Method and System for Risk Assessment of Bacterial Transmission in Chicken Coops

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses a method and system for evaluating the risk of bacterial transmission in a chicken coop. The method includes: collecting multi-modal perception data of bacteria in the chicken coop in real time through a distributed sensor array, extracting spatio-temporal features from the multi-modal perception data by using a dual attention spatio-temporal graph neural network to generate a bacterial transmission risk prediction map with confidence evaluation, inputting the bacterial transmission risk prediction map into a digital twin platform, simulating the dynamic simulation environment of the chicken coop by fusing a BIM model and CFD to obtain a CFD-BIM coupled dynamic simulation model, constructing a deep reinforcement learning decision model based on the CFD-BIM coupled dynamic simulation model to generate a bacterial prevention and control execution instruction for the chicken coop, and after feeding back the prevention and control execution instruction to the digital twin platform for virtual verification, outputting a visual decision report. The present invention can improve the accuracy of evaluating the risk of bacterial transmission in a chicken coop.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a method and system for evaluating the risk of bacterial transmission in a chicken coop. Background Art

[0002] Currently, the risk assessment of bacterial transmission in chicken coops mainly relies on manual inspections or single environmental parameter monitoring, which has problems such as insufficient data collection dimensions and poor real-time performance. Traditional methods are mostly based on static environmental models or simple statistical models, and it is difficult to capture the spatio-temporal transmission law of bacteria in a complex airflow field, resulting in limited risk assessment accuracy. In addition, in existing technologies, sensor networks mostly use single-modal data, lacking multi-modal collaborative perception of the characteristic spectra of bacteria in the air, the distribution of surface bacteria, and environmental parameters, and unable to comprehensively reflect the overall picture of the dynamic transmission of bacteria in the chicken coop, further restricting the accuracy of the assessment.

[0003] Existing prevention and control strategies mostly rely on empirical decision-making and do not combine dynamic simulation with multi-objective optimization. The integration degree of traditional CFD simulation and building information model (BIM) is low, and it is difficult to construct a high-fidelity dynamic environment of the chicken coop; and the selection of prevention and control measures often only considers a single objective (such as cost or risk), lacking comprehensive consideration of multiple factors such as delay time and coverage accuracy. This leads to insufficient robustness of the prevention and control strategy, making it difficult to adapt to complex and changeable actual scenarios, increasing the risk of chicken flock infection and breeding costs. Summary of the Invention

[0004] The present invention provides a method and system for evaluating the risk of bacterial transmission in a chicken coop, and its main purpose is to solve the problem of low accuracy in evaluating the risk of bacterial transmission in a chicken coop.

[0005] To achieve the above object, a method for evaluating the risk of bacterial transmission in a chicken coop provided by the present invention includes:

[0006] S1: Real-time collect multi-modal perception data of bacteria in the chicken coop through a distributed sensor array;

[0007] S2: Use a dual attention spatio-temporal graph neural network to extract spatio-temporal features from the multi-modal perception data, and generate a bacterial transmission risk prediction map with confidence evaluation;

[0008] S3: Input the bacterial transmission risk prediction map into a digital twin platform, and obtain a CFD-BIM coupled dynamic simulation model by fusing the BIM model and CFD to simulate the dynamic simulation environment of the chicken coop;

[0009] S4: Construct a deep reinforcement learning decision model based on the CFD-BIM coupled dynamic simulation model, where:

[0010] Using the real-time airflow field data and bacterial diffusion data in the dynamic simulation environment as the basic input of the state space;

[0011] Taking the multi-dimensional prevention and control measure combination as the action space;

[0012] Optimizing the bacterial prevention and control strategy through a multi-objective reward function to generate the bacterial prevention and control execution instructions for the chicken coop;

[0013] After feeding back the prevention and control execution instructions to the digital twin platform for virtual verification, an output of a visual decision-making report is generated.

[0014] Optionally, the distributed sensor array includes:

[0015] A miniaturized surface-enhanced Raman spectroscopy sensor for capturing the characteristic spectral data of bacteria in the air in the chicken coop in real time;

[0016] A microfluidic sampling chip modified with magnetic nanomaterials for adsorbing bacteria on the inner surface of the chicken coop and recording the spatio-temporal coordinates of the sampling points;

[0017] An environmental parameter monitoring module integrated with a temperature and humidity sensor, an ammonia concentration sensor, and a wind speed sensor for collecting the environmental parameters in the chicken coop, and the data of each sensor is encrypted and transmitted through the LoRa self-organizing network protocol.

[0018] Optionally, the dual attention spatio-temporal graph neural network includes a spatial attention layer and a temporal attention layer, where:

[0019] The spatial attention layer calculates the bacterial propagation path weights between sensor nodes in the distributed sensor array based on a path weight generation algorithm, where the path weight generation algorithm is as follows:

[0020] ;

[0021] Where, is the bacterial propagation path weight, is the normalization function, is the value matrix, is the dimension scaling factor, is the query matrix, is the key matrix, is the transpose of the key matrix;

[0022] The temporal attention layer extracts the temporal dependence relationship of the bacterial propagation risk through a temporal dependence generation algorithm, where the temporal dependence generation algorithm is:

[0023] ;

[0024] Where, is the timing dependence relationship, is a long short-term memory network, is a Hadamard product, is the multi-modal perception data, is a weight matrix, is a bias vector, is a time step index, is a Sigmoid activation function.

[0025] Optionally, the spatial attention layer further includes:

[0026] Construct a topological map of the chicken coop, where the topological nodes are the deployment positions of the sensors in the distributed sensor array, and the edge weights are jointly determined by the connection relationship of the ventilation ducts in the chicken coop and the historical bacterial transmission event data;

[0027] Weight and fuse the bacterial transmission path weights and the graph convolution output features to obtain a high-risk path identification matrix of the connection paths between the sensor nodes.

[0028] Optionally, the dynamic simulation environment of the chicken coop by fusing the BIM model and CFD includes:

[0029] Discretize the three-dimensional geometric structure of the chicken coop in the BIM model into an unstructured CFD calculation grid, with the chicken cage layout in the chicken coop as the solid obstacle boundary, the ventilation opening position as the velocity inlet boundary, and the pressure outlet boundary;

[0030] Simulate the aerosol diffusion field in the chicken coop by solving the Navier-Stokes equation, and map the bacterial concentration in the bacterial transmission risk prediction map to the grid nodes in the unstructured CFD calculation grid;

[0031] In the BIM visualization interface, overlay the dynamic diffusion particle trajectories simulated by CFD on the BIM model and render the risk levels of different regions in the chicken coop in the form of a heat map.

[0032] Optionally, the multi-objective reward function is designed as:

[0033] ;

[0034] Wherein, is the multi-objective reward function value, is the balance weight for adjusting the risk reduction and cost consumption, is the negative impact weight of the control delay time on the reward, is the reward weight for adjusting the accuracy of the coverage range of the prevention and control measures, is the amount of risk level reduction, is the cost of prevention and control measures, is the prevention and control delay time, is the actual biological contamination area, is the area where the prevention and control measures take effect, is the overlap degree between the actual biological contamination area and the area where the prevention and control measures take effect.

