A ventilation control method, control system for an animal farming site

Abstract

The application relates to the technical field of animal breeding environment control, and discloses a ventilation control method and system for an animal breeding site. The method collects animal behavior data and environmental factor data in the breeding site in real time, constructs an animal group behavior network based on the animal behavior data, and identifies a group behavior mode by using a graph neural network. The environmental factor data and the group behavior mode are coupled by a nonlinear weighted fusion algorithm, a multi-factor linkage prediction model is constructed to obtain a group stress response, historical stress data is further combined, a space-time graph neural network is used to establish a group stress response propagation model in the group to obtain a stress propagation path, and the control parameters of at least one environmental control device in a ventilation system and a temperature and humidity control system are adjusted based on the stress propagation path. The application can realize ventilation control based on linkage of group behavior and environmental factors and is suitable for complex breeding scenes.

Description

Technical Field

[0001] This invention relates to the field of animal husbandry environment control technology, specifically to a ventilation control method and control system for animal husbandry facilities. Background Technology

[0002] In modern animal husbandry, stress response is a key factor affecting animal growth, immunity, and productivity. Stress responses typically manifest as elevated body temperature, rapid breathing, decreased appetite, and reduced activity, directly impacting animal growth, health, and production efficiency. Especially in high-density farming environments, the dynamic behavior of animal groups, environmental changes, and the complex interactions between stressors place higher demands on environmental control.

[0003] Currently, traditional stress management methods largely rely on real-time environmental monitoring and single responses to animal physiological states, which often have a certain lag. For example, ventilation and humidity control systems are usually adjusted based on fixed settings or environmental parameters (such as temperature and humidity), failing to fully consider the stress response characteristics of individual animals and groups. More importantly, traditional methods typically rely solely on physiological responses (such as body temperature and respiratory rate) to determine an animal's stress state, making it difficult to detect signs of group stress responses in a timely manner, and thus hindering the implementation of effective early intervention measures.

[0004] Therefore, how to accurately predict group stress responses by monitoring animal physiological and behavioral data in real time, and take timely and effective intervention measures before stress responses occur, in order to reduce the negative impact of stress on animal health and production efficiency, has become a technical problem that urgently needs to be solved in the current breeding industry. Summary of the Invention

[0005] In view of the technical problems mentioned in the background, the present invention provides a ventilation control method and control system for animal breeding sites.

[0006] This invention provides a ventilation control method for animal breeding facilities, comprising the following steps:

[0007] S1, real-time collection of animal data and environmental factor data within the breeding site; wherein, the animal data includes animal location, density, movement trajectory and interaction behavior data between animals, and the environmental factor data includes temperature, humidity, gas concentration and light intensity;

[0008] S2, construct an animal group behavior network based on the animal data, and use a graph neural network to model and analyze the animal group behavior network to identify group behavior patterns, including group aggregation, dispersion or abnormal behavior patterns;

[0009] S3, The environmental factor data and the group behavior pattern are coupled through a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model for predicting group stress response;

[0010] S4. Based on historical stress data, the environmental factor data, and the group behavior pattern, a spatiotemporal graph neural network is used to establish a propagation model of the group stress response in the group and predict the stress propagation path.

[0011] S5, based on the stress propagation path, adjust the control parameters of at least one environmental control device in the ventilation system and temperature and humidity control system of the breeding site in real time.

[0012] As an example, an animal group behavior network is constructed based on the animal data, and a graph neural network is used to model and analyze the animal group behavior network to identify group behavior patterns, including:

[0013] S21. Each animal in the breeding area is treated as a node in the graph structure, and edges representing the relationships between animals are constructed based on the spatial distance, relative movement state and interaction intensity between animals, thus forming an animal group behavior network.

[0014] S22, Based on the animal group behavior network, a graph neural network is used to perform feature propagation and aggregation operations on nodes and their associated edges to obtain node representations that reflect the individual animal behavior characteristics and their group association characteristics;

[0015] S23. Based on the changes represented by the nodes, the overall behavioral state of the animal group is analyzed to obtain a group behavior pattern that characterizes the aggregation, dispersion, or abnormal changes of the animal group.

[0016] As an example, the environmental factor data and the group behavior pattern are coupled using a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model, including:

[0017] S31, the environmental factor data and the group behavior pattern are respectively represented by features, and the temperature, humidity, gas concentration and light intensity are converted into environmental feature vectors, and the group gathering, dispersal or abnormal behavior patterns are converted into behavior feature vectors;

[0018] S32, Based on the environmental feature vector and the behavioral feature vector, a nonlinear weighting function is introduced to assign corresponding weights to different types of features, and the feature vectors are fused through nonlinear mapping to obtain a joint feature representation;

[0019] S33, Construct a multi-factor linkage prediction model based on the joint feature representation, and output the group stress prediction result corresponding to the joint feature representation.

[0020] As an example, based on historical stress data, environmental factor data, and group behavior patterns, a spatiotemporal graph neural network is used to establish a propagation model of the group's stress response within the group, predicting the stress propagation path, including:

[0021] S41, acquire historical stress data and associate it with the corresponding environmental factor data and the group behavior pattern in chronological order to form time-series sample data for describing the group stress change process;

[0022] S42, construct a spatiotemporal graph structure based on the time-series sample data, wherein the nodes in the graph structure are used to represent animals or animal groups, the node attributes are used to characterize the environmental factor characteristics and group behavior pattern characteristics at the corresponding time, and the edges between nodes are used to represent the spatial adjacency relationship or behavioral association relationship between animals.

[0023] S43, Based on the spatiotemporal graph structure, a spatiotemporal graph neural network is used to model the node attributes and their temporal correlations, and output the stress propagation path that represents the propagation relationship of the group stress response in the time and space dimensions.

[0024] As an example, based on the aforementioned spatiotemporal graph structure, a spatiotemporal graph neural network is used to model node attributes and their temporal relationships, outputting stress propagation paths that characterize the propagation relationships of group stress responses in the temporal and spatial dimensions, including:

[0025] S431, In the spatiotemporal graph structure, a temporal association weight that changes with time is introduced for the edges between each node. The temporal association weight is dynamically updated according to the degree of historical stress association between animals corresponding to adjacent nodes, the trend of spatial distance change, and the similarity of group behavior patterns, so as to form a temporally weighted adjacency relationship.

