge attribute includes impedance parameters such as pipe length, diameter, inclination angle, and number of bends. The calculation of impedance parameters integrates fluid mechanics characteristics, with the main resistance coefficient determined by pipe material, internal smoothness, and fluid characteristics. During network construction, surveying and mapping data are used to establish an accurate coordinate reference system, ensuring consistency between the physical space and the digital model.

[0089] The mapping process of the feed difference coefficient matrix is implemented through a spatial registration algorithm. Each topological node is associated with the corresponding feed difference coefficient based on the physical area it serves. The mapping algorithm uses the nearest neighbor matching principle to establish a correspondence between each difference value in the matrix and the topologically closest node in space. For areas spanning multiple nodes, a weighted allocation strategy is used, with weights determined by the proportion of the service area. After mapping is complete, each topological node carries a difference coefficient value, which forms the node feature vector input to the graph neural network.

[0090] The graph neural network uses a message passing architecture to process the topological network. The network consists of an input layer, three layers of graph convolution layers, and an output layer. The graph convolution layer uses the following aggregation function to update the node representation:

[0091]

[0092] where: represents the feature vector of node in the layer, is the set of neighbor nodes of node , is the edge weight between nodes and is a trainable weight matrix, is an activation function. The edge weight is calculated from impedance parameters, reflecting the physical characteristics of the connection between nodes.

[0093] The abnormal propagation deduction adopts a random walk algorithm simulation. The algorithm starts from the high difference coefficient node and propagates along the topological edge simulation. An abnormal attenuation factor is introduced during the propagation process, which is calculated based on the impedance parameter: where represents the path impedance value from node to , is the attenuation coefficient, which is trained according to historical data. The propagation probability is inversely proportional to the attenuation factor. The larger the impedance, the lower the propagation probability.

[0094] The probability distribution map generation uses the kernel density estimation method, taking the node abnormal probability output by the graph neural network as input, and using the Gaussian kernel function for spatial smoothing processing. The bandwidth parameter of the kernel function is optimized according to the spatial scale of the farm to ensure the continuity and accuracy of the probability distribution. The generated probability distribution map is visualized in the form of a heat map, using a color gradient to represent the abnormal probability, and dark areas represent high-probability abnormal areas. The feed strategy adjustment module formulates the control strategy according to the probability distribution map, and the parameter configuration includes monitoring frequency, sampling accuracy and response threshold parameters. For areas with a probability value greater than 0.7, high-frequency monitoring mode is enabled, and the data collection interval is shortened from 5 minutes to 30 seconds. The monitoring range is expanded to the associated area, including the upstream feed path and adjacent chicken houses.

[0095] The swarm intelligence optimizer uses an improved particle swarm algorithm. During algorithm initialization, each particle represents a possible feed adjustment scheme, and the scheme dimension is the same as the number of topological nodes. The fitness function considers the difference coefficient, nutritional demand and operating cost:

[0096]

[0097] where: F(x) represents the value of the fitness function, is the feed amount vector, and each component xᵢ represents the adjusted feed amount recommended by the system for the i-th node (region) in the topological network, is the theoretical demand value, is the lower limit of nutritional demand, is the unit cost coefficient, ​is the weight parameter. The inertia weight and social learning mechanism are introduced in the particle updating process, gradually converging to the optimal solution, and n represents the total number of nodes (i.e., chicken coops or key control areas) in the farm topological network.

[0098] The generation of adaptive feeding control signals adopts a digital instruction format, and the signal contains target area identification, feeding amount value, execution timestamp, and priority flag. The instruction encoding adopts JSON format, which includes header information and data payload. The header information records the instruction sequence number and generation time, and the data payload contains specific control parameters. The signal is authenticated by digital signature before transmission, ensuring the integrity and authenticity of the instruction. When a high-probability abnormal area is detected, the system automatically triggers the early warning mechanism to inform the management personnel for on-site verification. At the same time, a detailed diagnostic report is generated, including the type of abnormality, possible causes, and treatment suggestions. All operation records are saved in the audit log, supporting post-analysis and process optimization.

