An intelligent decision-making management system for pig farming based on multimodal information monitoring

Abstract

The present invention discloses a pig farming intelligent decision-making management system based on multimodal information monitoring, which relates to the technical field of pig farming management. The intelligent decision-making management system includes: an information monitoring module, an abnormal warning module, an analysis and decision module, a management and control module, and an intelligent collaboration module, wherein the information monitoring module is used to divide the pig house into several enclosure areas through enclosure fences, and use the monitoring equipment configured in each enclosure area to monitor the multimodal information of the pig herd growth process in the enclosure area in real time. The present invention realizes the intelligent management of the whole process of pig farming by comprehensively applying information monitoring, abnormal warning, analysis and decision, management and control, and intelligent collaboration modules; comprehensively grasps the growth status of the pig herd by real-time collection of multimodal information in the enclosure area; analyzes information data in real time through edge computing equipment, quickly identifies and warns of abnormal situations, ensures timely handling of potential problems, and avoids the development of abnormal problems.

Description

Technical Field

[0001] The present invention relates to the technical field of pig breeding management, and in particular to a pig breeding intelligent decision-making management system based on multimodal information monitoring. Background Art

[0002] At present, pig farming is an important part of global agricultural production, providing an important guarantee for meeting the growing demand for meat. The development of pig farming shows a trend of scale, intensification and modernization, which leads to the continuous expansion of pig farming scale and the gradual improvement of technical level. However, with the expansion of farming scale, the challenges and risks faced in the farming process are also gradually increasing, such as disease prevention and control, environmental pollution, feed safety and other issues. Therefore, how to improve farming safety and sustainability while ensuring production efficiency has become the core concern in the current pig farming field.

[0003] With the expansion of breeding scale and the prominence of environmental problems, the safety warning and decision-making management of pig breeding are particularly important. Traditional breeding management methods mainly rely on breeding experience and intuition, lack of scientific basis and data support, and are prone to problems such as untimely discovery of problems and inaccurate decision-making.

[0004] At present, pig farming decision-making management technologies include sensor monitoring, data analysis and mining, intelligent early warning systems, and optimized decision support systems. These technologies provide timely early warning and decision support by real-time monitoring of multiple data such as the environment, feed, and pig health. However, current technologies still have some shortcomings and defects, such as incomplete and inaccurate data collection, high complexity of data processing, inadequate intelligent early warning and decision-making systems, lack of effective multimodal data fusion and analysis methods, information transmission delays, and insufficient real-time and accuracy. As a result, potential risks in the farming process cannot be discovered and handled in a timely manner, such as disease outbreaks, environmental deterioration, and feed waste, which in turn affects the health and production efficiency of pigs, increases farming costs, and reduces market competitiveness.

[0005] At the same time, the lag in data processing and decision-making may lead to an imbalance in market supply and demand, price fluctuations, and even food safety issues, endangering consumer health and industry reputation. Therefore, in order to solve these technical problems, building an efficient and intelligent pig farming decision-making management system is essential to promote the modernization and sustainable development of the pig farming industry.

[0006] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0007] 1. Technical issues to be resolved

[0008] In view of the shortcomings of the prior art, the present invention provides an intelligent decision-making management system for pig farming based on multimodal information monitoring, which has the advantages of improving the monitoring and management accuracy of pig farming, improving farming efficiency and pig health level, reducing feed waste and disease transmission, and optimizing resource allocation, thereby solving the problems of incomplete and inaccurate data collection, high complexity of data processing, inadequate intelligent early warning and decision-making systems, and lack of effective data processing means.

[0009] (II) Technical solution

[0010] In order to achieve the above advantages of improving the monitoring and management accuracy of pig farming, improving farming efficiency and pig health, reducing feed waste and disease transmission, and optimizing resource allocation, the specific technical solutions adopted by the present invention are as follows:

[0011] An intelligent decision-making management system for pig farming based on multimodal information monitoring, the intelligent decision-making management system includes: an information monitoring module, an abnormal warning module, an analysis and decision module, a management and control module and an intelligent collaboration module, wherein:

[0012] The information monitoring module is used to divide the pig house into several enclosure areas through enclosure fences, and use the monitoring equipment configured in each enclosure area to monitor the multimodal information of the pig herd growth process in the enclosure area in real time;

[0013] The abnormal warning module is used to monitor the multi-modal information collected in real time and the information transmission path through the edge computing devices deployed inside the pig house, using local computing resources, identifying abnormal phenomena in the information monitoring and transmission process, and triggering early warning notifications after abnormal phenomena are found;

[0014] The analysis and decision-making module is used to define each enclosure area as an intelligent entity, and use the cloud computing resources of the cloud computing platform to analyze the group growth status of the pigs in the enclosure area, as well as the activity trajectory and individual growth status of the pigs in the herd based on multimodal information; and based on the status analysis results, combined with market information and historical data, optimize personalized breeding strategies;

[0015] The management and control module is used to integrate the management of all enclosure areas, dynamically and visually display the breeding information of individual pigs in the breeding pig house, and implement personalized breeding strategies and regulate breeding parameters;

[0016] The intelligent collaboration module is used to manage the communication between edge computing devices and the cloud computing platform, and to distribute the computing tasks of the intelligent agents to the cloud computing platform to maintain the autonomous learning and collaborative work of the intelligent agents.

[0017] Furthermore, the information monitoring module includes: an intelligent ear tag unit, an environmental monitoring unit, a feeding metering unit, an excretion metering unit, a video monitoring unit and a wireless transmission unit, wherein:

[0018] Intelligent ear tag unit, used to collect pig vital signs information of individual pigs in real time;

[0019] Environmental monitoring unit, used to collect environmental information in the enclosure area in real time;

[0020] The feeding metering unit is used to set the metering cycle, collect the amount of feed fed and the amount of remaining feed in the enclosure area within the metering cycle, and integrate it into the feeding information in combination with the feed type;

[0021] Excretion metering unit, used to count the weight of excrement of pigs in the pen area during the metering period, defined as excretion information;

[0022] A video monitoring unit is used to monitor the pigs in the enclosure area using a camera and an infrared camera, respectively acquiring video images and infrared images, and defining them as monitoring information;

[0023] The wireless transmission unit is used to establish a communication channel between multimodal information and edge computing devices, and use wireless communication technology to achieve interactive communication of multimodal information; wherein the multimodal information includes pig vital signs information, environmental information, feeding information, excretion information and monitoring information.

[0024] Furthermore, the analysis and decision-making module includes: a storage management unit, a pig group analysis unit, an individual analysis unit, a market evaluation unit, a strategy matching unit, a self-learning optimization unit and an intelligent agent communication unit, wherein:

[0025] The storage management unit is used to receive the multimodal information after anomaly identification, upload it to the cloud computing platform, store it in the cloud database, and perform partition management according to data type and time series; at the same time, it stores historical growth data, historical multimodal information, historical strategy information and decision rule library;

[0026] The pig herd analysis unit is used to build a pig herd status model based on environmental information, feeding information, excretion information and the growth stage of the pig herd, and analyze the group growth status of the pig herd within the measurement period;

[0027] The individual analysis unit is used to record the activity trajectory of individual pigs based on the pig's vital signs parameters and monitoring information, and to construct an individual status model to analyze the individual growth status of individual pigs;

[0028] Market evaluation unit, used to set the purchase cycle, analyze the cost trend of various types of breeding materials in real time, and predict the material demand in the next purchase cycle and define market information;

[0029] Strategy matching unit, which is used to generate personalized breeding strategies based on the status analysis results of pig herds and individual pigs, combined with market information, matching decision information from the historical database;

[0030] The self-learning optimization unit is used to adaptively update the status information of the pig herd and individual pigs in the enclosure area and the personalized breeding strategy corresponding to the enclosure area based on the measurement period, and optimize the local model parameters of the intelligent body by taking the new status information and breeding strategy as new knowledge;

[0031] The intelligent agent communication unit is used to establish data sharing and information synchronization with intelligent agents in other captive areas, and to support collaborative decision-making and global resource optimization among multiple intelligent agents.