[0026] S432, Based on the time-weighted adjacency relationship, the information propagation process in the spatiotemporal graph neural network is constrained so that the propagation direction of node features in the time dimension is consistent with the direction of historical stress state change, thereby forming a spatiotemporal feature transmission path with propagation directionality;

[0027] S433, in the spatiotemporal graph neural network, the node attributes are modeled in parallel at different time scales, and the stress state change features in the short time scale and the stress evolution trend features in the long time scale are extracted respectively, and the features at different time scales are fused.

[0028] S434, based on the fused node features, determines the propagation order and propagation correlation of the group stress response between spatial nodes and in the time series, and outputs the stress propagation path to characterize the propagation process of the group stress response in the group.

[0029] The present invention also provides a ventilation control system for animal breeding facilities, the system comprising:

[0030] The data acquisition module is used to collect animal data and environmental factor data in the breeding site in real time; wherein, the animal data includes animal location, density, movement trajectory and interaction behavior data between animals, and the environmental factor data includes temperature, humidity, gas concentration and light intensity;

[0031] The group behavior modeling module is used to construct an animal group behavior network based on the animal data, and to use a graph neural network to model and analyze the animal group behavior network to identify group behavior patterns, including group aggregation, dispersion or abnormal behavior patterns.

[0032] The multi-factor fusion prediction module is used to couple the environmental factor data with the group behavior pattern through a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model for predicting group stress response.

[0033] The stress propagation modeling module is used to establish a propagation model of the group stress response in the group based on historical stress data, the environmental factor data and the group behavior pattern, and to predict the stress propagation path.

[0034] An environmental control module is used to adjust the control parameters of at least one environmental control device in the ventilation system and temperature and humidity control system of the breeding site in real time based on the stress propagation path.

[0035] As an example, the group behavior modeling module includes:

[0036] The behavioral network construction unit is used to treat each animal in the breeding site as a node in a graph structure, and constructs edges representing the relationships between animals based on the spatial distance, relative movement state and interaction intensity between animals, thereby forming an animal group behavioral network.

[0037] The behavior feature propagation unit is used to perform feature propagation and aggregation operations on nodes and their associated edges based on the animal group behavior network using a graph neural network, so as to obtain node representations that reflect the individual animal behavior features and their group association features;

[0038] The behavior pattern recognition unit is used to analyze the overall behavior state of the animal group based on the changes represented by the nodes, and obtain group behavior patterns that characterize the animal group's aggregation, dispersion, or abnormal changes.

[0039] As an example, the multi-factor fusion prediction module includes:

[0040] The feature representation unit is used to represent the environmental factor data and the group behavior pattern respectively, converting temperature, humidity, gas concentration and light intensity into environmental feature vectors, and converting group aggregation, dispersion or abnormal behavior patterns into behavioral feature vectors.

[0041] The nonlinear fusion unit is used to introduce a nonlinear weighting function to assign corresponding weights to different types of features based on the environmental feature vector and the behavioral feature vector, and to perform fusion processing on the feature vector through nonlinear mapping to obtain a joint feature representation;

[0042] The linkage prediction unit is used to construct a multi-factor linkage prediction model based on the joint feature representation and output the group stress prediction result corresponding to the joint feature representation.

[0043] As an example, the stress propagation modeling module includes:

[0044] A time-series sample construction unit is used to acquire historical stress data and associate it with the corresponding environmental factor data and the group behavior pattern in chronological order to form time-series sample data for describing the group stress change process.

[0045] The spatiotemporal graph construction unit is used to construct a spatiotemporal graph structure based on the time-series sample data. In the graph structure, nodes are used to represent animals or animal groups, node attributes are used to characterize the environmental factors and group behavior patterns at the corresponding time, and edges between nodes are used to represent the spatial adjacency or behavioral association between animals.

[0046] The propagation path prediction unit is used to model the node attributes and their temporal correlations based on the spatiotemporal graph structure using a spatiotemporal graph neural network, and output the stress propagation path that represents the propagation relationship of the group stress response in the time and space dimensions.

[0047] As an example, the propagation path prediction unit includes:

[0048] The temporal adjacency modulation unit is used to introduce temporal association weights that change with time to the edges between nodes in the spatiotemporal graph structure. The temporal association weights are dynamically updated according to the degree of historical stress association between animals corresponding to adjacent nodes, the trend of spatial distance change, and the similarity of group behavior patterns, so as to form a temporally weighted adjacency relationship.

[0049] The propagation direction constraint unit is used to constrain the information propagation process in the spatiotemporal graph neural network based on the time-weighted adjacency relationship, so that the propagation direction of node features in the time dimension is consistent with the historical stress state change direction, thereby forming a spatiotemporal feature transmission path with propagation directionality.

[0050] The multi-timescale modeling unit is used to perform parallel modeling of node attributes at different timescales in the spatiotemporal graph neural network, extract stress state change features in short timescales and stress evolution trend features in long timescales, and fuse the features at different timescales.