[0099] The maintenance mechanism includes regular calibration and model updating. The topological network is updated monthly according to the actual layout changes, and the impedance parameters are dynamically adjusted according to the pipeline usage. The graph neural network is retrained quarterly, incorporating the latest production data. The system supports remote upgrading and configuration updating, ensuring the reliability and accuracy of long-term operation. The monitoring interface provides interactive abnormality analysis tools, allowing users to view probability distribution graphs through the visual interface and drill down to detailed data for specific nodes, simulating the effects of different control strategies. The system provides historical comparison functions, supporting comparison analysis of the current state with the same period in history, helping to identify periodic abnormal patterns. The security mechanism includes access control and operation audit, and users with different permissions have different operation ranges. Key operations require double authentication. All configuration modification records are logged, supporting change tracking and responsibility identification. Data backup adopts an incremental backup strategy to ensure the integrity and recoverability of system data.

[0100] Example 4: Referring to Figure 5 , the implementation of Example 4 involves the coordinated operation of the temperature drop search algorithm and the gradient projection adjuster in the feeding strategy adjustment module. The module receives the abnormal feeding probability distribution map from the abnormal area positioning module, which identifies three high-probability abnormal areas (Area A, Area B, and Area C) with abnormal probabilities of 0.82, 0.76, and 0.68, respectively. In the system initialization stage, the parameters of the temperature drop search algorithm are set: the initial temperature T0 is 1000, the temperature decay coefficient α is 0.95, and the maximum number of iterations is 200. The algorithm takes the current feeding scheme as the initial solution, which includes the baseline feeding amount for each area.

[0101] The temperature-decreasing search algorithm starts the iteration process. In the first round of iteration, the algorithm randomly generates a new solution, which increases the supply amount of region A by 5%, decreases the supply amount of region B by 3%, and keeps the supply amount of region C unchanged. The target function value corresponding to the solution is calculated, which is the evaluation value considering the difference coefficient matrix, nutritional requirements, and operating costs. If the target function value of the new solution is better than the current solution, it is directly accepted. If it is worse than the current solution, it is accepted according to the Metropolis criterion with a certain probability, which decreases with the decrease of temperature. After multiple iterations, the algorithm gradually converges to a better solution, and outputs the supply amount adjustment value of each region.

[0102] The conversion process uses a 12-bit precision digital-to-analog converter to convert digital quantities into 4mA-20mA analog signals, which correspond to the full-closed to full-open state of the supply actuator. The signal transmission adopts the industrial field bus protocol, which contains check code and timing control information to ensure that the command is accurately sent to the actuator. The gradient projection adjuster then optimizes the distribution of these adjustment values, which sets three types of constraints: first, the upper and lower limits of the supply amount, the supply amount of each region must be between the minimum nutritional requirement and the maximum capacity; second, the total supply amount constraint, the sum of the supply amounts of all regions cannot exceed the current available amount of the feed bin; third, the priority constraint, high abnormal probability regions have priority in distribution. The adjuster first calculates the sub-gradient of the target function of the current supply scheme, and then maps the scheme to the feasible region through the projection operation. The projection operation uses an iterative algorithm, which moves the current solution along the feasible direction in each iteration to gradually satisfy all constraints. When a constraint violation is detected, the adjuster adjusts the distribution of each region according to the weight coefficient, giving priority to the supply demand of high abnormal regions. After multiple projection iterations, an optimized supply scheme that satisfies all constraints is obtained. The system continuously monitors the execution effect in actual operation. The supply amount adjustment scheme is shown in Table 1.

[0103] Table 1: Supply amount adjustment scheme.

[0104] Region number Abnormal probability Current feed amount (kg / h) Adjustment value (kg / h) Optimized feed amount (kg / h) State identification A-12 0.82 125.6 +8.4 134.0 Emergency B-07 0.76 98.3 +5.2 103.5 Emergency C-15 0.68 112.8 +3.1 115.9 Warning D-09 0.45 87.2 -2.8 84.4 Normal E-21 0.31 76.5 -1.5 75.0 Normal

[0105] The adjuster maintains a real-time constraint database during operation, which contains dynamic parameters for each region: the minimum supply amount is dynamically calculated based on the age of the chicken population and the egg production rate, the maximum supply amount considers the physical capacity of the feeder and the flow rate of the conveying pipeline, and the priority weight is updated in real time according to the abnormal probability. The database synchronizes the latest data every 5 minutes to ensure the timeliness of the constraint conditions. The generation of adaptive supply control signals uses a multi-level structure, the first layer contains basic control instructions that specify the target region and supply amount; the second layer adds execution parameters including supply speed, duration, and smooth transition curve; the third layer contains safety check information such as maximum allowed adjustment amplitude and emergency stop conditions. The signal is packaged in ASN.1 encoding format to ensure the reliability of data transmission and the consistency of parsing.