[0035] Optionally, the method for determining the edge weight includes:

[0036] Based on the ventilation duct connection relationship, an initial adjacency matrix is constructed, where the initial edge weight between adjacent nodes in the initial adjacency matrix is the product of the duct cross-sectional area and the real-time wind speed in the ventilation duct;

[0037] Based on the historical bacteria transmission event data, a historical transmission data correction factor for the initial edge weight is generated, and the initial edge weight is dynamically adjusted based on the historical transmission data correction factor. The calculation formula for the historical transmission data correction factor is:

[0038] ;

[0039] Wherein, is the balance factor for adjusting the physical properties of the ventilation duct and the historical bacteria transmission event data, is the duct cross-sectional area, is the real-time wind speed in the ventilation duct, is the number of times bacteria spread from node to node is the number of times of total transmission events, is the total number of transmission events, is the th topological node in the topological graph, is the th topological node in the topological graph.

[0040] Optionally, the feedback of the prevention and control execution instruction to the digital twin platform for virtual verification includes:

[0041] Set a Monte Carlo simulation module in the digital twin platform to randomly perturb the environmental parameters to evaluate the robustness of the bacteria prevention and control strategy;

[0042] Through the OPC-UA protocol, the verified prevention and control execution instruction is sent to the fan and spray disinfection equipment controller of the chicken coop, and the environmental feedback data after implementing the bacteria prevention and control strategy is collected in real time to update the input data of the dual attention spatio-temporal graph neural network in S2.

[0043] Optionally, the dynamic simulation environment of the chicken coop obtained by fusing the BIM model and CFD simulation to obtain a CFD-BIM coupled dynamic simulation model further includes:

[0044] Based on the aerosol diffusion field, extract the concentration gradient change rate of each grid node within the target time period, thereby constructing a spatio-temporal propagation tensor;

[0045] Locate the coordinates of the most probable pollution source within the chicken coop through the reverse particle tracking algorithm and the spatio-temporal propagation tensor.

[0046] Map the coordinates of the most probable pollution source located based on the grid nodes to the original BIM components through the local coordinate system transformation matrix of the BIM model;

[0047] Retrieve the subset of operation logs associated with the original BIM components, and associate the coordinates of the most probable pollution source with the subset of operation logs to obtain a propagation traceability analysis report.

[0048] To solve the above problems, the present invention also provides a bacterial transmission risk assessment system for a chicken coop, and the system includes:

[0049] A data acquisition module for real-time collecting multi-modal perception data of bacteria in the chicken coop through a distributed sensor array;

[0050] A feature extraction module for performing spatio-temporal feature extraction on the multi-modal perception data using a dual attention spatio-temporal graph neural network to generate a bacterial transmission risk prediction map with confidence assessment;

[0051] A dynamic simulation module for inputting the bacterial transmission risk prediction map into a digital twin platform, and obtaining a CFD-BIM coupled dynamic simulation model by fusing the BIM model and CFD simulation of the dynamic simulation environment of the chicken coop;

[0052] A decision model module for constructing a deep reinforcement learning decision model based on the CFD-BIM coupled dynamic simulation model, thereby generating a bacterial prevention and control execution instruction for the chicken coop, and after feeding back the prevention and control execution instruction to the digital twin platform for virtual verification, outputting a visual decision report.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. By collaborating with a distributed sensor array to collect multimodal data such as spectra, surface bacteria, and environmental parameters, and combining LoRa self-organizing network encrypted transmission, real-time and comprehensive monitoring of bacterial transmission is achieved. At the same time, based on a dual-attention spatio-temporal graph neural network, through a spatial attention layer and a temporal attention layer, the spatio-temporal dependence relationship of bacterial transmission is accurately captured, and a risk prediction map with confidence is generated, significantly improving the evaluation accuracy;

[0055] 2. The CFD-BIM coupling model integrates aerosol diffusion simulation with three-dimensional building information, locates the pollution source by combining the reverse particle tracking algorithm, and realizes the tracing of transmission through the association of BIM operation logs, providing a scientific basis for precise prevention and control;

[0056] 3. The deep reinforcement learning model optimizes the prevention and control strategy with a multi-objective reward function, and verifies the robustness through Monte Carlo simulation on the digital twin platform, realizing the intelligent generation and closed-loop optimization of the prevention and control strategy in a dynamic environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic flowchart of a method for evaluating the risk of bacterial transmission in a chicken coop provided by an embodiment of the present invention;

[0058] Figure 2 It is a functional module diagram of a system for evaluating the risk of bacterial transmission in a chicken coop provided by an embodiment of the present invention;

[0059] The realization, functional features, and advantages of the object of the present invention will be further described in conjunction with the embodiments with reference to the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0061] An embodiment of the present application provides a method for evaluating the risk of bacterial transmission in a chicken coop. The execution subject of the method for evaluating the risk of bacterial transmission in a chicken coop includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for evaluating the risk of bacterial transmission in a chicken coop can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0062] Refer toFigure 1 As shown in the figure, it is a schematic flowchart of a method for evaluating the risk of bacterial transmission in a chicken coop provided by an embodiment of the present invention. In this embodiment, the method for evaluating the risk of bacterial transmission in the chicken coop includes:

[0063] S1: Real-time collect multimodal perception data of bacteria in the chicken coop through a distributed sensor array.

[0064] In the embodiment of the present invention, the distributed sensor array is a system composed of a combination of multiple different types of sensors, distributed at various positions in the chicken coop, and is used to comprehensively collect various data related to bacteria in the chicken coop. It can obtain information from different dimensions and provide a rich data basis for subsequent analysis and decision-making.

[0065] In the embodiment of the present invention, the distributed sensor array includes:

[0066] A miniaturized surface-enhanced Raman spectroscopy sensor for real-time capturing the characteristic spectral data of bacteria in the air in the chicken coop;

[0067] A microfluidic sampling chip modified with magnetic nanomaterials for adsorbing bacteria on the surface of the chicken coop and recording the spatio-temporal coordinates of the sampling points;

[0068] An environmental parameter monitoring module integrated with temperature and humidity sensors, ammonia concentration sensors, and wind speed sensors for collecting the environmental parameters in the chicken coop, and the data of each sensor is encrypted and transmitted through the LoRa self-organizing network protocol.

[0069] Specifically, the miniaturized surface-enhanced Raman spectroscopy sensor (SERS) is a sensor that works based on the surface-enhanced Raman scattering effect. Raman scattering is an inelastic scattering phenomenon that occurs when light interacts with matter, and different molecules have specific Raman spectral characteristics. Surface-enhanced Raman scattering greatly enhances the Raman signal through special nanostructures, enabling the detection of extremely trace amounts of substances. The miniaturized SERS is convenient for flexible deployment in the chicken coop and can real-time detect the characteristic spectral data of bacteria in the air, and identify information such as the types and concentrations of bacteria by analyzing the spectral characteristics.