[0051] The propagation path generation unit is used to determine the propagation order and propagation correlation of the group stress response between spatial nodes and in the time series based on the fused node features, and outputs the stress propagation path to characterize the propagation process of the group stress response in the group.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] This invention introduces animal group behavior network modeling, multi-factor nonlinear weighted fusion, and spatiotemporal graph neural network propagation modeling mechanisms. This enables ventilation control in breeding farms to no longer rely solely on a single environmental parameter. Instead, it is based on a comprehensive analysis of animal group behavior, environmental factors, and their spatiotemporal correlations. This allows for the modeling and utilization of group stress responses and their propagation paths. Consequently, the control parameters of ventilation and related environmental control equipment can be adjusted in a targeted manner according to the spatial distribution and temporal evolution characteristics of the group state. This improves the matching degree between the breeding environment regulation process and changes in the group state, and enhances the adaptability and stability of the system in complex breeding scenarios. Attached Figure Description

[0054] Figure 1 This is a schematic flowchart of a ventilation control method for animal breeding facilities disclosed in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of a ventilation control system for animal breeding facilities disclosed in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of the structure of the group behavior modeling module disclosed in an embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of the structure of the multi-factor fusion prediction module disclosed in an embodiment of the present invention;

[0058] Figure 5 This is a schematic diagram of the stress propagation modeling module disclosed in an embodiment of the present invention;

[0059] Figure 6 This is a schematic diagram of the propagation path prediction unit disclosed in an embodiment of the present invention. Detailed Implementation

[0060] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0061] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0062] Please see Figure 1 This embodiment discloses a ventilation control method for animal breeding facilities. The method can be executed by a ventilation control system deployed on a local server, edge computing node, or cloud platform within the breeding facility. The ventilation control system establishes communication connections with ventilation equipment, temperature and humidity control equipment, and data acquisition devices within the breeding facility to acquire animal data and environmental factor data, and outputs control parameters based on this data. The method includes the following steps:

[0063] S1, real-time collection of animal data and environmental factor data within the breeding site; wherein, the animal data includes animal location, density, movement trajectory and interaction behavior data between animals, and the environmental factor data includes temperature, humidity, gas concentration and light intensity;

[0064] In this embodiment, the collection of animal data and environmental factor data needs to be time-aligned, that is, a set of animal data samples and a set of environmental factor data samples are formed within the same sampling period, and the same timestamp is attached to them. It is understood that animal data has significant time variability and population correlation: for example, an increase in population density at the same time may correspond to local crowding, chasing or gathering; and environmental factors (such as temperature, humidity, gas concentration) also change with ventilation conditions and animal activity.

[0065] In some optional implementations, animal locations can be represented as two-dimensional coordinates (e.g., using a plane coordinate system within the breeding shed), and movement trajectories can be represented as a sequence of locations over consecutive time intervals. Animal density can be obtained through zonal statistics, for example, by dividing the breeding shed into multiple spatial grids / regional units, counting the number of animals in each regional unit, and forming a density sequence. Animal interaction behavior data can include indicators characterizing the strength of the association between animals, such as the number of contacts, changes in relative distance, and the degree of synchronized movement. Environmental factor data can also be collected by regional unit or by the entire shed; when collected by regional unit, the temperature, humidity, gas concentration, and light intensity of each regional unit can form a spatial distribution vector.

[0066] In addition, basic preprocessing can be performed on the collected data, such as imputing missing values, removing or limiting obviously abnormal sensor readings, and normalizing features of different dimensions.

[0067] S2, construct an animal group behavior network based on the animal data, and use a graph neural network to model and analyze the animal group behavior network to identify group behavior patterns, including group aggregation, dispersion or abnormal behavior patterns;

[0068] This step transforms the animal behavior data collected over a continuous period into graph-structured data that reflects the relationships between individuals and the characteristics of the group structure. Feature propagation and aggregation operations are then performed on this graph structure to obtain computational results that characterize the group's behavioral state. Unlike analyses based solely on individual behavioral characteristics, this step explicitly models the relationships between animals, enabling the quantification of spatial structure and behavioral synergies at the group level.

[0069] As an example, an animal group behavior network is constructed based on the animal data, and a graph neural network is used to model and analyze the animal group behavior network to identify group behavior patterns, including:

[0070] S21. Each animal in the breeding area is treated as a node in the graph structure, and edges representing the relationships between animals are constructed based on the spatial distance, relative movement state and interaction intensity between animals, thus forming an animal group behavior network.

[0071] In this step, let's assume that at a certain sampling time t, a total of [number] samples were detected in the breeding area. For animals only, construct an undirected or directed graph. Among them, the node set Used to represent individual animals; edge set Used to indicate the relationships between individual animals.

[0072] For any two animals i and j, first calculate the Euclidean distance based on their spatial positions: .in, Let represent the two-dimensional position vector of animal i at time t. Less than the preset space threshold At that time, it was assumed that the two had a spatial adjacency relationship.

[0073] Simultaneously, relative motion states, such as relative velocity consistency, are calculated based on the animal's movement trajectory.

[0074]

[0075] in, This represents the velocity vector of animal i at time t; The closer to 1, the more consistent the directions of movement of the two animals.

[0076] In addition, it can also be based on the number of interactive behaviors within a unit time window. (For example, touching, chasing, or approaching behaviors) Calculate the interaction intensity index.

[0077] Taking into account the above factors, construct edge weights for nodes i and j: .in, These are preset weighting coefficients. If... Greater than the association threshold Then, an edge is established between node i and node j, thus obtaining an animal group behavior network that reflects the group structure and interaction relationships.

[0078] It is understandable that the animal group behavior network constructed in the above manner can simultaneously reflect the spatial structure characteristics and behavioral interaction characteristics of the group.

[0079] S22, Based on the animal group behavior network, a graph neural network is used to perform feature propagation and aggregation operations on nodes and their associated edges to obtain node representations that reflect the individual animal behavior characteristics and their group association characteristics;

[0080] In this step, for each node Constructing the initial node feature vector .in, This represents the local density of the spatial region where animal i is located.

[0081] Graph neural networks in the first The node feature update process of a layer can be represented as:

[0082]

[0083] in, Represents a node The set of neighboring nodes; and The learnable parameter matrix; It is a non-linear activation function.

[0084] Through the above propagation and aggregation operations, node representation... In each layer, behavioral information and association strength information from neighboring nodes are gradually fused. As the network stacks to the [number missing]th ... After layering, the final node representation is obtained: This final node represents the simultaneous encoding of an individual animal's behavioral state and its association with the group structure.

[0085] S23. Based on the changes represented by the nodes, the overall behavioral state of the animal group is analyzed to obtain a group behavior pattern that characterizes the aggregation, dispersion, or abnormal changes of the animal group.

[0086] In this step, the representations of all nodes are first aggregated at the group level, for example:

[0087]

[0088] in, Indicates time The group behavior embedding vector.