[0106] The system takes into account the operational needs of actual farms, and the parameters of the temperature drop search algorithm can be adjusted according to different seasons and chicken stages: a faster temperature decay rate is used in summer, and the number of iterations is increased during the egg-laying peak. The constraint set of the gradient projection adjuster supports online modification, and the farm management personnel can adjust the nutritional demand parameters or equipment capacity limits through the human-machine interface.

[0107] The abnormality handling mechanism includes multiple protection levels. When the adjuster detects that all the constraints cannot be met, it first attempts to reduce the feed supply in low-priority areas; if it is still not feasible, it triggers the renegotiation mechanism to request upstream modules to revise the abnormality probability threshold; the final solution is to start the hierarchical response strategy, which prioritizes the minimum feed supply requirements of high-abnormality areas.

[0108] The system deployment uses a distributed architecture. The temperature drop search algorithm runs on edge computing nodes for fast response; the gradient projection adjuster is deployed on regional servers to coordinate the feed allocation of multiple breeding houses; the control signal generation module is located in the on-site controller, which directly drives the actuator. This architecture not only ensures computing efficiency, but also ensures system reliability. The monitoring system records the complete adjustment process. The candidate solution, objective function value, and acceptance decision of each iteration are recorded in the operation log. The iteration process of the projection operation and the constraint satisfaction are also recorded in detail. These data are used for subsequent analysis of algorithm performance and optimization of parameter settings. The visualization interface displays the feed adjustment process in real time, including current temperature, iteration number, constraint satisfaction status, and other information.

[0109] The maintenance mechanism includes periodic calibration and algorithm update. The parameters of the temperature drop search algorithm are re-optimized every quarter, using historical data to test the effects of different parameter combinations. The constraint conditions of the gradient projection adjuster are reviewed and updated every month to reflect equipment modifications or process changes. The system supports remote diagnosis and parameter adjustment, reducing the need for on-site maintenance.

[0110] Example 5: The feedback detection module is implemented based on continuous monitoring of the system response after feed adjustment. This module collects chicken group response data after feed distribution through a sensor network deployed throughout the chicken house, including changes in feeding behavior, body weight growth trends, egg-laying performance, and group activity patterns. The sensor network consists of high-precision scales, infrared activity monitors, sound collectors, and video analysis equipment, covering the entire breeding area in a distributed architecture. The data collection frequency is dynamically adjusted according to the magnitude of the feed adjustment, and when a large magnitude of feed adjustment is detected, the system automatically increases the data sampling rate to once per minute to capture more detailed response characteristics.

[0111] The recursive state estimator employs a Kalman filter framework to process the flock response data. The estimator maintains a state-space model that includes hidden variables such as flock health status, nutrient intake level, and environmental fitness. Each estimation process is divided into two stages: prediction and update. In the prediction stage, the state change trend is inferred based on historical data, and in the update stage, the latest observation data is integrated to correct the estimation results. The state estimator can handle noise and missing values in sensor data and obtain the most likely state sequence through probabilistic reasoning. The estimation results are output in the form of confidence intervals, reflecting the credibility of the estimation results.

[0112] The state estimation results are transmitted in real-time to the cloud-edge collaborative analysis module and the feed strategy adjustment module. The transmission process uses a lightweight message protocol to ensure the timeliness and integrity of data transmission. After receiving the state estimation results, the cloud-edge collaborative analysis module combines them with historical operation data for parameter updating of the dynamic feed model. The model updating uses an online learning mechanism to adjust the neural network weights through incremental training, enabling the model to adapt to changes in flock status. The feed strategy adjustment module uses the state estimation results to reevaluate the effectiveness of the current control strategy and triggers the strategy adjustment process if necessary.