[0070] Specifically, magnetic nanomaterials have unique magnetism and can adsorb specific substances. Microfluidic technology can precisely control the fluid flow in microchannels. The chip modified with magnetic nanomaterials can specifically adsorb bacteria on the surface of the chicken coop. At the same time, it is equipped with a recording device that can accurately record the spatio-temporal coordinates of the sampling points, and these coordinate information is very crucial for subsequent analysis of the distribution and transmission path of bacteria.

[0071] Specifically, the environmental parameter monitoring module integrates a variety of environmental sensors, including temperature and humidity sensors, ammonia concentration sensors, and wind speed sensors. The temperature and humidity sensors are used to measure the temperature and humidity in the chicken coop. These two environmental factors have an important impact on the growth and spread of bacteria. The ammonia concentration sensor monitors the ammonia content in the chicken coop. Excessive ammonia concentration not only affects the health of chickens but is also related to the growth of bacteria. The wind speed sensor is used to measure the air flow speed in the chicken coop. Air flow affects the diffusion range and speed of bacteria. These sensors work together to comprehensively collect the environmental parameters in the chicken coop.

[0072] Specifically, LoRa (Long Range) is a low-power, long-distance wireless communication technology. The self-organizing network protocol allows each sensor node to automatically form a network without relying on a pre-established infrastructure. Using the LoRa self-organizing network protocol to encrypt and transmit data can, on the one hand, ensure the stability and reliability of communication between sensors, enabling data to be accurately transmitted to the data processing center; on the other hand, the encryption mechanism can effectively protect the security of data and prevent data from being stolen or tampered with during transmission.

[0073] Specifically, miniaturized surface-enhanced Raman spectroscopy sensors are reasonably deployed in the chicken coop so that they can fully come into contact with the air in the chicken coop. The light source inside the sensor emits laser light, which irradiates the bacteria in the air, generating Raman scattered light. The scattered light is collected and processed by the optical system and received by the detector. The detector converts the optical signal into an electrical signal, and then through the signal amplification and processing circuit, the electrical signal is converted into a digital signal. These digital signals represent the characteristic spectral data of bacteria. These data are encoded and encrypted according to the LoRa self-organizing network protocol and then sent out through the wireless communication module.

[0074] Specifically, a microfluidic sampling chip modified with magnetic nanomaterials is arranged at the surface positions to be monitored in the chicken coop, such as walls and chicken coops. When the chip comes into contact with the inner surface of the chicken coop, the magnetic nanomaterials will adsorb the surrounding bacteria. The positioning system integrated inside the chip (such as based on GPS or other indoor positioning technologies) will real-time obtain the spatial coordinate information of the sampling point, and at the same time, the clock module on the chip records the sampling time. After sampling, the chip processes and transmits the bacterial sample information and spatio-temporal coordinate data according to the LoRa self-organizing network protocol.

[0075] Specifically, the temperature and humidity sensor utilizes principles such as thermistors and capacitive humidity sensors to convert the temperature and humidity changes in the chicken coop into changes in resistance or capacitance values, and then converts these changes into standard electrical signals through a signal conditioning circuit. The ammonia concentration sensor uses electrochemical or semiconductor sensing technologies to detect the changes in electrical signals generated by the reaction between ammonia in the chicken coop and the sensitive material on the sensor surface, thereby obtaining ammonia concentration data. The wind speed sensor measures the force of the wind on its internal blades or other sensing components and converts it into an electrical signal for output. These environmental parameter data are aggregated and processed inside the module, encrypted according to the LoRa self-organizing network protocol, and then sent out through a wireless communication module.

[0076] Furthermore, the data collected by each sensor is processed according to the LoRa self-organizing network protocol at its respective node and then sent out through wireless signals. A data receiving base station is set up inside or near the chicken coop to receive the data sent by each sensor node. The base station decodes and integrates the received data, and then transmits the complete multi-modal perception data to the data processing center for subsequent analysis and processing.

[0077] Generally speaking, through the collaborative work of multiple sensors in the distributed sensor array, it is possible to comprehensively collect information related to bacteria in the chicken coop from different angles, including the characteristic spectral data of the bacteria themselves, the distribution information of bacteria on the inner surface, and the environmental parameters in the chicken coop, providing a rich data basis for accurately evaluating the risk of bacteria transmission in the subsequent stage.

[0078] Specifically, the real-time working characteristics of the sensors can continuously obtain data, enabling breeders or managers to timely understand the dynamic changes of bacteria and the environment in the chicken coop, so as to take corresponding measures in a timely manner, such as adjusting the ventilation system, disinfecting, etc., to prevent the disease risk caused by bacteria transmission.

[0079] Specifically, the spatio-temporal coordinates of the sampling points recorded by the microfluidic sampling chip modified with magnetic nanomaterials help to determine the source and transmission path of bacteria in subsequent analysis, providing strong support for precise prevention and control. For example, by analyzing bacterial samples at different times and locations, it can be judged that the bacteria start to spread from a specific area, and then targeted key prevention, control and disinfection can be carried out on this area.

[0080] Specifically, encrypting and transmitting data using the LoRa self-organizing network protocol ensures the security and reliability of the data during transmission. In the complex chicken coop environment, the risk of data being interfered with or stolen is avoided, ensuring the integrity and accuracy of the data, making the decisions based on these data more reliable.

[0081] Specifically, a miniaturized surface-enhanced Raman spectroscopy sensor, a microfluidic sampling chip modified with magnetic nanomaterials, and an environmental parameter monitoring module are combined into a distributed sensor array. This way of multi-sensor collaborative acquisition of multi-modal data is different from the monitoring method of traditional single-type sensors. Traditional methods may only be able to obtain single-dimensional data and cannot comprehensively understand the bacteria situation in the chicken coop. However, through the organic combination of multiple sensors in the present invention, all-round and multi-level monitoring of bacteria and their environment in the chicken coop is achieved, greatly improving the richness and accuracy of data, and providing the possibility for more accurate subsequent risk assessment and prevention and control.

[0082] Specifically, the miniaturized surface-enhanced Raman spectroscopy sensor utilizes surface-enhanced Raman scattering technology and can detect the characteristic spectral data of bacteria in the air in real time and quickly in a complex chicken coop environment, which is difficult to achieve by traditional detection methods. Traditional bacteria detection methods may require sample collection and analysis in the laboratory, which not only takes a long time but also cannot reflect the bacteria situation in the chicken coop in real time. The microfluidic sampling chip modified with magnetic nanomaterials can specifically adsorb bacteria on the surface of the chicken coop and record the spatio-temporal coordinates, providing key data for the traceability and precise prevention and control of bacteria.

[0083] Generally speaking, the multi-modal perception data collected is the original input data for subsequent spatio-temporal feature extraction (generating a bacteria transmission risk prediction map) using a dual-attention spatio-temporal graph neural network. Data such as temperature, humidity, ammonia concentration, and wind speed collected by the environmental parameter monitoring module will affect the growth and transmission characteristics of bacteria and have an important impact on analyzing the bacteria transmission risk. The characteristic spectral data of bacteria obtained by the miniaturized surface-enhanced Raman spectroscopy sensor and the bacteria sample and spatio-temporal coordinate information collected by the microfluidic sampling chip modified with magnetic nanomaterials are directly related to the understanding of the types, quantities, and distributions of bacteria. These information combined provide indispensable basic data for subsequent construction of a CFD-BIM coupled dynamic simulation model and a deep reinforcement learning decision model.