[0089] Additionally, the spatial discreteness of the node representation can be calculated:

[0090]

[0091] when When the average edge weight between nodes is relatively small and the average edge weight between nodes is relatively large, it indicates that the animal group exhibits a high degree of consistency in behavior and structure, which corresponds to a group aggregation pattern; when When the edge weights are large and the overall edge weights are low, it indicates that the animal group's behavior is dispersed; when or A significant shift occurs within a short time window, i.e. If so, it can be identified as an abnormal change pattern in group behavior.

[0092] S3, The environmental factor data and the group behavior pattern are coupled through a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model for predicting group stress response;

[0093] Environmental factor data and group behavior patterns differ significantly in data structure, statistical characteristics, and scale of change. Directly performing linear concatenation or simple weighting can easily lead to one type of feature dominating the prediction results, or cause the structural information contained in the group behavior patterns to be masked by environmental noise. Therefore, this embodiment introduces a nonlinear weighted fusion mechanism to couple environmental factors and group behavior patterns within a unified feature space, enabling both to participate in the modeling of group stress states in a dynamic and adjustable manner within the prediction model.

[0094] As an example, the environmental factor data and the group behavior pattern are coupled using a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model, including:

[0095] S31, the environmental factor data and the group behavior pattern are respectively represented by features, and the temperature, humidity, gas concentration and light intensity are converted into environmental feature vectors, and the group gathering, dispersal or abnormal behavior patterns are converted into behavior feature vectors;

[0096] The environmental factor data is vectorized. Let's assume that at time... Environmental factors collected from aquaculture sites include temperature. ,humidity gas concentration and light intensity Then an environmental feature vector can be constructed. .

[0097] In some alternative implementations, when environmental factors are collected by spatial region, the environmental feature vector can be expanded into a regional feature matrix. .in, Indicates the number of spatial region units. Indicates the first Each regional unit at time The environmental feature vector.

[0098] Simultaneously, the group behavior pattern output in step S2 is characterized. Let's assume that at time... If the identified group behavior pattern is a categorical result (clustering, dispersion, or anomaly), it can be represented using one-hot encoding or probability vector form. .in, , , These represent the probability values ​​of a group being in a clustered, dispersed, or abnormal behavior pattern, respectively.

[0099] In another alternative implementation, the group behavior pattern can also be directly adopted from the group-level embedding vector obtained in step S2. As behavioral feature vector .

[0100] S32, Based on the environmental feature vector and the behavioral feature vector, a nonlinear weighting function is introduced to assign corresponding weights to different types of features, and the feature vectors are fused through nonlinear mapping to obtain a joint feature representation;

[0101] In this step, nonlinear mappings are applied to the environmental feature vector and the behavioral feature vector respectively to eliminate the influence of different dimensions and distribution characteristics. Specifically, environmental feature mapping functions and behavioral feature mapping functions can be defined as follows:

[0102]

[0103] in, , It is a nonlinear mapping function, which can be implemented by a multilayer perceptron, a gated network, or other nonlinear functions.

[0104] Based on the current environmental state and the group behavior state, dynamic weights are assigned to two types of features. These weights can be calculated using a weighting function. .in, Represents the weight of environmental features. Represents the weights of behavioral features, and in some implementations, satisfies normalization constraints: .

[0105] Based on the above weights, the two types of features are fused to obtain a joint feature representation: .in, This represents a vector concatenation or weighted summation operation. In spatial region-level modeling scenarios, the fusion operation can also be performed separately on each spatial region unit to obtain a set of joint feature representations at the region level.

[0106] Through this nonlinear weighted fusion process, the contribution ratio of environmental factors and group behavior patterns in the joint features can be adaptively adjusted at different times and under different group conditions, so that the joint features can simultaneously retain information on environmental changes and group structure changes.

[0107] S33, Construct a multi-factor linkage prediction model based on the joint feature representation, and output the group stress prediction result corresponding to the joint feature representation.

[0108] In this step, the joint feature representation obtained in step S32 is used. As input, construct a multi-factor linkage prediction model To output the group stress prediction results: .in, Indicates time The corresponding group stress prediction results can be represented as stress probability values, stress levels, or stress state vectors.

[0109] In some optional implementations, the prediction model Multi-layer nonlinear mapping structures (such as MLPs) can be employed to enable models to learn the complex interactions between environmental factors and group behavior; in other implementations, the predictive model can also output node-level or region-level stress prediction results. .in, Indicates the first A spatial region or group unit at time The stress prediction results.

[0110] S4. Based on historical stress data, the environmental factor data, and the group behavior pattern, a spatiotemporal graph neural network is used to establish a propagation model of the group stress response in the group and predict the stress propagation path.

[0111] In high-density farming environments, stress responses often do not occur in isolation, but rather spread gradually along spatial adjacency relationships, behavioral correlations, and environmental gradients among individual animals. Analyzing stress states based solely on a single moment or region is insufficient to reflect the evolutionary process of stress within the population. Therefore, this embodiment introduces a spatiotemporal graph neural network to model historical and current stress states within a unified spatiotemporal graph structure, thereby obtaining stress propagation paths that can be used for subsequent control decisions.

[0112] As an example, based on historical stress data, environmental factor data, and group behavior patterns, a spatiotemporal graph neural network is used to establish a propagation model of the group's stress response within the group, predicting the stress propagation path, including:

[0113] S41, acquire historical stress data and associate it with the corresponding environmental factor data and the group behavior pattern in chronological order to form time-series sample data for describing the group stress change process;

[0114] In this step, stress data records generated within a historical time period are acquired. These records can originate from historical operational logs, historical prediction results, or historical annotation information, and can be in the form of stress probability, stress level, or stress state vector. Let's assume a time series... The corresponding historical stress data is as follows: , Historical stress data were compared with environmental factor characteristics at the same time point. and group behavior characteristics Perform correlation to construct joint time series samples Furthermore, a sliding time window approach can be used to divide the continuous time window into segments. A time series sample segment is composed of joint samples from several time points: Understandably, time-series sample data is used to describe the evolution of a group's stress state over time.