[0113] The feedback data is iteratively analyzed in the multi-dimensional difference evaluation module, which compares the state estimation results with the expected flock response and calculates a new difference coefficient matrix. The iterative analysis uses a sliding window mechanism, with the window size dynamically adjusted based on data characteristics, typically containing the last 24 hours of data records. The analysis process pays special attention to the time delay effect between feed adjustment and flock response, establishing a delay compensation mechanism to accurately evaluate the control effect. The new difference coefficient matrix not only reflects the current supply-demand difference but also contains information on the effectiveness of control measures. The system establishes a closed-loop control architecture, enabling the feed strategy to be continuously optimized based on real-time feedback. After each feed adjustment, the system starts a complete monitoring-evaluation-adjustment cycle. This cycle usually lasts 6 to 12 hours, with the specific duration depending on the degree of flock response and the sufficiency of data collection. During the cycle, the system records all relevant data, including feed quantity changes, flock behavior responses, environmental parameter fluctuations, etc., providing complete data support for subsequent analysis.

[0114] The data management uses a time-series database to store all feedback data. The database design supports efficient time range queries and stream data processing, enabling quick retrieval of flock response records within a specific time period. The data index is based on timestamps and sensor types, supporting multi-dimensional data aggregation analysis. All data records include quality identifiers indicating the reliability and completeness of the data.

[0115] The anomaly detection mechanism continuously monitors the feedback data stream, and when an abnormal response pattern is detected, the system automatically triggers an early warning process. The early warning is divided into multiple levels, from slight deviation to severe anomaly, each level corresponds to different handling strategies. For slight deviation, the system automatically performs parameter fine-tuning; for severe anomaly, manual intervention is required for in-depth analysis and processing. All early warning events are recorded in the event log, including the occurrence time, anomaly type, handling measures and final result. System performance evaluation is based on statistical analysis of long-term running data. By comparing the chicken group state indicators before and after feed adjustment, the effectiveness of the control strategy is evaluated. Evaluation indicators include feed conversion rate, egg production consistency, group uniformity and other production performance parameters. These evaluation results are used to optimize system parameters and improve control algorithms, forming a virtuous cycle of continuous improvement.

[0116] The maintenance mechanism ensures the long-term stable operation of the system. Regular sensor calibration and data quality check are performed to ensure the accuracy of the input data. The parameters of the recursive state estimator are retrained every quarter, incorporating the latest production data. The system provides complete diagnostic tools to detect data stream anomalies, calculation biases and communication failures, etc. All maintenance operations are recorded in the system log to support fault tracing and performance analysis.

[0117] The human-computer interaction interface provides visual display of feedback data. The breeding management personnel can view the chicken group response trend chart, state estimation results and difference analysis report. The interface supports drilling query, which can go from summary data to detailed records of individual henhouse. The system also provides automatic report generation function, which generates operation summary and effect evaluation report regularly. The security mechanism protects the integrity and privacy of the feedback data. Data transmission uses encryption protocol to prevent unauthorized access. Access control system limits the data viewing and operation permissions of different users. Data backup adopts incremental backup strategy to ensure quick recovery in case of system failure. All data operations are recorded in the audit log to support security event investigation and responsibility tracing.

[0118] It should be noted that, in this text, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0119] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A large data cloud edge collaboration-based precise feed supply adjustment system for laying hen breeding, characterized in that, The method comprises the following steps: An environmental data acquisition module is configured with chicken health sensors, feed consumption sensors, and environmental monitoring sensors. The chicken multi-source data stream obtained by the environmental data acquisition module is initially processed to generate standard chicken data. A cloud-edge collaborative analysis module is used to construct a dynamic feeding model based on the standard chicken data and output the theoretical feed demand. A multi-dimensional difference evaluation module is used to compare and analyze the theoretical feed demand and the actual feed consumption value in multiple dimensions to generate a feed difference coefficient matrix. An abnormal area positioning module is used to generate an abnormal feed probability distribution map based on the feed difference coefficient matrix and the topological information of the farm. A feed strategy adjustment module is used to generate an adaptive feed control signal based on the abnormal feed probability distribution map. An encrypted communication link module is used to transmit the adaptive feed control signal to the feed distribution executor through a quantum encryption channel. The multi-dimensional difference evaluation module comprises the following steps: A time domain cumulative deviation calculation is performed on the theoretical feed demand and the actual feed consumption value to generate a time domain deviation vector. A frequency domain energy offset detection is performed on the theoretical feed demand and the actual feed consumption value to generate a frequency domain offset vector. A sequence similarity evaluation is performed on the theoretical feed demand and the actual feed consumption value based on a sequence structure matching algorithm to generate a sequence similarity vector. The time domain deviation vector, the frequency domain offset vector, and the sequence similarity vector are fused to generate the feed difference coefficient matrix.