[0084] S2: Use a dual-attention spatio-temporal graph neural network to perform spatio-temporal feature extraction on the multi-modal perception data and generate a bacteria transmission risk prediction map with confidence evaluation.

[0085] In the embodiment of the present invention, the dual-attention spatio-temporal graph neural network includes: a spatial attention layer and a temporal attention layer, where:

[0086] The spatial attention layer calculates the bacteria transmission path weights between sensor nodes in the distributed sensor array based on a path weight generation algorithm, where the path weight generation algorithm is as follows:

[0087] ;

[0088] Among them, is the weight of the bacterial transmission path, is the normalization function, is the value matrix, is the dimension scaling factor, is the query matrix, is the key matrix, is the transpose of the key matrix;

[0089] The time attention layer extracts the temporal dependence relationship of the bacterial transmission risk through a temporal dependence generation algorithm. Among them, the temporal dependence generation algorithm is:

[0090] ;

[0091] Among them, is the temporal dependence relationship, is the long short-term memory network, is the Hadamard product, is the multi-modal perception data, is the weight matrix, is the bias vector, is the time step index, is the Sigmoid activation function.

[0092] Specifically, the dual attention spatio-temporal graph neural network is a neural network structure specially designed to process multi-modal perception data related to bacterial transmission in chicken coops. It consists of a spatial attention layer and a time attention layer, which can extract data features from spatial and temporal dimensions respectively, and capture the laws and trends of bacterial transmission.

[0093] Specifically, the spatial attention layer is a part of the dual attention spatio-temporal graph neural network, mainly used to analyze the spatial relationship between sensor nodes in a distributed sensor array, determine the possibility of bacterial transmission between different positions, and measure the size of this possibility by calculating the weight of the bacterial transmission path.

[0094] Specifically, the time attention layer is another part of this network, focusing on mining the dependence relationship of bacterial transmission risk in time series. Considering the changing law of bacterial transmission over time, it helps to predict future transmission trends.

[0095] Specifically, the path weight generation algorithm is a mathematical method used in the spatial attention layer to calculate the weight of the bacterial transmission path between sensor nodes. Based on matrix operations and the Softmax function, it comprehensively considers the query matrix, key matrix, value matrix, and dimension scaling factor to determine the weight.

[0096] Specifically, the Softmax function is a commonly used normalization function that converts an input numerical vector into a probability distribution vector, such that all elements have values between 0 and 1 and the sum of all elements is 1. It is used in the path weight generation algorithm to convert the calculation result into a probability-form weight.

[0097] Specifically, the value matrix (V), query matrix (Q), and key matrix (K) are key matrices used for matrix operations in the path weight generation algorithm. Their specific values are determined by the input data and network parameters, and the weights of the bacterial propagation paths are calculated through operations among them.

[0098] Specifically, the dimension scaling factor is a parameter used to scale the result of matrix operations, preventing calculation problems caused by dimension differences and ensuring the stability and rationality of the calculation.

[0099] Specifically, the temporal dependence generation algorithm is an algorithm used in the time attention layer to extract the temporal dependence relationship of the bacterial propagation risk, and it is calculated in combination with the long short-term memory network (LSTM) and the Sigmoid activation function, etc.

[0100] Specifically, the long short-term memory network (LSTM) is a special recurrent neural network that can effectively process time series data. It solves the problems of gradient vanishing and gradient explosion in traditional recurrent neural networks through memory units and gating mechanisms, and is used in the temporal dependence generation algorithm to capture the long-term and short-term dependence relationships of the bacterial propagation risk in the time series.

[0101] Specifically, the Hadamard product ( ) is a matrix operation where two matrices of the same dimension are multiplied element by element to obtain a new matrix. It is used in the temporal dependence generation algorithm to multiply the output of the LSTM element by element with the result processed by the Sigmoid activation function.

[0102] Specifically, in the time attention layer, the weight matrix and bias vector are parameters of the network, which are continuously adjusted through training and are used to perform a linear transformation on the input data, affecting the input of the Sigmoid activation function and thus affecting the extraction of the temporal dependence relationship.

[0103] Specifically, the Sigmoid activation function is a function that maps any real number to the range between 0 and 1. It is used in the temporal dependence generation algorithm to perform a non-linear transformation on the result of the linear transformation, enabling the network to learn more complex relationships.

[0104] In the embodiment of the present invention, the spatial attention layer further includes:

[0105] Construct the topological graph of the chicken coop, where the topological nodes are the deployment positions of the sensors in the distributed sensor array, and the edge weights are jointly determined by the connection relationship of the ventilation ducts in the chicken coop and the historical bacterial transmission event data;

[0106] Perform weighted fusion on the bacterial transmission path weight and the graph convolution output feature to obtain the high-risk path identification matrix of the connection paths between the sensor nodes.

[0107] Specifically, the method for determining the edge weights includes:

[0108] Based on the ventilation duct connection relationship, construct an initial adjacency matrix, where the initial edge weight between adjacent nodes in the initial adjacency matrix is the product of the duct cross-sectional area and the real-time wind speed in the ventilation duct;

[0109] Generate a historical transmission data correction factor for the initial edge weight based on the historical bacterial transmission event data, and dynamically adjust the initial edge weight based on the historical transmission data correction factor. The calculation formula for the historical transmission data correction factor is:

[0110] ;

[0111] Where, is the balance factor for adjusting the physical properties of the ventilation duct and the historical bacterial transmission event data, is the duct cross-sectional area, is the real-time wind speed in the ventilation duct, is the number of times bacteria spread from node to node ; is the total number of transmission events, is the th topological node in the topological graph, is the th topological node in the topological graph.

[0112] Specifically, the topological graph of the chicken coop is a model that graphically represents the distribution and connection relationship of sensors in the chicken coop. The topological nodes represent the deployment positions of the sensors, the edges represent the connection relationships between the nodes, and the edge weights reflect the likelihood of bacterial transmission between the nodes.

[0113] Specifically, graph convolution is a technique for performing convolution operations on graph-structured data. By aggregating and transforming the features of nodes and their neighbor nodes, it extracts the feature representation of the graph and is used to process the features of sensor nodes in the spatial attention layer.

[0114] Specifically, the high-risk path identification matrix is a matrix obtained by weighted fusion of the bacterial transmission path weights and the graph convolution output features, and is used to identify the paths with higher bacterial transmission risks among the connection paths between sensor nodes. The element values in the matrix represent the risk levels.

[0115] Specifically, the initial adjacency matrix is a matrix representing the initial connection situation between sensor nodes constructed based on the ventilation duct connection relationship. The elements in the matrix represent whether there is a connection between nodes and the initial weight of the connection.

[0116] Specifically, the historical transmission data correction factor is a factor generated according to the historical bacterial transmission event data for adjusting the initial edge weights. Considering the physical attributes of the ventilation ducts and the historical transmission situation, the edge weights are made more in line with the actual possibility of bacterial transmission.