[0115] S42, construct a spatiotemporal graph structure based on the time-series sample data, wherein the nodes in the graph structure are used to represent animals or animal groups, the node attributes are used to characterize the environmental factor characteristics and group behavior pattern characteristics at the corresponding time, and the edges between nodes are used to represent the spatial adjacency relationship or behavioral association relationship between animals.

[0116] In this step, at each time slice Construct a spatial map Among them, the node set Represents an individual animal or a predefined group unit (e.g., a spatial region unit); edge set It represents the spatial or behavioral relationships between nodes.

[0117] For nodes Its time The node attributes can be represented as: .in, This indicates the environmental factor characteristics of the region or animal corresponding to the node. Indicates characteristics of group behavior. This indicates the stress prediction result obtained by the node in step S3.

[0118] In the time dimension, the states of the same node in adjacent time slices are connected by time edges, thus forming a spatiotemporal graph structure. Spatial edges describe the relationships between nodes at the same time point, while temporal edges describe the evolutionary relationships of the same node across different time slices.

[0119] S43, Based on the spatiotemporal graph structure, a spatiotemporal graph neural network is used to model the node attributes and their temporal correlations, and output the stress propagation path that represents the propagation relationship of the group stress response in the time and space dimensions.

[0120] In this step, a spatiotemporal graph neural network is used to analyze the spatiotemporal graph structure. Modeling is performed. Spatiotemporal graph neural networks perform information propagation between adjacent nodes in the spatial dimension and information transfer across time slices in the temporal dimension. Their node feature update process can be represented as:

[0121]

[0122] in, Represents a node The set of neighboring nodes in a spatial graph; The weights are the spatial edge weights. , These are learnable parameters.

[0123] After multi-layer spatiotemporal propagation, a high-dimensional representation of each node in each time slice is obtained, and the propagation relationship of stress state between nodes is inferred based on this representation.

[0124] Conventional spatiotemporal graph neural networks typically propagate features based solely on fixed or static adjacency relationships, making it difficult to accurately characterize the propagation characteristics of group stress in real-world aquaculture scenarios, which vary over time and are influenced by both behavioral states and spatial structures. Therefore, this embodiment introduces dynamic association weights, propagation direction constraints, and multi-timescale modeling mechanisms during the spatiotemporal graph structure and propagation modeling process to enhance the structural expressiveness of stress propagation paths in both temporal and spatial dimensions.

[0125] As an example, based on the aforementioned spatiotemporal graph structure, a spatiotemporal graph neural network is used to model node attributes and their temporal relationships, outputting stress propagation paths that characterize the propagation relationships of group stress responses in the temporal and spatial dimensions, including:

[0126] S431, In the spatiotemporal graph structure, a temporal association weight that changes with time is introduced for the edges between each node. The temporal association weight is dynamically updated according to the degree of historical stress association between animals corresponding to adjacent nodes, the trend of spatial distance change, and the similarity of group behavior patterns, so as to form a temporally weighted adjacency relationship.

[0127] In this step, for any two nodes i and j in the spatiotemporal graph that are spatially or behaviorally related, a dynamic edge weight is defined at time t to characterize the probability of stress propagation association between the two at the current moment.

[0128] Dynamic edge weights can be expressed as: .in, This indicates the degree of stress correlation between node i and node j within a historical time window, and can be calculated from the correlation of the stress prediction sequence, for example:

[0129]

[0130] in, This represents the trend of spatial distance changes, used to characterize the impact of changes in distance between nodes over time, and can be expressed by the following formula:

[0131]

[0132] in For a moment The distance between nodes; This represents the similarity of group behavior patterns, used to characterize the consistency of the behavioral states of two nodes. It can be calculated from the similarity of the behavior embedding vectors corresponding to the nodes:

[0133] .

[0134] This step transforms the originally static spatial adjacency relationship into a weighted adjacency relationship that changes over time, enabling the spatiotemporal graph structure to reflect the dynamic evolution of stress propagation correlation.

[0135] S432, Based on the time-weighted adjacency relationship, the information propagation process in the spatiotemporal graph neural network is constrained so that the propagation direction of node features in the time dimension is consistent with the direction of historical stress state change, thereby forming a spatiotemporal feature transmission path with propagation directionality;

[0136] This step further distinguishes the possible directional relationships of propagation between nodes. Specifically, it compares the temporal order of historical stress state changes between node i and node j. When the stress prediction result of node i changes significantly earlier than that of node j, information is only allowed to propagate along the path of the stress state change. The direction of propagation. This direction constraint can be expressed as:

[0137]

[0138] in, and These represent the time and position of the change in stress state of node i and node j within the historical window, respectively.

[0139] By introducing temporal constraints into the feature propagation process in the above manner, the propagation links formed in the spatiotemporal graph neural network have a clear temporal direction, thus avoiding disordered or reverse propagation of propagation relationships in the time dimension.

[0140] S433, in the spatiotemporal graph neural network, the node attributes are modeled in parallel at different time scales, and the stress state change features in the short time scale and the stress evolution trend features in the long time scale are extracted respectively, and the features at different time scales are fused.

[0141] In this step, for the same node i, temporal feature representations are constructed for both short-term and long-term time scales. Let the length of the short-term window be... The long window length is ,and Then the features of a node at different scales can be represented as: .in, and These represent time series modeling functions for different time scales.

[0142] By fusing the two types of time-scale features, a comprehensive temporal representation of the nodes is obtained: .in, This is a fusion coefficient used to balance the contributions of short-term changes and long-term trends.

[0143] S434, based on the fused node features, determines the propagation order and propagation correlation of the group stress response between spatial nodes and in the time series, and outputs the stress propagation path to characterize the propagation process of the group stress response in the group.

[0144] In this step, the dynamic association weights between nodes are used. Propagation direction constraint results and fused node features This involves identifying node pairs that exhibit propagation associations within adjacent time slices. Specifically, a node is considered to be associated with a propagation association within a time slice when the following conditions are met. There exists a propagation relationship from node i to node j: and .in, The threshold for determining propagation.