[0032] Furthermore, based on the environmental information, feeding information, excretion information and the growth stage of the pigs, a pig herd status model is constructed to analyze the group growth status of the pigs during the measurement period, including:

[0033] A weighing cycle is set to regularly measure the weight of each individual pig in the pen area, obtain the growth stage identification of the pig herd, and add up the individual weights of each individual pig as the herd weight information of the pig herd in the pen area; wherein the weighing cycle includes multiple measurement cycles;

[0034] The environmental information, feeding information, excretion information, pig weight information and growth stage identification are standardized, and effective features are extracted to construct feature vectors to form a pig sample set;

[0035] Set the number of categories of the population growth state as the number of clusters, set the original cluster center and the randomly initialized membership matrix, and then define the fuzzy factor and the maximum number of iterations;

[0036] Calculate the membership matrix of the group growth status of each sample in the pig herd sample set;

[0037] Use the cluster center iteration formula to calculate the cluster center of the next iteration;

[0038] Iterate until the membership matrix changes less than the set membership threshold or reaches the maximum number of iterations, then stop iterating, obtain the membership of each sample, and determine the cluster to which each sample belongs;

[0039] Based on the cluster to which the sample belongs, the group growth status of the pig herd in the pen area during the current measurement period is obtained, and the group growth index of the group growth status is calculated.

[0040] Furthermore, based on the cluster to which the sample belongs, the group growth status of the pig herd in the pen area during the current measurement period is obtained, and the group growth indicators of the group growth status are calculated, including:

[0041] If all samples belong to the same cluster, the population growth state corresponding to the cluster is used as the population growth state of the enclosure area in the current measurement period; if there are samples belonging to different clusters, the average membership method is used to calculate the average membership of the samples to each cluster, and the cluster with the largest average membership is selected as the population growth state of the enclosure area in the current measurement period;

[0042] Obtain the ideal state index of the current group growth state of the pig herd, and calculate the deviation between the standardized environmental information, feeding information and excretion information and the ideal state index, which is defined as ( o , p , q ); then calculate ( o , p , q ) and the coordinate origin (0, 0, 0), and the distance value is defined as the population growth index of the population growth state, where the calculation formula of the population growth index is:

[0043]

[0044] In the formula, d express( o , p , q ) and the coordinate origin (0, 0, 0); o Indicates the degree of deviation between environmental information and ideal state indicators; p Indicates the degree of deviation between feeding information and ideal state indicators; q Indicates the degree of deviation between excretion information and ideal state indicators.

[0045] Furthermore, based on the pig's vital signs parameters and monitoring information, the activity trajectory of individual pigs is recorded, and an individual status model is constructed to analyze the individual growth status of individual pigs, including:

[0046] The location information of individual pigs is obtained through the smart ear tags worn by individual pigs, and the distinguishing marks of individual tags are used to record the activity trajectory of each individual pig in the enclosure area and measurement period;

[0047] Obtain a set of thermal infrared images of the enclosure area taken by an infrared camera. In the thermal infrared images, classify the pigs according to their excretion posture and excrement form, and mark them as excretion behavior; classify the pigs according to their eating posture and feed form, and mark them as eating behavior; classify the pigs according to their sleeping posture and sleeping heart rate, and mark them as sleeping behavior;

[0048] Synchronize the timestamps between the thermal infrared image and the location information recorded by the smart ear tag, draw the enclosure area as a plane coordinate system, define the location information of each individual pig as a coordinate point, and map the coordinate point location information to the thermal infrared image, and annotate the individual labels of the individual pigs in the thermal infrared image in real time;

[0049] Based on the annotation results of thermal infrared images, an individual state model with an efficient neural network as the backbone network is established, and a weighted bidirectional feature fusion network is used as the feature network; the activity behaviors of the annotated individual pigs are extracted from the thermal infrared images collected in real time, and the number and duration of the activity behaviors are recorded;

[0050] Statistics are kept of the activity behavior timetable of each individual pig in each measurement period. A standard behavior timetable is set according to the growth stage of the pigs. By comparing the activity behavior timetable with the standard behavior timetable, the individual growth status of each individual pig in the measurement period is analyzed.

[0051] Furthermore, we set the purchasing cycle, analyze the cost trend of various types of breeding materials in real time, and predict the material input in the next purchasing cycle. The market information is defined as follows:

[0052] Set the purchase cycle of breeding materials for piggery, obtain real-time material price data of breeding materials through the data interface, and record the purchase information of historical purchase cycles; then use the time series analysis method to draw the future cost trend of breeding materials and predict the purchase cost of the purchase cycle;

[0053] Based on the herd growth status of the pigs in the enclosure area and combined with historical growth data, the material demand of the pigs in the next procurement cycle is predicted, and the total material demand of all enclosure areas is counted;

[0054] Get the maximum storage capacity and the remaining amount in the cycle of the material warehouse in the pig breeding house, take the difference between the maximum storage capacity and the remaining amount in the cycle as the upper limit of the material input, and then take the difference between the total material demand and the remaining amount in the cycle as the lower limit of the material input, and define the value range of the material input in the next procurement cycle;

[0055] Integrate the purchase cost and total material requirement for the next purchase cycle and define it as market information.

[0056] Furthermore, based on the status analysis results of the pig herd and individual pigs, combined with market information, the decision information is matched from the historical database to generate personalized breeding strategies including:

[0057] Obtain the analysis results of the group growth status of the pig herds and the individual growth status of the pigs in all the enclosures in the current measurement period, identify whether there are abnormal conditions in each enclosure area, if there are abnormal conditions, capture the location of the abnormal pig herds or abnormal pigs, and issue emergency response strategies, if there are no abnormal conditions, execute conventional strategy decisions;

[0058] Define the breeding strategy decision for the next metering period as the target strategy, and integrate the multimodal information and market information in each enclosure area to define it as the strategy attribute of the target strategy;

[0059] Connect to the historical database and calculate the attribute distance between the corresponding strategy attributes of the target strategy and the historical case strategy. The calculation formula of the attribute distance is:

[0060]

[0061] In the formula, D ( Q ik , Q xk ) indicates the target strategy C i Case Studies and Strategies C x The attribute distance between Q xk The target strategy k Item policy attributes; Q ik Indicates i The historical case strategy k The policy attribute corresponding to k Item strategy attribute; G ik Indicates the target strategy k The value of the policy attribute; G xk Represents historical case strategy k The value of the item policy attribute;

[0062] Calculate the similarity between the target strategy and the historical case strategy, where the similarity calculation formula is:

[0063]

[0064] In the formula, S ( C i , C x ) indicates the target strategy C i Case Studies and Strategies C x The similarity between ωk Indicates the weight value of each policy attribute; N x Indicates the number of policy attributes;

[0065] A similarity threshold is set, the historical database is traversed, historical case strategies that meet the similarity threshold are screened, and a historical strategy set is established; and the simulated annealing algorithm is used to select the optimal reuse strategy from the historical strategy set as the personalized breeding strategy for each captive area.

[0066] Furthermore, the intelligent collaboration module includes: a communication management unit, a task allocation unit, a synchronous learning unit and a collaborative work unit, wherein:

[0067] The communication management unit is used to manage the data transmission between the edge computing device and the cloud computing platform using the message queue telemetry transmission protocol; at the same time, a lightweight machine learning model is deployed on the edge computing device to handle the data preprocessing and decision-making tasks of multimodal information;

[0068] A task allocation unit is used to dynamically adjust computing tasks and allocate them to various edge computing devices using a load balancing method based on the availability of computing resources and data transmission delay;

[0069] Synchronous learning unit, used to deploy federated learning framework, support collaborative training of multiple agents, regularly aggregate local models of agents to update to the cloud computing platform, and dynamically adjust model parameters;

[0070] The collaborative working unit is used to create a digital twin model of the pig house, establish real-time synchronization and interaction between the virtual environment and the actual environment, synchronize multimodal information, and use digital twin technology to simulate and predict humanized breeding strategies to further optimize breeding strategies and management plans.