[0117] Specifically, first, the topological nodes are determined according to the actual deployment positions of the distributed sensor arrays in the chicken coop. Then, an initial adjacency matrix is constructed based on the ventilation duct connection relationship, and the initial edge weights between adjacent nodes are calculated, which is the product of the duct cross-sectional area and the real-time wind speed in the ventilation duct. For example, if node A and node B are connected by a ventilation duct, the duct cross-sectional area is , and the real-time wind speed is , then the initial edge weight between them is .

[0118] Specifically, the multi-modal perception data is subjected to a specific transformation to obtain a query matrix , a key matrix , and a value matrix , as well as a dimension scaling factor .

[0119] Specifically, the bacterial transmission path weights are calculated according to the path weight generation algorithm . Assuming that the input data is processed to obtain , , first calculate , then calculate , and finally normalize it through the Softmax function and multiply it by V to obtain

[0120] Specifically, graph convolution operations are performed on the features of the sensor nodes to obtain the graph convolution output features. The calculated bacterial transmission path weights are weighted and fused with the graph convolution output features. For example, set the weights as and ( ), then the high-risk path identification matrix , where Represent the graph convolution output features to obtain the high-risk path identification matrix of the connection paths between sensor nodes.

[0121] Specifically, according to the historical bacterial transmission event data, calculate the historical transmission data correction factor , assuming , then , use this correction factor to dynamically adjust the initial edge weights and update the edge weights of the topological graph.

[0122] Specifically, input the multi-modal perception data into the long short-term memory network LSTM. LSTM processes the data according to the time step index t, captures the long-term and short-term dependencies of the data in the time series, and obtains the output .

[0123] Meanwhile, perform a linear transformation on the input data X, that is , and then perform a non-linear transformation through the Sigmoid activation function to obtain . Finally, perform the Hadamard product operation on and , that is , to obtain the temporal dependence relationship of the bacterial transmission risk.

[0124] Generally speaking, the spatial and temporal attention layers of the dual attention spatio-temporal graph neural network extract features from the multi-modal perception data in the spatial and temporal dimensions respectively, and can accurately capture the distribution law of bacterial transmission in space and the change trend in time, providing strong support for accurately evaluating the bacterial transmission risk.

[0125] Furthermore, by generating a bacterial transmission risk prediction map with confidence evaluation, the breeding personnel can intuitively understand the bacterial transmission risk levels in different regions and at different times, and know the credibility of the prediction results, which helps to take targeted prevention and control measures in advance and reduce the risk of chicken flocks being infected with bacterial diseases.

[0126] Specifically, in the spatial attention layer, when constructing the chicken coop topological graph, determine the edge weights by combining the ventilation duct connection relationship and the historical bacterial transmission event data, making the model more in line with the actual bacterial transmission situation and improving the accuracy of the bacterial transmission path and risk assessment.

[0127] Specifically, dynamically adjusting the initial edge weights based on the historical transmission data correction factor can adapt to the changes in the chicken coop environment and the changes in the bacterial transmission law, making the model have better adaptability and robustness, and maintaining a high risk assessment accuracy for a long time.

[0128] Specifically, the calculation results of the spatial attention layer and the temporal attention layer complement each other and jointly constitute a prediction map of the risk of bacterial transmission. The spatial attention layer determines the spatial transmission possibility of bacteria and high-risk paths, and the temporal attention layer analyzes the variation law of the risk of bacterial transmission over time. The combination of the two can more comprehensively evaluate the risk of bacterial transmission, laying a foundation for formulating prevention and control strategies based on the risk assessment results.

[0129] Specifically, the spatial attention layer and the temporal attention layer have clear division of labor. Compared with conventional spatio-temporal graph neural networks, this structure can more effectively extract the spatio-temporal features of bacterial transmission. Conventional networks may not be able to simultaneously consider the complex relationships in both the spatial and temporal dimensions, while the present invention deeply analyzes data from spatial and temporal perspectives through two specialized attention layers, improving the accuracy and efficiency of the risk assessment of bacterial transmission.

[0130] Specifically, in the spatial attention layer, the edge weights are determined based on the connection relationship of ventilation ducts and historical bacterial transmission event data, and are dynamically adjusted through a specific formula. This method fully considers the actual physical environment in the chicken coop and the history of bacterial transmission, enabling the model to more realistically reflect the possibility of bacterial transmission. In contrast, traditional methods do not consider the impact of actual factors such as ventilation on bacterial transmission.

[0131] Specifically, multiple technologies such as graph convolution, LSTM, Softmax function, and Sigmoid activation function are integrated into a network structure, and a prediction map of the risk of bacterial transmission with confidence assessment is generated through the collaborative work of each part.

[0132] S3: Input the prediction map of the risk of bacterial transmission into the digital twin platform, and obtain a CFD-BIM coupled dynamic simulation model by fusing the BIM model and CFD to simulate the dynamic simulation environment of the chicken coop.

[0133] In the embodiment of the present invention, the simulation of the dynamic simulation environment of the chicken coop by fusing the BIM model and CFD includes:

[0134] Discretize the three-dimensional geometric structure of the chicken coop in the BIM model into an unstructured CFD calculation grid, with the chicken cage layout in the chicken coop as the solid obstacle boundary, the ventilation port position as the velocity inlet boundary, and the pressure outlet boundary;

[0135] Simulate the aerosol diffusion field in the chicken coop by solving the Navier-Stokes equation, and map the bacterial concentration in the prediction map of the risk of bacterial transmission to the grid nodes in the unstructured CFD calculation grid;

[0136] In the BIM visualization interface, the dynamic diffusion particle trajectories of the CFD simulation are superimposed on the BIM model, and the risk levels of different areas in the chicken coop are rendered in the form of a heat map.

[0137] Specifically, the digital twin platform is a digital technology that digitally models a physical entity (such as a chicken coop) to reflect its state, behavior, and performance in real time. In the present invention, it is used to integrate and process the bacterial transmission risk prediction map, the BIM model, and the CFD simulation results, providing a virtual environment for subsequent analysis and decision-making.

[0138] Specifically, the BIM model (Building Information Model) is a model that stores detailed information such as the geometric structure of the chicken coop, the layout of chicken cages, and the positions of ventilation openings in a three-dimensional digital form. It not only contains geometric data but also integrates various attribute information related to the chicken coop, providing a basic framework for CFD simulation and visualization.

[0139] Specifically, CFD (Computational Fluid Dynamics) is a numerical calculation method for solving the control equations of fluid flow to simulate physical phenomena such as fluid flow, heat transfer, and mass transfer. In the chicken coop simulation scenario, it is used to study the air flow and the diffusion of aerosols (including bacteria) in the chicken coop.

[0140] Specifically, the unstructured CFD computational grid divides the continuous computational domain into discrete grid cells in the CFD simulation. Compared with structured grids, unstructured grids can better adapt to complex geometric shapes. In the present invention, it is used to discretize the three-dimensional geometric structure of the chicken coop in the BIM model for CFD numerical calculation.