[0145] The nodes that meet the conditions are organized in chronological order to form a set of propagation paths. The propagation path set is used to characterize the propagation process of group stress response between different spatial nodes over time.

[0146] S5, based on the stress propagation path, adjust the control parameters of at least one environmental control device in the ventilation system and temperature and humidity control system of the breeding site in real time.

[0147] The stress propagation path not only reflects which spatial nodes or group units are currently under stress, but also further reflects the propagation sequence of stress in the temporal dimension and the spatial correlation. Therefore, this step, combining the propagation direction, propagation intensity, and propagation sequence contained in the propagation path, adjusts environmental control parameters to reflect spatial differences and temporal order.

[0148] In this embodiment, the environmental control equipment includes at least one of a ventilation system and a temperature and humidity control system; the control parameters may include, but are not limited to: the air volume, fan speed, supply and exhaust ratio, air direction or damper opening of the ventilation equipment, and the target temperature, target humidity, humidification or dehumidification intensity of the temperature and humidity control equipment.

[0149] First, based on the stress transmission pathway Identify the spatial nodes or group units that need to participate in control. Nodes can be screened based on propagation weights, for example, when a node satisfies the following formula in the propagation path,

[0150]

[0151] This method identifies the node as one of the nodes that should prioritize participation in environmental control within the current time window. It distinguishes upstream, midstream, and downstream nodes in the propagation path, providing a structured basis for subsequent control parameter allocation.

[0152] After determining the set of nodes participating in the control, the adjustment of control parameters for each node is further calculated based on the stress propagation path. Let the stress prediction result of node i at time t be... Its propagation intensity index in the propagation path can be defined as:

[0153]

[0154] The propagation intensity index is used to comprehensively characterize the degree of participation of node i in the propagation path.

[0155] Based on this propagation intensity index, adjustment functions can be defined for different types of environmental control parameters. For example, for the airflow parameter of a ventilation system. Its adjustment amount can be expressed as: .in, This is a preset parameter mapping function used to map stress prediction results and propagation intensity to specific airflow adjustment values.

[0156] Similarly, for the target humidity parameter of the temperature and humidity control system Its adjustment amount can be expressed as: .

[0157] It should be noted that the above mapping function is not limited to a specific form, and it can be implemented using piecewise functions, linear functions, or nonlinear functions according to the engineering implementation requirements.

[0158] In this embodiment, the temporal sequence information contained in the stress propagation path is further considered to avoid control conflicts caused by simultaneous parameter adjustments in multiple regions. Specifically, the control parameter adjustments are scheduled in a time-sharing manner according to the propagation time sequence of nodes in the propagation path. Specifically, let the time when node i first appears in the propagation path be... , can be according to Sort the nodes according to their order of appearance: This generates a time-sharing control sequence. During execution, control parameter adjustments are prioritized for nodes that are ranked higher, and subsequent nodes are adjusted only after the control action for that node is completed. In this way, the control actions remain consistent with the stress propagation sequence in the time dimension.

[0159] In some implementations, the ventilation and temperature / humidity control systems can support differentiated control based on spatial zones. In this case, control parameters are assigned to zones based on the spatial structure of the propagation path. Specifically, let the base adjustment of a certain control parameter across the entire field be... Based on the weights of each node in the propagation path, a local adjustment ratio can be assigned to node i:

[0160]

[0161] Thus, the final control parameter adjustment amount for node i is obtained: .

[0162] This method enables the control parameters to be spatially differentiated, allowing the adjustment action to match the spatial correlation reflected by the propagation path.

[0163] Finally, the calculated control parameter adjustments are converted into corresponding equipment control commands and sent to the ventilation equipment or temperature and humidity control equipment for execution via the communication interface. After execution, animal data and environmental factor data are collected again according to step S1, and steps S2 to S5 are repeated in subsequent cycles, thus forming a control process that is continuously updated based on the propagation path. It should be noted that the above closed-loop update is one operating mode of this embodiment, used to illustrate the data flow relationship between steps, and does not constitute a limitation on the order or number of steps in the claims.

[0164] Please see Figure 2 This invention also provides a ventilation control system 100 for animal husbandry facilities, the system comprising:

[0165] The data acquisition module 11 is used to collect animal data and environmental factor data in the breeding site in real time; wherein, the animal data includes animal location, density, movement trajectory and interaction behavior data between animals, and the environmental factor data includes temperature, humidity, gas concentration and light intensity.

[0166] The group behavior modeling module 12 is used to construct an animal group behavior network based on the animal data, and to use a graph neural network to model and analyze the animal group behavior network to identify group behavior patterns, including group aggregation, dispersion or abnormal behavior patterns.

[0167] The multi-factor fusion prediction module 13 is used to couple the environmental factor data with the group behavior pattern through a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model for predicting group stress response;

[0168] The stress propagation modeling module 14 is used to establish a propagation model of the group stress response in the group based on historical stress data, the environmental factor data and the group behavior pattern, and to predict the stress propagation path.

[0169] The environmental control module 15 is used to adjust the control parameters of at least one environmental control device in the ventilation system and temperature and humidity control system of the breeding site in real time based on the stress propagation path.

[0170] As an example, please refer to Figure 3 The group behavior modeling module 12 includes:

[0171] The behavior network construction unit 121 is used to treat each animal in the breeding site as a node in the graph structure, and constructs edges representing the relationship between animals based on the spatial distance, relative movement state and interaction intensity between animals, thereby forming an animal group behavior network.

[0172] The behavior feature propagation unit 122 is used to perform feature propagation and aggregation operations on nodes and their associated edges based on the animal group behavior network using a graph neural network, so as to obtain node representations that reflect the individual animal behavior features and their group association features.

[0173] The behavior pattern recognition unit 123 is used to analyze the overall behavior state of the animal group based on the changes represented by the nodes, and obtain the group behavior pattern that characterizes the animal group's aggregation, dispersion or abnormal changes.