[0071] Furthermore, a federated learning framework is deployed to support collaborative training of multiple agents, and local models of agents are aggregated and updated to the cloud computing platform regularly. Model parameters are dynamically adjusted, including:

[0072] Deploy the federated learning framework on the cloud computing platform, generate an initial global model, and distribute it to each agent. Each agent receives the initial global model as the initial version of the local model.

[0073] The agent uses the local model to process the multimodal information in the corresponding enclosure area, obtains new knowledge after processing, updates the local model parameters, and then uploads them to the cloud computing platform;

[0074] The cloud computing platform receives the model parameters uploaded by each intelligent agent, generates new global model parameters using an aggregation algorithm, updates the global model, distributes it to each intelligent agent, and optimizes the local model.

[0075] (III) Beneficial effects

[0076] Compared with the prior art, the present invention provides an intelligent decision-making management system for pig farming based on multimodal information monitoring, which has the following beneficial effects:

[0077] (1) Through the comprehensive application of information monitoring, abnormal warning, analysis and decision-making, management and control, and intelligent collaboration modules, intelligent management of the entire pig farming process is realized. First, by real-time collection of multimodal information in the enclosure area, the growth status of the pig herd is fully understood. The local computing resources of the edge computing device are used to analyze the information data in real time, quickly identify and warn of abnormal situations, ensure that potential problems are dealt with in a timely manner, and avoid the development and spread of abnormal problems. Then, with the help of the cloud computing platform, the multimodal information is further analyzed in depth to evaluate the growth status of the pig herd and individuals, and combined with market trends and historical data, personalized breeding strategies are optimized to ensure that pigs at each growth stage can obtain the best breeding plan. This has greatly improved the refinement and scientific level of breeding management, reduced labor costs and resource waste, improved the health and production efficiency of pigs, and ensured food safety.

[0078] (2) Through integrated management and dynamic visualization, the system can fully control various information in the breeding process, adjust breeding parameters in a timely manner, and optimize resource allocation; the intelligent collaboration module ensures efficient and reliable data transmission between edge computing devices and cloud computing platforms, and through the federated learning framework, realizes collaborative training and autonomous learning of multiple agents, improves the overall intelligence level and decision-making ability of the system, and promotes the breeding industry to move towards intelligent, modern and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0080] Figure 1 It is a system principle block diagram of a pig farming intelligent decision-making management system based on multimodal information monitoring according to an embodiment of the present invention.

[0081] In the figure:

[0082] 1. Information monitoring module; 101. Intelligent ear tag unit; 102. Environmental monitoring unit; 103. Feeding metering unit; 104. Excretion metering unit; 105. Video monitoring unit; 106. Wireless transmission unit; 2. Abnormal warning module; 3. Analysis and decision-making module; 301. Storage management unit; 302. Pig herd analysis unit; 303. Individual analysis unit; 304. Market evaluation unit; 305. Strategy matching unit; 306. Self-learning optimization unit; 307. Intelligent communication unit; 4. Management and control module; 5. Intelligent collaboration module; 501. Communication management unit; 502. Task allocation unit; 503. Synchronous learning unit; 504. Collaborative work unit. DETAILED DESCRIPTION

[0083] According to an embodiment of the present invention, a pig farming intelligent decision-making management system based on multimodal information monitoring is provided.

[0084] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an embodiment of the present invention, an intelligent decision-making management system for pig farming based on multimodal information monitoring includes: an information monitoring module 1, an abnormal warning module 2, an analysis and decision module 3, a management and control module 4 and an intelligent collaboration module 5.

[0085] The information monitoring module 1 is used to divide the pig house into several enclosure areas through enclosure fences, and use the monitoring equipment configured in each enclosure area to monitor the multimodal information of the pig herd growth process in the enclosure area in real time.

[0086] In the description of the present invention, the information monitoring module 1 includes: an intelligent ear tag unit 101, an environment monitoring unit 102, a feeding metering unit 103, an excretion metering unit 104, a video monitoring unit 105 and a wireless transmission unit 106.

[0087] The intelligent ear tag unit 101 is used to collect the vital signs information of individual pigs in real time.

[0088] Specifically, the smart ear tag unit 101 is a device worn on the pig's ear and is equipped with a variety of sensors for real-time collection of individual pig vital signs, such as body temperature, heart rate, position, etc. For example, a small low-power LIS2DH three-axis acceleration sensor is used to monitor and identify pig movement behavior, and an I 2 The MAX30205 and MAX30102 sensors of C communication mode are used to monitor the body temperature and heart rate parameters of pigs respectively. The collected vital sign data of each pig is transmitted from the controller CC2640 to the wireless transmission unit 106 via the LoRa protocol SX1278 for data fusion.

[0089] The environmental monitoring unit 102 is used to collect environmental information in the enclosure area in real time.

[0090] Specifically, the environmental monitoring unit 102 monitors the environmental information in the enclosure area in real time, such as temperature, humidity, gas concentration (such as ammonia), light intensity, etc. It installs environmental sensors in the enclosure area, including temperature sensors, humidity sensors, gas sensors, and light sensors. The sensors regularly collect corresponding environmental information and transmit it to the edge computing device.

[0091] The feeding metering unit 103 is used to set a metering cycle, collect the amount of feed fed and the amount of remaining feed in the enclosure area within the metering cycle, and integrate them into the feeding information in combination with the feed type.

[0092] Specifically, a metering cycle is set, and the amount of feed fed and the amount of remaining feed in the enclosure area during the metering cycle are collected, and combined with the feed type, they are integrated and defined as feeding information. A weight sensor and an RFID reader need to be installed in the feed trough. The weight sensor monitors the weight changes of the feed in real time and records the amount of feed added and consumed each time; the RFID reader identifies the feed type. In the subsequent process, the edge computing device aggregates the weight sensor data and RFID data, calculates the total amount of feed fed and the remaining amount in each metering cycle, and integrates them into feeding information.

[0093] The excretion metering unit 104 is used to count the weight of excrement of the pigs in the pen area within a metering period, which is defined as excretion information.

[0094] Specifically, the weight of the excrement of the pigs in the pen area during the measurement period is defined as excrement information. A weight sensor needs to be installed in the excrement collection area in the pen area. The weight sensor records the weight of the excrement at regular intervals and transmits it to the edge computing device.

[0095] The video monitoring unit 105 is used to monitor the pigs in the enclosure area using a camera and an infrared camera, respectively acquiring video images and infrared images, and defining them as monitoring information.

[0096] The wireless transmission unit 106 is used to establish a communication channel between the multimodal information and the edge computing device, and use wireless communication technology to realize the interactive communication of the multimodal information. The multimodal information includes pig vital signs information, environmental information, feeding information, excretion information and monitoring information.

[0097] Specifically, the wireless transmission unit 106 can be configured with a communication module that supports Wi-Fi, LoRa or other wireless communication protocols. Through the wireless communication module 106, pig vital signs information, environmental information, feeding information, excretion information and monitoring information are transmitted to the edge computing device in real time. The edge computing device receives and integrates multimodal information, performs data analysis and processing, and ensures the real-time and reliability of information interaction.

[0098] Abnormal warning module 2 is used to monitor the multimodal information and information transmission path collected in real time through the edge computing equipment deployed inside the pig house, using local computing resources, identify abnormal phenomena in the information monitoring and transmission process, and trigger early warning notifications after abnormal phenomena are found.

[0099] Specifically, edge computing devices are deployed in each enclosure area or within a certain range, and they must have sufficient computing and storage capabilities. The edge computing devices obtain multimodal data from the information monitoring module in real time, and identify abnormal data in the multimodal information by configuring the early warning trigger mechanism. When the abnormal phenomenon cannot be solved by automatic processing, an early warning notification is sent to the breeding management personnel.

[0100] Early warning notifications are sent through a variety of methods, such as SMS, email or dedicated early warning applications, and describe the abnormal phenomenon in detail, including the type of abnormality, the time and place of occurrence, possible causes and recommended treatment measures. For example, if the body temperature of a pig is abnormally high, the notification content should include the abnormal pig's number, body temperature value, normal range, possible disease risks and recommended treatment measures (such as immediate isolation, contacting a veterinarian, etc.).