2. The precise feeding regulation system for layer chicken breeding based on big data cloud edge collaboration according to claim 1, wherein The environmental data acquisition module comprises the following steps: The chicken body weight data, feeding frequency data, and egg production rate data are collected through the chicken health sensors. The temperature and humidity data and the light intensity data are collected through the environmental monitoring sensors. Noise filtering is performed on the collected chicken multi-source data stream, and abnormal data points are removed according to a preset data quality threshold. The filtered chicken multi-source data stream is time series aligned to output the standard chicken data.

3. The precise feeding regulation system for layer chicken breeding based on big data cloud edge collaboration according to claim 2, characterized in that, The cloud-edge collaborative analysis module comprises the following steps: The dynamic feeding model is trained based on historical chicken operation data, and the dynamic feeding model comprises a time series prediction unit and a feature compensation unit. The standard chicken data is obtained in real time through an edge computing node. Incremental data batch processing algorithm is used to process the standard chicken data in batches to extract chicken state features. The chicken state features are input into the dynamic feeding model to output the theoretical feed demand.

4. The precise feeding regulation system for layer chicken breeding based on big data cloud edge collaboration according to claim 1, characterized in that, The abnormal area positioning module comprises the following steps: A node topology network is constructed according to the layout of the farm, and the location information of the chicken coop and the impedance parameters of the feed conveying path are labeled. The feed difference coefficient matrix is mapped to the corresponding nodes of the node topology network. An abnormal feed propagation path is deduced using a graph neural network, and an abnormal attenuation factor is calculated based on the node impedance parameters. The abnormal feed probability distribution map covering the entire farm is generated to identify high-probability abnormal areas.

5. The precise feeding regulation system for layer farming based on big data cloud edge collaboration according to claim 4, characterized in that, The feed strategy adjustment module comprises the following steps: Feed verification parameters are configured according to the abnormal feed probability distribution map. High-frequency monitoring mode is enabled for the high-probability abnormal areas. A swarm intelligence optimizer is used to adjust the feed control rules. The adaptive feed control signal is generated based on the adjusted feed control rules.

6. The precise feeding regulation system for layer farming based on big data cloud edge collaboration according to claim 5, characterized in that, The feed strategy adjustment module further comprises: calculating a feed amount adjustment value through a temperature drop search algorithm; converting the feed amount adjustment value into the adaptive feed control signal; optimizing the feed amount allocation based on preset feed constraints using a gradient projection adjuster.

7. The precise feeding regulation system for layer farming based on big data cloud edge collaboration according to claim 6, characterized in that, The gradient projection adjuster comprises: setting a set of feed amount constraints; calculating a sub-gradient of a target function according to a current feed plan; updating the feed plan through a projection operation to meet all the constraints; outputting the optimized feed plan to the adaptive feed control signal.

8. The precise feeding regulation system for layer farming based on big data cloud edge collaboration according to claim 1, wherein, Further comprising: a feedback detection module for monitoring adjusted feed consumption data through a recursive state estimator; feeding the monitoring data back to the cloud-edge collaborative analysis module and the feed strategy adjustment module; updating the feed strategy based on the feedback data.

9. The precise feeding regulation system for layer farming based on big data cloud edge collaboration according to claim 8, characterized in that, The feedback detection module comprises: collecting chicken group response data after feed allocation; processing the chicken group response data through a recursive state estimator to generate a state estimation result; inputting the state estimation result into the multi-dimensional difference evaluation module for difference analysis iteration.

Images