[0174] As an example, please refer to Figure 4 The multi-factor fusion prediction module 13 includes:

[0175] The feature representation unit 131 is used to perform feature representation on the environmental factor data and the group behavior pattern respectively, converting temperature, humidity, gas concentration and light intensity into environmental feature vectors, and converting group aggregation, dispersion or abnormal behavior patterns into behavior feature vectors;

[0176] The nonlinear fusion unit 132 is used to introduce a nonlinear weighting function to assign corresponding weights to different types of features based on the environmental feature vector and the behavioral feature vector, and to perform fusion processing on the feature vector through nonlinear mapping to obtain a joint feature representation;

[0177] The linkage prediction unit 133 is used to construct a multi-factor linkage prediction model based on the joint feature representation and output the group stress prediction result corresponding to the joint feature representation.

[0178] As an example, please refer to Figure 5 The stress propagation modeling module 14 includes:

[0179] The time-series sample construction unit 141 is used to acquire historical stress data and associate it with the corresponding environmental factor data and the group behavior pattern in chronological order to form time-series sample data for describing the group stress change process.

[0180] The spatiotemporal graph construction unit 142 is used to construct a spatiotemporal graph structure based on the time-series sample data. In the graph structure, nodes are used to represent animals or animal groups, node attributes are used to characterize the environmental factor characteristics and group behavior pattern characteristics at the corresponding time, and edges between nodes are used to represent the spatial adjacency relationship or behavioral association relationship between animals.

[0181] The propagation path prediction unit 143 is used to model the node attributes and their temporal correlations using a spatiotemporal graph neural network based on the spatiotemporal graph structure, and output the stress propagation path that represents the propagation relationship of the group stress response in the time and space dimensions.

[0182] As an example, please refer to Figure 6 The propagation path prediction unit 143 includes:

[0183] The temporal adjacency modulation unit 1431 is used to introduce temporal association weights that change with time to the edges between nodes in the spatiotemporal graph structure. The temporal association weights are dynamically updated according to the degree of historical stress association between animals corresponding to adjacent nodes, the trend of spatial distance change, and the similarity of group behavior patterns, so as to form a temporally weighted adjacency relationship.

[0184] The propagation direction constraint unit 1432 is used to constrain the information propagation process in the spatiotemporal graph neural network based on the time-weighted adjacency relationship, so that the propagation direction of node features in the time dimension is consistent with the historical stress state change direction, thereby forming a spatiotemporal feature transmission path with propagation directionality.

[0185] The multi-timescale modeling unit 1433 is used to perform parallel modeling of node attributes at different timescales in the spatiotemporal graph neural network, extract stress state change features in short timescales and stress evolution trend features in long timescales, and fuse the features at different timescales.

[0186] The propagation path generation unit 1434 is used to determine the propagation order and propagation correlation of the group stress response between spatial nodes and in the time series based on the fused node features, and output the stress propagation path to characterize the propagation process of the group stress response in the group.

[0187] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A ventilation control method for animal husbandry facilities, characterized in that, The method includes the following steps: S1, real-time collection of animal data and environmental factor data within the breeding site; wherein, the animal data includes animal location, density, movement trajectory and interaction behavior data between animals, and the environmental factor data includes temperature, humidity, gas concentration and light intensity; S2, construct an animal group behavior network based on the animal data, and use a graph neural network to model and analyze the animal group behavior network to identify group behavior patterns, including group aggregation, dispersion or abnormal behavior patterns; S3, The environmental factor data and the group behavior pattern are coupled through a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model for predicting group stress response; S4. Based on historical stress data, the environmental factor data, and the group behavior pattern, a spatiotemporal graph neural network is used to establish a propagation model of the group stress response in the group and predict the stress propagation path. S5, based on the stress propagation path, adjust the control parameters of at least one environmental control device in the ventilation system and temperature and humidity control system of the breeding site in real time.

2. The ventilation control method for animal breeding facilities according to claim 1, characterized in that: Based on the animal data, an animal group behavior network is constructed. A graph neural network is then used to model and analyze the animal group behavior network to identify group behavior patterns, including: S21. Each animal in the breeding area is treated as a node in the graph structure, and edges representing the relationships between animals are constructed based on the spatial distance, relative movement state and interaction intensity between animals, thus forming an animal group behavior network. S22, Based on the animal group behavior network, a graph neural network is used to perform feature propagation and aggregation operations on nodes and their associated edges to obtain node representations that reflect the individual animal behavior characteristics and their group association characteristics; S23. Based on the changes represented by the nodes, the overall behavioral state of the animal group is analyzed to obtain a group behavior pattern that characterizes the aggregation, dispersion, or abnormal changes of the animal group.

3. The ventilation control method for animal breeding facilities according to claim 1, characterized in that: The environmental factor data and the group behavior pattern are coupled using a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model, including: S31, the environmental factor data and the group behavior pattern are respectively represented by features, and the temperature, humidity, gas concentration and light intensity are converted into environmental feature vectors, and the group gathering, dispersal or abnormal behavior patterns are converted into behavior feature vectors; S32, Based on the environmental feature vector and the behavioral feature vector, a nonlinear weighting function is introduced to assign corresponding weights to different types of features, and the feature vectors are fused through nonlinear mapping to obtain a joint feature representation; S33, Construct a multi-factor linkage prediction model based on the joint feature representation, and output the group stress prediction result corresponding to the joint feature representation.

4. The ventilation control method for animal breeding facilities according to claim 1, characterized in that: Based on historical stress data, environmental factor data, and group behavior patterns, a spatiotemporal graph neural network is used to establish a propagation model of the group's stress response within the group, predicting the stress propagation path, including: S41, acquire historical stress data and associate it with the corresponding environmental factor data and the group behavior pattern in chronological order to form time-series sample data for describing the group stress change process; S42, construct a spatiotemporal graph structure based on the time-series sample data, wherein the nodes in the graph structure are used to represent animals or animal groups, the node attributes are used to characterize the environmental factor characteristics and group behavior pattern characteristics at the corresponding time, and the edges between nodes are used to represent the spatial adjacency relationship or behavioral association relationship between animals. S43, Based on the spatiotemporal graph structure, a spatiotemporal graph neural network is used to model the node attributes and their temporal correlations, and output the stress propagation path that represents the propagation relationship of the group stress response in the time and space dimensions.