[0101] The analysis and decision-making module 3 is used to define each enclosure area as an intelligent entity, and use the cloud computing resources of the cloud computing platform to analyze the group growth status of the pigs in the enclosure area, as well as the activity trajectory and individual growth status of the pigs in the pig herd based on multimodal information. Based on the status analysis results, combined with market information and historical data, the personalized breeding strategy is optimized.

[0102] Among them, the intelligent agent is an intelligent entity, and its English name is Agent. It is based on the cloud and takes AI as the core to build a three-dimensional perception, global collaboration, accurate judgment, continuous evolution, and open intelligent system. In the present invention, each enclosure area is defined as an intelligent agent, which means that the pig herd and its environmental data in each area are regarded as an independent computing and management unit. The definition of an intelligent agent is an entity that can perceive, make decisions and perform tasks autonomously. In pig farming, defining the enclosure area as an intelligent agent helps to refine management and facilitate the implementation of personalized breeding strategies for pig herds in different areas. Through this definition, multimodal information in each enclosure area (such as environmental data, feed consumption, excrement, vital signs parameters, etc.) can be independently processed and analyzed, so as to understand the group growth status and individual growth status of the pig herds in each area in real time.

[0103] With the help of cloud computing platform, centralized management of multiple intelligent agents and efficient use of computing resources are achieved. Cloud computing platform provides powerful data processing and storage capabilities, which can integrate the data of multiple intelligent agents for in-depth analysis, and combine market information with historical data to generate optimized personalized breeding strategies. This connection enables the system to excel in large-scale data processing and real-time decision-making, ensuring that each enclosure area can be optimally managed and monitored, thereby improving breeding efficiency and pig health.

[0104] In the description of the present invention, the analysis and decision module 3 includes: a storage management unit 301, a pig group analysis unit 302, an individual analysis unit 303, a market evaluation unit 304, a strategy matching unit 305, a self-learning optimization unit 306 and an intelligent agent communication unit 307.

[0105] The storage management unit 301 is used to receive the multimodal information after abnormality identification, upload it to the cloud computing platform, store it in the cloud database, and perform partition management according to data type and time series. At the same time, it stores historical growth data, historical multimodal information, historical strategy information and decision rule library.

[0106] Specifically, the edge computing device sends the multimodal information after anomaly identification, uploads it to the cloud computing platform, stores it in the cloud database, and manages it in partitions according to data type and time series. The unit also stores historical growth data, historical multimodal information, historical strategy information, and a decision rule library.

[0107] The pig herd analysis unit 302 is used to construct a pig herd status model based on environmental information, feeding information, excretion information and the growth stage of the pig herd, and analyze the group growth status of the pig herd within the measurement period.

[0108] Among them, the pig herd status model is a data analysis model built based on the fuzzy C-means clustering algorithm (FCM). Using the fuzzy C-means clustering algorithm (FCM), it can analyze and evaluate the group growth status of the pig herd in the penned area during the current measurement period based on the environmental information, feeding information and excretion information collected in the penned area and combined with the growth stage of the pig herd.

[0109] In the description of the present invention, based on environmental information, feeding information, excretion information and the growth stage of the pig herd, a pig herd state model is constructed, and the group growth state of the pig herd within the measurement period is analyzed, including:

[0110] S321, set a weighing cycle, regularly measure the weight of each individual pig in the pen area, obtain the growth stage identification of the pig group, and add up the individual weight of each individual pig as the herd weight information of the pig group in the pen area. The weighing cycle includes multiple measurement cycles.

[0111] S321. Standardize the environmental information, feeding information, excretion information, pig weight information and growth stage identification, extract effective features to construct feature vectors, and form a pig sample set.

[0112] Specifically, multimodal information within the current metering cycle is collected, including:

[0113] 1. Environmental information: temperature, humidity, air quality (ammonia, carbon dioxide concentration, etc.), light intensity, ventilation conditions, etc.

[0114] 2. Feeding information: feed type, feed amount, feeding time and frequency.

[0115] 3. Excretion information: stool volume, urine volume, and excretion frequency.

[0116] 4. Growth stage information: piglet, fattening pig, adult pig and other stage identification.

[0117] It is necessary to remove invalid data and noise from the above multimodal information and process missing values; then standardize the data of different dimensions to ensure the comparability of the data.

[0118] Extract features from preprocessed data and construct feature vectors for cluster analysis. For example:

[0119] Environmental characteristics: average temperature, average humidity, air quality index, etc.

[0120] Feeding characteristics: daily feed amount, feeding frequency, etc.

[0121] Excretion characteristics: daily excretion volume, excretion frequency, etc.

[0122] Growth stage: The identifier of different growth stages (can be represented by a dummy variable).

[0123] S322. Set the number of categories of the group growth state as the number of clusters, set the original cluster centers and the randomly initialized membership matrix, and then define the fuzzy factor and the maximum number of iterations.

[0124] S323, calculating the membership matrix of the group growth status of each sample in the pig sample set; wherein the membership matrix calculation formula is:

[0125]

[0126] In the formula, u ij Representation sample x i Cluster Center v j The degree of membership; C The number of categories representing the growth status of the population; x i Indicates the first i samples; v j and v k Respectively represent j , k cluster centers; m represents the fuzzy factor.

[0127] S324, using the cluster center iteration formula to calculate the cluster center of the next iteration; wherein the cluster center iteration formula is:

[0128]

[0129] In the formula, N represents the total number of samples in the pig sample set;

[0130] S325, iterate until the membership matrix changes less than the set membership threshold or reaches the maximum number of iterations, stop iterating, obtain the membership of each sample, and determine the cluster to which each sample belongs.

[0131] S326. Based on the cluster to which the sample belongs, the group growth status of the pig herd in the pen area during the current measurement period is obtained, and the group growth index of the group growth status is calculated.

[0132] Among them, in fuzzy C-means clustering (FCM), sample refers to the basic unit for cluster analysis. In the pig farming scenario, each sample can represent an instance of multimodal information collected within a certain period of time. This information includes environmental information, feeding information, excretion information, and growth stage information. The membership degree indicates the degree to which a sample belongs to a cluster. In FCM, each sample can belong to multiple clusters at the same time, and each cluster corresponds to a membership value ranging from 0 to 1. The sum of the memberships of all clusters is 1. The higher the membership, the more the data point belongs to the cluster. Through the FCM algorithm (pig herd state model), the group growth status of the pig herd in the pen area can be clustered based on multimodal information, and groups with different growth states can be identified to provide scientific breeding management decision support. This helps to improve breeding efficiency, optimize breeding strategies, and ensure the health and growth of the pig herd.

[0133] In the description of the present invention, based on the cluster to which the sample belongs, the group growth status of the pig herd in the pen area in the current measurement period is obtained, and the group growth index of the group growth status is calculated, including:

[0134] S3261. If all samples belong to the same cluster, the population growth state corresponding to the cluster is used as the population growth state of the enclosure area in the current measurement period. If there are samples belonging to different clusters, the average membership method is used to calculate the average membership of the samples to each cluster, and the cluster with the largest average membership is selected as the population growth state of the enclosure area in the current measurement period.

[0135] S3262. Obtain the ideal state index of the current group growth state of the pig herd, and calculate the degree of deviation between the standardized environmental information, feeding information and excretion information and the ideal state index, which is defined as ( o , p , q ); then calculate ( o , p , q ) and the coordinate origin (0, 0, 0), and the distance value is defined as the population growth index of the population growth state, where the calculation formula of the population growth index is:

[0136]

[0137] In the formula, d express( o , p , q ) and the coordinate origin (0, 0, 0); o Indicates the degree of deviation between environmental information and ideal state indicators; pIndicates the degree of deviation between feeding information and ideal state indicators; q Indicates the degree of deviation between excretion information and ideal state indicators.