[0141] Specifically, the Navier-Stokes equation is the equation of motion that describes the momentum conservation of viscous incompressible fluids and is the basic equation for CFD simulation. By solving this equation, the distributions of physical quantities such as fluid velocity and pressure can be obtained, and then the diffusion field of aerosols in the chicken coop can be simulated.

[0142] Specifically, the aerosol diffusion field refers to the distribution and diffusion of aerosols (including bacteria) in the air in the chicken coop, which is one of the important results of CFD simulation and reflects the transmission path and concentration change of bacteria in the air.

[0143] Specifically, the heat map is a data visualization method that represents the size or intensity of data through the depth of different colors. In the present invention, it is used to intuitively display the bacterial transmission risk levels of different areas in the chicken coop, facilitating users to quickly understand the risk distribution.

[0144] Specifically, using a professional mesh generation software or the mesh generation function in a CFD simulation tool, the three-dimensional geometric structure of the chicken coop in the BIM model is transformed into an unstructured CFD calculation mesh. According to the actual layout of chicken cages and the positions of ventilation openings in the chicken coop, corresponding boundary conditions are set in the CFD simulation software. The chicken cage layout is defined as a solid obstacle boundary, and the positions of the ventilation openings are set as a velocity inlet boundary and a pressure outlet boundary respectively. For example, determine the wind speed and pressure values of the ventilation openings, as well as the shape, size, and position information of the chicken cages to ensure that the simulation environment is consistent with the actual situation.

[0145] Specifically, in the CFD simulation software, select a suitable numerical solution method (such as the finite volume method, finite element method, etc.) to solve the Navier-Stokes equations. According to the air flow characteristics and bacteria diffusion laws in the chicken coop, set relevant physical parameters (such as air density, viscosity coefficient, etc.). During the simulation, map the bacteria concentration data in the bacteria propagation risk prediction map obtained in step S2 to each node in the unstructured CFD calculation mesh according to the position correspondence of the grid nodes. Through iterative calculations, the simulation results of the aerosol diffusion field in the chicken coop are obtained, including the bacteria concentration distribution at different times and positions.

[0146] Specifically, use a BIM visualization software or the visualization module integrated with the CFD simulation software to import the dynamic diffusion particle trajectories obtained from the CFD simulation into the BIM model. According to the bacteria concentration data, set the color mapping rules for the heat map. For example, set the low concentration area as green, the high concentration area as red, and the medium concentration area as yellow, etc. Render the heat map for different areas of the chicken coop in the BIM visualization interface according to the set rules to visually present the bacteria propagation risk level.

[0147] Specifically, the step of obtaining the CFD-BIM coupled dynamic simulation model by fusing the BIM model and the CFD simulation of the dynamic simulation environment of the chicken coop further includes:

[0148] Extract the concentration gradient change rate of each grid node within the target time period based on the aerosol diffusion field, thereby constructing a spatio-temporal propagation tensor;

[0149] Locate the coordinates of the maximum probability pollution source in the chicken coop through the reverse particle tracking algorithm and the spatio-temporal propagation tensor.

[0150] Map the coordinates of the maximum probability pollution source located based on the grid nodes to the original BIM components through the local coordinate system transformation matrix of the BIM model;

[0151] Retrieve the subset of operation logs associated with the original BIM components, and associate the coordinates of the maximum probability pollution source with the subset of operation logs to obtain a propagation traceability analysis report.

[0152] In detail, the concentration gradient change rate is the rate of change of the bacterial concentration at a certain grid node in the aerosol diffusion field over space and time. It reflects the diffusion trend of bacteria at different locations and times and is an important basis for constructing the space-time propagation tensor.

[0153] In detail, the space-time propagation tensor is a mathematical quantity containing both time and space information, which is constructed from the rate of change of concentration gradients and is used to describe the propagation characteristics of bacteria in the space-time dimension, providing data support for the inverse particle tracking algorithm.

[0154] In detail, the reverse particle tracking algorithm is an algorithm that determines the source of particles by tracking their motion trajectory in the flow field. In the present invention, according to the spatiotemporal propagation tensor, the maximum probability contamination source coordinates in the chicken house are located by reverse tracking from the current bacterial distribution state.

[0155] In detail, the local coordinate system conversion matrix is used to convert the coordinates of the maximum probability pollution source based on grid node positioning from the local coordinate system of the CFD calculation grid to the coordinate system of the BIM model, realizing the mapping of the pollution source coordinates between different models.

[0156] In detail, the operation log subset is a data set that records operation information associated with the original BIM component, including information such as time, personnel, and operation content related to the operation, maintenance, and cleaning of the chicken house, which is used for traceability analysis.

[0157] In detail, the concentration gradient change rate data of each grid node in the target period are extracted from the aerosol diffusion field data obtained by CFD simulation. Based on these data, the space-time propagation tensor is constructed according to the construction method of the space-time propagation tensor (such as combining the concentration gradient change rate at different times and in different directions into a tensor form). Using the reverse particle tracking algorithm, starting from the current bacterial concentration distribution, along the direction opposite to the bacterial diffusion, according to the information provided by the space-time propagation tensor, the movement trajectory of the particles is gradually tracked, and finally the coordinates of the maximum probability pollution source in the chicken house are located. Through the local coordinate system conversion matrix of the BIM model, the pollution source coordinates are converted from the local coordinate system of the CFD calculation grid to the coordinate system of the BIM model. In the database, according to the identification information of the original component in the BIM model, the subset of operation logs associated with it is retrieved. The coordinates of the maximum probability pollution source are associated with the time, operation content and other information in the operation log for analysis. For example, if a feed delivery operation is performed at a certain time and location, and the location is close to the pollution source coordinates, then it can be inferred that the feed delivery operation may be related to bacterial transmission. In this way, a transmission traceability analysis report is generated to provide a basis for preventing and controlling bacterial transmission.

[0158] Generally speaking, integrating the BIM model with CFD simulation can comprehensively consider various factors such as the geometric structure of the chicken house, the layout of chicken cages, and ventilation conditions, and truly simulate the air flow and bacteria diffusion in the chicken house, providing a more reliable basis for the assessment of bacteria transmission risks; by rendering the risk level in the form of a heat map in the BIM visualization interface and superimposing the dynamic diffusion particle trajectories of CFD simulation, it enables breeders to intuitively understand the degree of bacteria transmission risks and the diffusion process in different areas of the chicken house, facilitating the timely discovery of high-risk areas and taking targeted prevention and control measures; by constructing a spatio-temporal propagation tensor and using the reverse particle tracking algorithm to locate the most probable pollution source, and combining with the BIM model and operation logs for traceability analysis, it is possible to identify the source of bacteria transmission, help breeders understand the reasons for bacteria transmission, and thus prevent bacteria transmission from the source and improve the biological safety management level of the chicken house.

[0159] In this step, sub-steps such as discretization, CFD simulation, visualization, pollution source location, and traceability analysis are interrelated and gradually in-depth. Discretization and boundary setting provide the basic conditions for CFD simulation, the results of CFD simulation are used for visualization and constructing a spatio-temporal propagation tensor, visualization helps users intuitively understand the risk situation, while pollution source location and traceability analysis provide the key basis for the formulation of prevention and control strategies, and the whole process forms a complete simulation and analysis chain.