5. A ventilation control method for animal breeding facilities according to claim 4, characterized in that: Based on the aforementioned spatiotemporal graph structure, a spatiotemporal graph neural network is used to model node attributes and their temporal relationships, outputting stress propagation paths that characterize the propagation relationships of group stress responses in the temporal and spatial dimensions, including: S431, In the spatiotemporal graph structure, a temporal association weight that changes with time is introduced for the edges between each node. The temporal association weight is dynamically updated according to the degree of historical stress association between animals corresponding to adjacent nodes, the trend of spatial distance change, and the similarity of group behavior patterns, so as to form a temporally weighted adjacency relationship. S432, Based on the time-weighted adjacency relationship, the information propagation process in the spatiotemporal graph neural network is constrained so that the propagation direction of node features in the time dimension is consistent with the direction of historical stress state change, thereby forming a spatiotemporal feature transmission path with propagation directionality; S433, in the spatiotemporal graph neural network, the node attributes are modeled in parallel at different time scales, and the stress state change features in the short time scale and the stress evolution trend features in the long time scale are extracted respectively, and the features at different time scales are fused. S434, based on the fused node features, determines the propagation order and propagation correlation of the group stress response between spatial nodes and in the time series, and outputs the stress propagation path to characterize the propagation process of the group stress response in the group.

6. A ventilation control system for animal husbandry facilities, characterized in that, The system includes: The data acquisition module is used to collect animal data and environmental factor data in the breeding site in real time; wherein, the animal data includes animal location, density, movement trajectory and interaction behavior data between animals, and the environmental factor data includes temperature, humidity, gas concentration and light intensity; The group behavior modeling module is used to construct an animal group behavior network based on the animal data, and to use a graph neural network to model and analyze the animal group behavior network to identify group behavior patterns, including group aggregation, dispersion or abnormal behavior patterns. The multi-factor fusion prediction module is used to couple the environmental factor data with the group behavior pattern through a nonlinear weighted fusion algorithm to construct a multi-factor linkage prediction model for predicting group stress response. The stress propagation modeling module is used to establish a propagation model of the group stress response in the group based on historical stress data, the environmental factor data and the group behavior pattern, and to predict the stress propagation path. An environmental control module is used to adjust the control parameters of at least one environmental control device in the ventilation system and temperature and humidity control system of the breeding site in real time based on the stress propagation path.

7. A ventilation control system for animal husbandry facilities according to claim 6, characterized in that: The group behavior modeling module includes: The behavioral network construction unit is used to treat each animal in the breeding site as a node in a graph structure, and constructs edges representing the relationships between animals based on the spatial distance, relative movement state and interaction intensity between animals, thereby forming an animal group behavioral network. The behavior feature propagation unit is used to perform feature propagation and aggregation operations on nodes and their associated edges based on the animal group behavior network using a graph neural network, so as to obtain node representations that reflect the individual animal behavior features and their group association features; The behavior pattern recognition unit is used to analyze the overall behavior state of the animal group based on the changes represented by the nodes, and obtain group behavior patterns that characterize the animal group's aggregation, dispersion, or abnormal changes.

8. A ventilation control system for animal husbandry facilities according to claim 6, characterized in that: The multi-factor fusion prediction module includes: The feature representation unit is used to represent the environmental factor data and the group behavior pattern respectively, converting temperature, humidity, gas concentration and light intensity into environmental feature vectors, and converting group aggregation, dispersion or abnormal behavior patterns into behavioral feature vectors. The nonlinear fusion unit is used to introduce a nonlinear weighting function to assign corresponding weights to different types of features based on the environmental feature vector and the behavioral feature vector, and to perform fusion processing on the feature vector through nonlinear mapping to obtain a joint feature representation; The linkage prediction unit is used to construct a multi-factor linkage prediction model based on the joint feature representation and output the group stress prediction result corresponding to the joint feature representation.

9. A ventilation control system for animal husbandry facilities according to claim 6, characterized in that: The stress propagation modeling module includes: A time-series sample construction unit is used to acquire historical stress data and associate it with the corresponding environmental factor data and the group behavior pattern in chronological order to form time-series sample data for describing the group stress change process. The spatiotemporal graph construction unit is used to construct a spatiotemporal graph structure based on the time-series sample data. In the graph structure, nodes are used to represent animals or animal groups, node attributes are used to characterize the environmental factors and group behavior patterns at the corresponding time, and edges between nodes are used to represent the spatial adjacency or behavioral association between animals. The propagation path prediction unit is used to model the node attributes and their temporal correlations based on the spatiotemporal graph structure using a spatiotemporal graph neural network, and output the stress propagation path that represents the propagation relationship of the group stress response in the time and space dimensions.

10. A ventilation control system for animal husbandry facilities according to claim 9, characterized in that: The propagation path prediction unit includes: The temporal adjacency modulation unit is used to introduce temporal association weights that change with time to the edges between nodes in the spatiotemporal graph structure. The temporal association weights are dynamically updated according to the degree of historical stress association between animals corresponding to adjacent nodes, the trend of spatial distance change, and the similarity of group behavior patterns, so as to form a temporally weighted adjacency relationship. The propagation direction constraint unit is used to constrain the information propagation process in the spatiotemporal graph neural network based on the time-weighted adjacency relationship, so that the propagation direction of node features in the time dimension is consistent with the historical stress state change direction, thereby forming a spatiotemporal feature transmission path with propagation directionality. The multi-timescale modeling unit is used to perform parallel modeling of node attributes at different timescales in the spatiotemporal graph neural network, extract stress state change features in short timescales and stress evolution trend features in long timescales, and fuse the features at different timescales. The propagation path generation unit is used to determine the propagation order and propagation correlation of the group stress response between spatial nodes and in the time series based on the fused node features, and outputs the stress propagation path to characterize the propagation process of the group stress response in the group.

Images