[0138] Specifically, it is necessary to define various indicators under ideal conditions, which are the best conditions defined by scientific research or industry standards. The purpose of standardization is to convert data of different dimensions to the same scale. Commonly used standardization methods include Z-score standardization and Min-Max standardization. The degree of deviation between the standardized data and the ideal state indicators can be expressed by absolute error. The absolute error between each actual information data and the ideal state indicator is calculated as the degree of deviation between the data and the ideal indicator. Degree of deviation ( o , p , q ) represents the deviation between the actual state and the ideal state during the current measurement cycle.

[0139] The individual analysis unit 303 is used to record the activity trajectory of individual pigs based on the pig's vital sign parameters and monitoring information, and to construct an individual state model to analyze the individual growth state of the individual pigs.

[0140] In the description of the present invention, based on the pig's vital signs parameters and monitoring information, the activity trajectory of individual pigs is recorded, and an individual state model is constructed to analyze the individual growth state of individual pigs, including:

[0141] S331. The location information of individual pigs is obtained through the smart ear tags worn by individual pigs, and the distinguishing marks of the individual tags are used to record the activity trajectory of each individual pig in the enclosure area and the measurement period.

[0142] S332, obtain a set of thermal infrared images in the enclosure area taken by an infrared camera, and classify the thermal infrared images according to the excretion posture and excrement form of the pigs, and mark them as excretion behavior. Classify the images according to the feeding posture and feed form of the pigs, and mark them as feeding behavior. Classify the images according to the sleeping posture and sleeping heart rate of the pigs, and mark them as sleeping behavior.

[0143] S333. Synchronize the timestamps between the thermal infrared image and the position information recorded by the smart ear tag, draw the enclosure area as a plane coordinate system, define the position information of each individual pig as a coordinate point, and map the coordinate point position information to the thermal infrared image, and annotate the individual labels of the individual pigs in the thermal infrared image in real time.

[0144] S334. Based on the labeling results of the thermal infrared images, an individual state model with an efficient neural network as the backbone network is established, and a weighted bidirectional feature fusion network is used as the feature network. The activity behaviors of the labeled individual pigs are extracted from the real-time collected thermal infrared images, and the number and duration of the activity behaviors are recorded.

[0145] Specifically, the present invention uses an efficient target detection model (EfficientDet) to perform the detection function of individual pig activity behaviors. EfficientDet is an efficient and accurate target detection model, a deep learning model for solving target detection tasks. EfficientDet is based on the EfficientNet (efficient neural network) architecture and uses a series of optimization techniques to achieve efficient and accurate target detection.

[0146] EfficientDet uses EfficientNet as its backbone network, fuses feature maps through a technology called BiFPN (Bi-directional Feature Pyramid Network), and introduces a compound scaling strategy in network design. These technologies enable EfficientDet to greatly reduce the consumption of computing resources while maintaining high accuracy.

[0147] The main features of EfficientDet include the following aspects.

[0148] 1. Efficient backbone network: Using EfficientNet as the backbone network improves computing efficiency.

[0149] 2. BiFPN (Weighted Bidirectional Feature Fusion Network): A bidirectional feature pyramid network that can better fuse features at different levels and improve detection performance.

[0150] 3. Compound scaling: Simultaneously scale the depth, width, and resolution of the network to achieve the best balance between performance and efficiency.

[0151] 4. Flexibility: It can handle target objects of various sizes, and has good detection effects from small objects to large objects.

[0152] S335. Count the activity behavior timetable of each individual pig in each measurement period, set a standard behavior timetable according to the growth stage of the pigs, and analyze the individual growth status of each individual pig in the measurement period by comparing the activity behavior timetable with the standard behavior timetable.

[0153] Specifically, according to the different growth stages of pigs (such as lactation, weaning, fattening, etc.), corresponding standard behavior schedules are set. These standard schedules can be formulated based on expert experience, literature research and historical data, including standard eating time, excretion frequency, sleep time, etc. for pigs at different stages. The standard behavior schedules for different growth stages are stored in the system's database for subsequent comparison and analysis.

[0154] The actual behavior schedule of each individual pig is then compared with the standard behavior schedule for its growth stage. Algorithms can be used to compare the behavior type, frequency and duration, and analyze the deviation between the actual and the standard. Based on the comparison results, the growth status of each individual pig is evaluated.

[0155] The specific assessment is done by setting thresholds or using a scoring mechanism. For example, short eating time and abnormal excretion frequency can be considered abnormal. The growth status of individual pigs is divided into normal, sub-healthy, abnormal and other categories, and corresponding reports are generated based on the assessment results. If the deviation of the activity behavior of an individual pig exceeds the set threshold, the system will automatically trigger an abnormal warning and notify the breeder to intervene.

[0156] The market evaluation unit 304 is used to set the purchase cycle, analyze the cost trend of various types of breeding materials in real time, predict the material demand in the next purchase cycle, and define market information.

[0157] In the description of the present invention, a purchasing cycle is set, the cost trend of various types of breeding materials is analyzed in real time, and the material input amount of the next purchasing cycle is predicted. The market information is defined to include:

[0158] S341. Set the purchase cycle of breeding materials for the piggery, obtain the real-time price data of breeding materials through the data interface, and record the purchase information of the historical purchase cycle. Then use the time series analysis method to draw the future cost trend of breeding materials and predict the purchase cost of the purchase cycle.

[0159] Among them, the purchasing cycle and the weighing cycle are two independent time periods, both of which can use months, weeks or days as cycle units. For example, if the month is used as the unit, the purchasing cycle requires monthly material purchases, while the weighing cycle weighs the pig herd and individual pigs every month. The two are independent and do not interfere with each other. As the most basic observation unit, the measurement cycle is smaller than the purchasing cycle and the weighing cycle. For example, when the purchasing cycle and the weighing cycle are in months or quarters, the measurement cycle usually uses weeks or days as units to calculate the growth status of the pig herd every day to reduce the observation cost while ensuring the detection accuracy.

[0160] S342. Based on the group growth status of the pigs in the enclosure area and combined with historical growth data, predict the material demand of the pigs in the next procurement cycle and calculate the total material demand of all enclosure areas.

[0161] S343. Obtain the maximum storage capacity and the cycle surplus of the material warehouse in the pig breeding house, take the difference between the maximum storage capacity and the cycle surplus as the upper limit of the material input, and then take the difference between the total material demand and the cycle surplus as the lower limit of the material input, and define the value range of the material input in the next procurement cycle.

[0162] S344. Integrate the purchase cost and total material demand of the next purchase cycle and define it as market information.

[0163] The strategy matching unit 305 is used to generate personalized breeding strategies based on the status analysis results of the pig herd and individual pigs, combined with market information, and matching decision information from the historical database.

[0164] In the description of the present invention, based on the status analysis results of the pig herd and individual pigs, combined with market information, matching decision information from the historical database, generating personalized breeding strategies includes:

[0165] S351. Obtain the analysis results of the group growth status of the pig herd and the individual growth status of the individual pigs in all the penned areas within the current metering period, identify whether there are any abnormal conditions in each penned area, and if so, capture the location of the abnormal pig herd or abnormal individual pigs and issue an emergency response strategy; if not, execute conventional strategy decisions.

[0166] S352. Define the breeding strategy decision for the next metering period as the target strategy, and integrate the multimodal information and market information in each enclosure area to define the strategy attributes of the target strategy.

[0167] S353, connecting to the historical database, calculating the attribute distance between the corresponding strategy attributes of the target strategy and the historical case strategy, wherein the calculation formula of the attribute distance is:

[0168]

[0169] In the formula, D ( Q ik , Q xk ) indicates the target strategy C i Case Studies and Strategies C x The attribute distance between Q xk The target strategyk Item policy attributes; Q ik Indicates i The historical case strategy k The policy attribute corresponding to k Item policy attributes; G ik Indicates the target strategy k The value of the item policy attribute; G xk Represents historical case strategy k The value of a policy attribute.