[0160] Generally speaking, the BIM model is combined with CFD simulation and applied in the field of chicken house bacteria transmission risk assessment. This integration method breaks through the limitations of traditional single technologies, gives full play to the advantages of BIM in geometric modeling and information integration, as well as the capabilities of CFD in simulating fluid flow and diffusion, realizes the accurate simulation and analysis of the dynamic simulation environment of the chicken house, and provides a new technical means for the assessment of bacteria transmission risks.

[0161] Generally speaking, a method is proposed to construct a spatio-temporal propagation tensor based on the aerosol diffusion field, combine the reverse particle tracking algorithm for pollution source location, and associate with the BIM model and operation logs for traceability analysis. This method can accurately identify the source of bacteria transmission, is innovative in the field of chicken house bacteria prevention and control, is different from traditional simple speculation or empirical judgment methods, and provides a scientific basis for precise prevention and control.

[0162] Generally speaking, by superimposing the results of CFD simulation in the BIM visualization interface and rendering the risk level with a heat map, it provides users with intuitive and clear visualization results of risk assessment. This visualization method not only facilitates breeders to understand complex simulation data but also provides direct support for decision-making, and is innovative in the visualization and decision-making support of chicken house bacteria transmission risk assessment.

[0163] S4: Construct a deep reinforcement learning decision-making model based on the CFD-BIM coupled dynamic simulation model.

[0164] In the embodiment of the present invention, the CFD-BIM coupled dynamic simulation model integrates the advantages of computational fluid dynamics (CFD) and building information modeling (BIM). By simulating the airflow field and bacteria diffusion in the chicken coop through CFD, and combining the three-dimensional structure, layout and other information of the chicken coop provided by BIM, a model that can truly reflect the dynamic environment of the chicken coop is constructed. It provides data support close to the actual scenario for the deep reinforcement learning decision-making model.

[0165] Specifically, the deep reinforcement learning decision-making model is a model constructed based on the reinforcement learning theory. By interacting with the environment (here refers to the chicken coop simulation environment), it continuously tries different prevention and control strategies (actions), and learns the optimal bacteria prevention and control strategy according to the reward signal feedback from the environment, so as to achieve the purpose of effectively controlling the spread of bacteria under various constraints.

[0166] Specifically, according to the data provided by the CFD-BIM coupled dynamic simulation model, determine the dimension and specific characteristics of the state space. For example, the wind speed and direction in the real-time airflow field data, as well as the bacteria concentration and distribution area in the bacteria diffusion data, etc. are used as the characteristics of the state space. Then, define the action space and clarify all possible multi-dimensional prevention and control measure combinations. For example, the fan speed can be set to several different gears, and there are also various options for the opening time and range of the spray disinfection equipment. Combine these different prevention and control measure parameters to form the action space. Then, set the balance weights in the multi-objective reward function according to actual needs and experience. Finally, select a suitable deep reinforcement learning algorithm (such as deep Q network, policy gradient algorithm, etc.), build the network structure of the deep reinforcement learning decision-making model, and complete the construction of the model.

[0167] Specifically, when the deep reinforcement learning decision-making model starts to run, at each time step, the model selects a prevention and control strategy (action) from the action space according to the current state space (real-time airflow field data and bacteria diffusion data). After executing this action, calculate the reward value according to the multi-objective reward function. For example, if after taking a certain prevention and control measure, the risk level is significantly reduced, and the cost consumption is within an acceptable range, the prevention and control delay time is short, and the overlap degree between the action area of the prevention and control measure and the actual biological pollution area is high, then the reward value will be large. The model adjusts its own strategy according to the reward value, continuously tries different actions, and gradually optimizes the bacteria prevention and control strategy. When the model converges to a better strategy, generate specific prevention and control execution instructions according to the current environmental state, such as "adjust the fan speed to gear X, turn on the spray disinfection equipment, the disinfection time is Y minutes, and the disinfection range covers area Z".

[0168] In the embodiments of the present invention, the real-time airflow field data and bacteria diffusion data in the dynamic simulation environment are used as the basic inputs of the state space.

[0169] Specifically, in reinforcement learning, the state space represents the perception of the environment by the agent (here, the deep reinforcement learning decision model). In the present invention, the real-time airflow field data and bacteria diffusion data provided by the CFD-BIM coupled dynamic simulation model are used as the basic inputs of the state space. These data describe the real-time situation of bacteria propagation in the chicken coop, enabling the model to understand the current environmental state.

[0170] In the embodiments of the present invention, the multi-dimensional prevention and control measure combination is used as the action space.

[0171] Specifically, the action space refers to the set of all possible actions that the agent can take in the environment. In the scenario of bacteria prevention and control in the chicken coop, the multi-dimensional prevention and control measure combination constitutes the action space, such as the combination ways of different prevention and control measures like adjusting the fan speed, controlling the opening time and range of the spray disinfection equipment, etc.

[0172] In the embodiments of the present invention, the bacteria prevention and control strategy is optimized through a multi-objective reward function, thereby generating the bacteria prevention and control execution instructions for the chicken coop.

[0173] Specifically, the multi-objective reward function is a mathematical function used to guide the deep reinforcement learning decision model to learn the optimal strategy. It comprehensively considers multiple objectives, such as risk reduction, cost consumption, prevention and control delay time, and the accuracy of the coverage range of prevention and control measures, etc. By performing weighted summation on these factors, it provides a reward value for each action of the model, prompting the model to search for the optimal prevention and control strategy that achieves a balance among multiple objectives.

[0174] Specifically, the multi-objective reward function is designed as:

[0175] ;

[0176] Wherein, is the multi-objective reward function value, is the balance weight for adjusting risk reduction and cost consumption, is the negative impact weight of the control delay time on the reward, is the reward weight for adjusting the accuracy of the coverage range of prevention and control measures, is the amount of risk level reduction, is the cost of prevention and control measures, is the prevention and control delay time, is the actual biological pollution area, is the area where the prevention and control measures act, is the overlap degree between the actual biological pollution area and the area where the prevention and control measures act.

[0177] Specifically, in the multi-objective reward function, it is used to adjust the balance between risk reduction and cost consumption, it is used to control the degree of negative impact of the delay time on the reward, it is used to adjust the contribution of the accuracy of the coverage range of prevention and control measures to the reward. These weight values are set according to actual needs and experience to ensure that the model makes a reasonable trade-off between different objectives.

[0178] Specifically, the amount of risk level reduction represents the degree of decrease in the risk level of bacterial transmission in the chicken house after taking a certain prevention and control measure. The risk level is usually comprehensively evaluated based on factors such as bacterial concentration and transmission speed. The greater the amount of risk level reduction, the better the effect of the prevention and control measure in reducing the risk.

[0179] Specifically, the cost of prevention and control measures is the resources required to implement various bacterial prevention and control measures, including equipment operation cost, disinfectant cost, labor cost, etc. When optimizing the prevention and control strategy, it is necessary to consider the cost factor while reducing the risk to achieve the balance between economic benefits and prevention and control effects.