[0170] S354: Calculate the similarity between the target strategy and the historical case strategy, where the similarity calculation formula is:

[0171]

[0172] In the formula, S ( C i , C x ) indicates the target strategy C i Case Studies and Strategies C x The similarity between ω k Indicates the weight value of each policy attribute; N x Indicates the number of policy attributes.

[0173] S355, setting a similarity threshold, traversing the historical database, screening historical case strategies that meet the similarity threshold, and establishing a historical strategy set; and using a simulated annealing algorithm, selecting the optimal reuse strategy from the historical strategy set as the personalized breeding strategy for each enclosure area.

[0174] Specifically, the simulated annealing algorithm (SA) is a random search algorithm for global optimization. It simulates the physical annealing process, the process in which an object gradually cools down from high temperature to low temperature and reaches thermal equilibrium, and finds the optimal solution to the problem by gradually lowering the "temperature" of the search space; after screening the historical strategies that meet the similarity threshold in the historical database, the historical strategy set may contain multiple potential reuse strategies. The simulated annealing algorithm is used to select the best strategy from these candidate strategies and optimize the personalized farming strategy to make it more consistent with the attributes and requirements of the current target strategy.

[0175] Using the simulated annealing algorithm, the optimal reuse strategy is selected from the historical strategy set as the personalized breeding strategy for each enclosure area, including the following aspects:

[0176] 1. Initialization

[0177] Initial solution: Randomly select a strategy from the historical strategy set as the initial solution.

[0178] Initial Temperature: Set a higher initial temperature.

[0179] Cooling rate: Set the cooling rate, usually between 0.8 and 0.99.

[0180] 2. Iteration process

[0181] Neighborhood search: randomly select a new solution in the neighborhood of the current solution (in this case, it can be another strategy in the collection of historical strategies).

[0182] Evaluation function: Calculate the evaluation function of the current solution and the new solution (usually a comprehensive score of the policy attributes, considering the matching degree between the target policy attributes and the actual policy attributes).

[0183] Acceptance criteria: If the evaluation function of the new solution is better than the current solution, the new solution is accepted; otherwise, the new solution is accepted with a certain probability, which is related to the current temperature and the difference between the evaluation function.

[0184] Cooling: Updates the temperature according to the cooling rate.

[0185] 3. Termination conditions

[0186] The iteration is terminated when the preset number of times is reached or the temperature is lower than the threshold, and the current optimal solution is output as the reuse strategy.

[0187] Through the simulated annealing algorithm, the optimal strategy is finally selected as the personalized breeding strategy to achieve precise and optimized breeding management.

[0188] The self-learning optimization unit 306 is used to adaptively update the status information of the pig herd and individual pigs in the enclosure area and the personalized breeding strategy corresponding to the enclosure area based on the metering period, and use the new status information and breeding strategy as new knowledge to optimize the local model parameters of the intelligent body.

[0189] Specifically, at the end of each metering cycle, the latest multimodal information, including pig vital signs parameters, environmental information, feeding information, excretion information and monitoring information, is obtained from the information monitoring module 1. By analyzing the pre-processed multimodal information, the status information of the pig herd and individual pigs is updated; and the new status information is stored in the cloud database to maintain data continuity and historical records.

[0190] The intelligent agent communication unit 307 is used to establish data sharing and information synchronization with intelligent agents in other enclosure areas, and to support collaborative decision-making and global resource optimization among multiple intelligent agents.

[0191] Specifically, a communication network based on the Internet of Things (IoT) needs to be established between multiple agents to ensure that data between agents in each enclosure area can be transmitted quickly and securely. Distributed computing and federated learning technologies are used to support collaborative decision-making among multiple agents; and through data sharing, the status information of each enclosure area is integrated to optimize the global breeding strategy.

[0192] The management and control module 4 is used to integrate the management of all enclosure areas, dynamically and visually display the breeding information of individual pigs in the breeding pig house, and implement personalized breeding strategies and regulate breeding parameters.

[0193] Specifically, the management and control module 4 integrates the data of the information monitoring module 1, the abnormal warning module 2 and the analysis and decision-making module 3, and summarizes the multimodal information, status analysis results and personalized breeding strategies into a unified platform for management, so as to realize the classified storage and hierarchical management of data and ensure the integrity and traceability of data.

[0194] Using data visualization technology, the breeding information of individual pigs in the enclosure area (such as physical data, activity trajectory, feed consumption, excretion, etc.) is dynamically displayed in the form of charts, dashboards, etc. A real-time monitoring interface is provided to display the environmental parameters of each enclosure area (such as temperature, humidity, air quality, etc.) and the health status of pigs.

[0195] According to the personalized breeding strategy generated by the analysis and decision-making module, feed feeding, environmental control and other parameters are adjusted. Automatic execution strategy adjustment, such as feed formula adjustment, temperature adjustment, ventilation control, etc., ensures that pigs grow in the best environment. Achieve precise control of breeding parameters, and adjust various parameters of the enclosure area in real time through IoT devices (such as smart feeders, environmental sensors, etc.). Provide manual control and automatic control modes, allowing administrators to adjust parameters according to actual needs. Combined with the warning notification of the abnormal warning module, respond to abnormal situations in the breeding process in real time. Provide emergency treatment solutions and quickly adjust relevant parameters to ensure the health of pigs and the stability of the breeding process.

[0196] The intelligent collaboration module 5 is used to manage the communication between the edge computing device and the cloud computing platform, and to distribute the computing tasks of the intelligent agent to the cloud computing platform to maintain the autonomous learning and collaborative work of the intelligent agent.

[0197] In the description of the present invention, the intelligent collaboration module 5 includes: a communication management unit 501 , a task allocation unit 502 , a synchronous learning unit 503 and a collaborative work unit 504 .

[0198] The communication management unit 501 is used to manage the data transmission between the edge computing device and the cloud computing platform using the message queue telemetry transmission protocol. At the same time, a lightweight machine learning model is deployed on the edge computing device to handle the data preprocessing and decision-making tasks of multi-modal information.

[0199] Specifically, the communication management unit 501 realizes low-latency and high-efficiency data transmission, which is suitable for resource-constrained edge devices, deploys optimized machine learning models on edge devices, processes real-time data, and reduces cloud computing load.

[0200] The task allocation unit 502 is used to dynamically adjust the computing tasks based on the availability of computing resources and data transmission delay using a load balancing method and allocate them to each edge computing device.

[0201] The synchronous learning unit 503 is used to deploy a federated learning framework, support collaborative training of multiple agents, and regularly aggregate local models of agents to update to the cloud computing platform to dynamically adjust model parameters.

[0202] In the description of the present invention, a federated learning framework is deployed to support collaborative training of multiple agents, and local models of agents are aggregated and updated to the cloud computing platform regularly. Dynamic adjustment of model parameters includes:

[0203] S531. Deploy the federated learning framework on the cloud computing platform, generate an initial global model, and distribute it to each intelligent agent. Each intelligent agent receives the initial global model as the initial version of the local model.

[0204] S532. The intelligent agent uses the local model to process the multimodal information in the corresponding enclosure area, obtains new knowledge after the processing is completed, updates the local model parameters, and then uploads it to the cloud computing platform.

[0205] S533. The cloud computing platform receives the model parameters uploaded by each intelligent agent, generates new global model parameters using an aggregation algorithm, updates the global model, distributes it to each intelligent agent, and optimizes the local model.

[0206] The initial global model is an initial version of a machine learning model created on a cloud computing platform, which includes the initial model architecture and parameter settings. The model has not been trained or has only been preliminarily trained, and is used to distribute to various agents under the federated learning framework. The initial global model provides a unified starting point, allowing all agents to share the same model structure and initial parameters during the collaborative training process, thereby ensuring that model parameters from different agents can be effectively aggregated in subsequent training.

[0207] The global model is a comprehensive model obtained by aggregating the local model parameters uploaded by each agent in the federated learning framework. The global model represents the training results of all agents and is the final model formed after multiple rounds of training and parameter aggregation. The global model integrates the local model parameters from different agents and integrates the multimodal information and training data of each captive area to form a more comprehensive and accurate model. This model will be updated periodically and distributed to each agent to further optimize the local model.

[0208] The local model is a model that each agent uses to train and update in its own enclosure area using local data (corresponding multimodal information) and the initial global model. Specifically, the local model includes the models used by the herd analysis unit, individual analysis unit, market evaluation unit, and strategy matching unit in the analysis and decision-making module. At the agent level, the local model is trained and optimized based on real-time multimodal information in the enclosure area. The agent uses the local model to process local data, identify the growth status of the herd and individual pigs, and formulate corresponding breeding strategies. After each round of training, the agent uploads the updated local model parameters to the cloud computing platform for aggregation and updating of the global model.

[0209] In summary, the initial global model provides a unified model structure and initial parameters as the starting point for all agents. The global model is a comprehensive model formed by aggregating the local model parameters of each agent, representing the overall training results, and is updated and distributed periodically. The local model is trained and optimized by each agent based on local data in its own enclosure area, and is used to process the local information of the agent corresponding to the enclosure area and formulate personalized breeding strategies.

[0210] The collaborative working unit 504 is used to create a digital twin model of the pig house, establish real-time synchronization and interaction between the virtual environment and the actual environment, synchronize multimodal information, and use digital twin technology to simulate and predict the humanized breeding strategy to further optimize the breeding strategy and management plan.

[0211] Specifically, professional 3D modeling software (such as Blender, Unity3D) is used to create a virtual environment for the pig house, and sensor data in the actual environment is integrated into the digital twin model to achieve real-time data update and synchronization, and then the real-time collected data is transmitted to the digital twin model to ensure that the digital model reflects the current actual breeding situation. Mathematical modeling and simulation technology are used to simulate and predict breeding strategies, and breeding strategies are adjusted and optimized based on simulation results to improve production efficiency and animal welfare.

[0212] To sum up, with the help of the above-mentioned technical scheme of the present invention, the intelligent management of the whole process of pig farming is realized through the comprehensive application of information monitoring, abnormal warning, analysis and decision-making, management and control, and intelligent collaboration modules; first, by real-time collection of multimodal information in the confined area, the growth status of the pig herd is fully grasped; through the local computing resources of the edge computing device, the information data is analyzed in real time, and abnormal situations are quickly identified and warned to ensure timely handling of potential problems and avoid the development and spread of abnormal problems; and with the help of the cloud computing platform, the multimodal information is further analyzed in depth, the growth status of the pig herd and individuals is evaluated, and the personalized breeding strategy is optimized in combination with market dynamics and historical data to ensure that pigs at each growth stage can obtain the best breeding plan; thereby greatly improving the refinement and scientific level of breeding management, reducing labor costs and resource waste, improving the health and production efficiency of pigs, and ensuring food safety. Through integrated management and dynamic visualization, the system can fully control various information in the breeding process, adjust breeding parameters in a timely manner, and optimize resource allocation. The intelligent collaboration module ensures efficient and reliable data transmission between edge computing devices and cloud computing platforms, and through the federated learning framework, it realizes collaborative training and autonomous learning of multiple intelligent agents, improves the overall intelligence level and decision-making ability of the system, and promotes the breeding industry to move towards intelligent, modern and sustainable development.

[0213] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent decision-making management system for pig farming based on multimodal information monitoring, characterized in that: The intelligent decision-making management system includes: information monitoring module, abnormal warning module, analysis and decision module, management and control module and intelligent collaboration module, among which: The information monitoring module is used to divide the pig house into several enclosure areas through enclosure fences, and use the monitoring equipment configured in each enclosure area to monitor the multimodal information of the pig herd growth process in the enclosure area in real time; The abnormal warning module is used to monitor the multimodal information and information transmission path collected in real time through the edge computing device deployed inside the pig house, using local computing resources, identify abnormal phenomena in the information monitoring and transmission process, and trigger an early warning notification after the abnormal phenomenon is found; The analysis and decision-making module is used to define each enclosure area as an intelligent entity, and use the cloud computing resources of the cloud computing platform to analyze the group growth status of the pig herd in the enclosure area, as well as the activity trajectory and individual growth status of the individual pigs in the pig herd based on the multimodal information; and optimize the personalized breeding strategy based on the status analysis results, combined with market information and historical data; The management and control module is used to integrate and manage all enclosure areas, dynamically and visually display the breeding information of individual pigs in the breeding pig house, and implement the personalized breeding strategy to regulate breeding parameters; The intelligent collaboration module is used to manage the communication between the edge computing device and the cloud computing platform, and to distribute the computing tasks of the intelligent agent to the cloud computing platform, so as to maintain the autonomous learning and collaborative work of the intelligent agent; The analysis and decision-making module includes: a storage management unit, a pig group analysis unit, an individual analysis unit, a market evaluation unit, a strategy matching unit, a self-learning optimization unit and an intelligent agent communication unit, wherein: The storage management unit is used to receive the multimodal information after abnormality identification, upload it to the cloud computing platform, store it in the cloud database, and perform partition management according to data type and time series; at the same time, store historical growth data, historical multimodal information, historical strategy information and decision rule library; The pig herd analysis unit is used to construct a pig herd state model based on environmental information, feeding information, excretion information and the growth stage of the pig herd, and analyze the group growth state of the pig herd within the measurement period; The individual analysis unit is used to record the activity trajectory of individual pigs based on the pig's vital sign parameters and monitoring information, and to construct an individual state model to analyze the individual growth state of the individual pigs; The market evaluation unit is used to set the purchase cycle, analyze the cost trend of various types of breeding materials in real time, and predict the material demand in the next purchase cycle and define market information; The strategy matching unit is used to match decision information from the historical database based on the status analysis results of the pig herd and individual pigs, combined with market information, to generate a personalized breeding strategy; The self-learning optimization unit is used to adaptively update the status information of the pig herd and individual pigs in the enclosure area and the personalized breeding strategy corresponding to the enclosure area with the metering period as the time unit, and optimize the local model parameters of the intelligent body by taking the new status information and breeding strategy as new knowledge; The agent communication unit is used to establish data sharing and information synchronization with agents in other enclosure areas, and to support collaborative decision-making and global resource optimization among multiple agents; The pig herd status model is constructed based on the environmental information, feeding information, excretion information and the growth stage of the pig herd, and the group growth status of the pig herd within the measurement period is analyzed, including: A weighing cycle is set to regularly measure the weight of each individual pig in the pen area, obtain the growth stage identifier of the pig herd, and add up the individual weights of each individual pig as the herd weight information of the pig herd in the pen area; wherein the weighing cycle includes multiple measurement cycles; The environmental information, feeding information, excretion information, pig weight information and growth stage identification are standardized, and effective features are extracted to construct feature vectors to form a pig sample set; Set the number of categories of the population growth state as the number of clusters, set the original cluster center and the randomly initialized membership matrix, and then define the fuzzy factor and the maximum number of iterations; Calculate the membership matrix of the group growth status of each sample in the pig sample set; the membership matrix calculation formula is: In the formula, u ij Representation sample x i Cluster Center v j The degree of membership; C The number of categories representing the growth status of the population; x i Indicates the first i samples; v j and v k Respectively represent j , k Cluster centers; m represents the fuzzy factor; The cluster center iteration formula is used to calculate the cluster center of the next iteration; the cluster center iteration formula is: In the formula, N Indicates the total number of samples in the pig herd sample set; Iterate until the membership matrix changes less than the set membership threshold or reaches the maximum number of iterations, then stop iterating, obtain the membership of each sample, and determine the cluster to which each sample belongs; Based on the cluster to which the sample belongs, the group growth status of the pig herd in the pen area during the current measurement period is obtained, and the group growth index of the group growth status is calculated; The method of obtaining the group growth status of the pig herd in the pen area in the current measurement period based on the cluster to which the sample belongs, and calculating the group growth index of the group growth status includes: If all samples belong to the same cluster, the population growth state corresponding to the cluster is used as the population growth state of the enclosure area in the current measurement period; if there are samples belonging to different clusters, the average membership method is used to calculate the average membership of the samples to each cluster, and the cluster with the largest average membership is selected as the population growth state of the enclosure area in the current measurement period; Obtain the ideal state index of the current group growth state of the pig herd, and calculate the deviation between the standardized environmental information, feeding information and excretion information and the ideal state index, which is defined as ( o , p , q ); then calculate ( o , p , q ) and the coordinate origin (0, 0, 0), and the distance value is defined as the population growth index of the population growth state, where the calculation formula of the population growth index is: In the formula, d express( o , p , q ) and the coordinate origin (0, 0, 0); o Indicates the degree of deviation between environmental information and ideal state indicators; p Indicates the degree of deviation between feeding information and ideal state indicators; q Indicates the degree of deviation between excretion information and ideal state indicators; The method of recording the activity trajectory of individual pigs based on the pig vital signs parameters and monitoring information, and constructing an individual state model to analyze the individual growth state of individual pigs includes: The location information of individual pigs is obtained through the smart ear tags worn by individual pigs, and the distinguishing marks of individual tags are used to record the activity trajectory of each individual pig in the enclosure area and measurement period; Obtain a set of thermal infrared images of the enclosure area taken by an infrared camera. In the thermal infrared images, classify the pigs according to their excretion posture and excrement form, and mark them as excretion behavior; classify the pigs according to their eating posture and feed form, and mark them as eating behavior; classify the pigs according to their sleeping posture and sleeping heart rate, and mark them as sleeping behavior; Synchronize the timestamps between the thermal infrared image and the location information recorded by the smart ear tag, draw the enclosure area as a plane coordinate system, define the location information of each individual pig as a coordinate point, and map the coordinate point location information to the thermal infrared image, and annotate the individual labels of the individual pigs in the thermal infrared image in real time; Based on the annotation results of thermal infrared images, an individual state model with an efficient neural network as the backbone network is established, and a weighted bidirectional feature fusion network is used as the feature network; the activity behaviors of the annotated individual pigs are extracted from the thermal infrared images collected in real time, and the number and duration of the activity behaviors are recorded; Count the activity behavior timetable of each individual pig in each measurement period, set a standard behavior timetable according to the growth stage of the pig, and analyze the individual growth status of each individual pig in the measurement period by comparing the activity behavior timetable with the standard behavior timetable; The above-mentioned personalized breeding strategy is generated based on the status analysis results of the pig herd and individual pigs, combined with market information, matching decision information from the historical database, and comprising: Obtain the analysis results of the group growth status of the pig herds and the individual growth status of the pigs in all the enclosures in the current measurement period, identify whether there are abnormal conditions in each enclosure area, if there are abnormal conditions, capture the location of the abnormal pig herds or abnormal pigs, and issue emergency response strategies, if there are no abnormal conditions, execute conventional strategy decisions; Define the breeding strategy decision of the next metering period as the target strategy, and integrate the multimodal information and market information in each enclosure area to define it as the strategy attribute of the target strategy; Connect to the historical database and calculate the attribute distance between the corresponding strategy attributes of the target strategy and the historical case strategy, where the calculation formula of the attribute distance is: In the formula, D ( Q ik , Q xk ) indicates the target strategy C i Strategy with historical cases C x The attribute distance between Q xk The target strategy k Item policy attributes; Q ik Indicates i The historical case strategy k The policy attribute corresponding to k Item policy attributes; G ik Indicates the target strategy k The value of the item policy attribute; G xk Represents historical case strategy k The value of the item policy attribute; Calculate the similarity between the target strategy and the historical case strategy, where the similarity calculation formula is: In the formula, S ( C i , C x ) indicates the target strategy C i Strategy with historical cases C x The similarity between ω k Indicates the weight value of each policy attribute; N x Indicates the number of policy attributes; A similarity threshold is set, the historical database is traversed, historical case strategies that meet the similarity threshold are screened, and a historical strategy set is established; and the simulated annealing algorithm is used to select the optimal reuse strategy from the historical strategy set as the personalized breeding strategy for each captive area.

2. According to claim 1, the intelligent decision-making management system for pig farming based on multimodal information monitoring is characterized in that: The information monitoring module includes: an intelligent ear tag unit, an environmental monitoring unit, a feeding metering unit, an excretion metering unit, a video monitoring unit and a wireless transmission unit, wherein: The smart ear tag unit is used to collect the pig vital signs information of individual pigs in real time; The environmental monitoring unit is used to collect environmental information in the enclosure area in real time; The feeding metering unit is used to set a metering cycle, collect the amount of feed fed and the amount of remaining feed in the enclosure area within the metering cycle, and integrate and define the information as feeding information in combination with the feed type; The excretion metering unit is used to count the weight of excrement of the pigs in the pen area within the metering period, which is defined as excretion information; The video monitoring unit is used to monitor the pigs in the enclosure area using a camera and an infrared camera, respectively acquiring video images and infrared images, and defining them as monitoring information; The wireless transmission unit is used to establish a communication channel between the multimodal information and the edge computing device, and use wireless communication technology to realize the interactive communication of multimodal information; wherein the multimodal information includes pig vital signs information, environmental information, feeding information, excretion information and monitoring information.

3. According to claim 2, a pig farming intelligent decision-making management system based on multimodal information monitoring is characterized in that: The setting of the procurement cycle, real-time analysis of the cost trend of various types of aquaculture materials, and prediction of the material input in the next procurement cycle, defines the market information including: Set the purchase cycle of breeding materials for piggery, obtain real-time material price data of breeding materials through the data interface, and record the purchase information of historical purchase cycles; then use the time series analysis method to draw the future cost trend of breeding materials and predict the purchase cost of the purchase cycle; Based on the herd growth status of the pigs in the enclosure area and combined with historical growth data, the material demand of the pigs in the next procurement cycle is predicted, and the total material demand of all enclosure areas is counted; Get the maximum storage capacity and the remaining amount in the cycle of the material warehouse in the pig breeding house, take the difference between the maximum storage capacity and the remaining amount in the cycle as the upper limit of the material input, and then take the difference between the total material demand and the remaining amount in the cycle as the lower limit of the material input, and define the value range of the material input in the next procurement cycle; Integrate the purchase cost and total material requirement for the next purchase cycle and define it as market information.

4. The intelligent decision-making management system for pig farming based on multimodal information monitoring according to claim 1 is characterized in that: The intelligent collaboration module includes: a communication management unit, a task allocation unit, a synchronous learning unit and a collaborative work unit, wherein: The communication management unit is used to manage the data transmission between the edge computing device and the cloud computing platform using the message queue telemetry transmission protocol; at the same time, a lightweight machine learning model is deployed on the edge computing device to handle the data preprocessing and decision-making tasks of multimodal information; The task allocation unit is used to dynamically adjust the computing tasks based on the availability of computing resources and data transmission delay using a load balancing method and allocate them to each edge computing device; The synchronous learning unit is used to deploy a federated learning framework, support collaborative training of multiple agents, and regularly aggregate local models of agents to update to the cloud computing platform to dynamically adjust model parameters; The collaborative working unit is used to create a digital twin model of the pig house, establish real-time synchronization and interaction between the virtual environment and the actual environment, synchronize multimodal information, and use digital twin technology to simulate and predict humanized breeding strategies to further optimize breeding strategies and management plans.

5. The intelligent decision-making management system for pig farming based on multimodal information monitoring according to claim 4 is characterized in that: The deployment of the federated learning framework supports the collaborative training of multiple agents, and regularly aggregates the local models of the agents to update to the cloud computing platform, and dynamically adjusts the model parameters including: Deploy the federated learning framework on the cloud computing platform, generate an initial global model, and distribute it to each agent. Each agent receives the initial global model as the initial version of the local model. The agent uses the local model to process the multimodal information in the corresponding enclosure area, obtains new knowledge after processing, updates the local model parameters, and then uploads them to the cloud computing platform; The cloud computing platform receives the model parameters uploaded by each intelligent agent, generates new global model parameters using an aggregation algorithm, updates the global model, distributes it to each intelligent agent, and optimizes the local model.

Images