[0180] Specifically, the prevention and control delay time is the time elapsed from the decision to take prevention and control measures to the actual implementation of the measures. The longer the delay time, the more widely the bacteria may spread, and the greater the threat to the health of the chicken flock. Therefore, in the reward function, it is necessary to punish the delay time to prompt the model to take effective prevention and control measures as soon as possible.

[0181] Specifically, the actual biological pollution area refers to the area in the chicken house that is actually contaminated by bacteria after taking prevention and control measures. This is a dynamically changing area that will change with the implementation of prevention and control measures and the spread of bacteria.

[0182] Specifically, the area affected by prevention and control measures refers to the area covered and affected by prevention and control measures (such as spray disinfection, ventilation, etc.). Ideally, the area affected by prevention and control measures should coincide with the actual biological pollution area as much as possible to improve the prevention and control effect.

[0183] Specifically, the overlap degree (IoU), that is, the intersection over union, is used to measure the overlap degree between the actual biological pollution area and the area affected by prevention and control measures. The higher the IoU value, the better the accuracy of the coverage range of prevention and control measures, that is, the prevention and control measures can act more effectively on the contaminated area.

[0184] In the embodiment of the present invention, after the prevention and control execution instruction is fed back to the digital twin platform for virtual verification, a visual decision report is output.

[0185] Specifically, feeding back the prevention and control execution instruction to the digital twin platform for virtual verification includes:

[0186] Setting a Monte Carlo simulation module in the digital twin platform to randomly perturb environmental parameters to evaluate the robustness of the bacteria prevention and control strategy;

[0187] Issuing the verified prevention and control execution instruction to the fan and spray disinfection equipment controller in the chicken coop through the OPC-UA protocol, and real-time collecting the environmental feedback data after implementing the bacteria prevention and control strategy to update the input data of the dual attention spatio-temporal graph neural network in S2.

[0188] Specifically, the digital virtual model corresponding to the digital twin platform and the actual chicken coop can reflect the physical state, operation conditions, etc. of the chicken coop in real time. In the present invention, it is used to perform virtual verification on the generated prevention and control execution instruction, simulate the execution effect of the instruction in the actual chicken coop, and evaluate the feasibility and effectiveness of the prevention and control strategy.

[0189] Specifically, the simulation module based on the Monte Carlo method evaluates the performance and uncertainty of the system through multiple random samplings and simulation experiments. Setting this module in the digital twin platform, by randomly perturbing environmental parameters (such as temperature, humidity, wind speed, etc.), different environmental conditions are simulated to test the robustness of the bacteria prevention and control strategy in various situations, that is, whether the effectiveness of the strategy is stable and whether it can adapt to environmental changes.

[0190] Specifically, the OPC-UA protocol is a communication protocol used in the field of industrial automation, with advantages such as platform independence, security, and interoperability. In the present invention, it is used to issue the prevention and control execution instruction verified through virtual verification from the digital twin platform to the fan and spray disinfection equipment controller in the chicken coop, realizing remote control of actual equipment and ensuring the accuracy and reliability of data transmission.

[0191] Specifically, feeding back the generated prevention and control execution instruction to the digital twin platform and starting the Monte Carlo simulation module. During the simulation process, randomly change environmental parameters (such as the temperature fluctuates within a certain range, the wind speed is randomly adjusted, etc.) to simulate different actual environmental situations. Observe the execution effect of the prevention and control strategy in these different environments, calculate relevant indicators (such as risk level change, cost consumption, overlap degree, etc.) to evaluate the robustness of the prevention and control strategy. If the prevention and control strategy can show good effects in multiple simulated environments, it indicates that the strategy has high robustness. The verified prevention and control execution instruction is issued to the fan and spray disinfection equipment controller in the chicken coop through the OPC-UA protocol to control the actual equipment to implement corresponding prevention and control measures.

[0192] Specifically, during the implementation of prevention and control measures, environmental feedback data is collected in real time through sensors installed in the chicken coop. For example, a bacteria concentration sensor measures changes in bacteria concentration, and an air velocity sensor monitors changes in the airflow field. The collected data is transmitted back to the system to update the input data of the dual-attention spatio-temporal graph neural network in step S2. The dual-attention spatio-temporal graph neural network re-performs spatio-temporal feature extraction based on the new input data to generate a more accurate prediction map of bacteria transmission risk. Then, the deep reinforcement learning decision model continues to optimize the prevention and control strategy based on the updated risk prediction map and new environmental feedback data, realizing the dynamic cyclic optimization of the entire risk assessment and prevention and control system.

[0193] Generally speaking, by constructing a deep reinforcement learning decision model and using a multi-objective reward function to comprehensively consider factors such as risk reduction, cost consumption, prevention and control delay time, and accuracy of the coverage of prevention and control measures, the optimal bacteria prevention and control strategy can be found in a complex chicken coop environment, reducing costs while improving the prevention and control effect, and achieving a balance between economic benefits and biosecurity. By using the Monte Carlo simulation module in the digital twin platform to randomly perturb environmental parameters for virtual verification, the effectiveness of the prevention and control strategy under different environmental conditions can be tested, ensuring that the prevention and control strategy has high robustness, can adapt to changes in the actual chicken coop environment, and improve the reliability of prevention and control.

[0194] Generally speaking, using the environmental feedback data after implementing the prevention and control strategy to update the input data of the dual-attention spatio-temporal graph neural network, and then optimizing the deep reinforcement learning decision model, enables the entire risk assessment and prevention and control system to continuously adjust and optimize according to the actual situation, continuously improving the prevention and control ability of the bacteria transmission risk in the chicken coop. By remotely issuing prevention and control execution instructions through the OPC-UA protocol, it is possible to automatically control the fans, spray disinfection equipment, etc. in the chicken coop, improving the timeliness and accuracy of the implementation of prevention and control measures, reducing manual intervention, and enhancing the automation level of breeding production.

[0195] Specifically, relying on the real-time airflow field data and bacteria diffusion data provided by the CFD-BIM coupled dynamic simulation model as the input of the state space, and at the same time, the prevention and control execution instructions generated by the deep reinforcement learning decision model are fed back to the digital twin platform for virtual verification. The verified environmental feedback data is then used to update the input data of the dual-attention spatio-temporal graph neural network in step S2, forming a data cycle and interdependent relationship to realize the dynamic operation of the entire risk assessment and prevention and control system.

[0196] Specifically, a deep reinforcement learning decision model is constructed to optimize the bacterial prevention and control strategy and generate execution instructions. Virtual verification is used to test the effectiveness and robustness of these instructions in different environments. Instruction issuance and environmental feedback data collection are to apply the optimized strategy to the actual situation and obtain actual effect data for further optimizing the model. Each link is closely connected and logically progressive, jointly achieving effective prevention and control of the risk of bacterial transmission in the chicken coop.

[0197] Generally speaking, combining CFD digital twin with the optimization of the deep reinforcement learning prevention and control strategy forms a complete decision-making chain from perception (collecting data through sensors and generating risk prediction maps), simulation (CFD-BIM coupled dynamic simulation model), decision-making (deep reinforcement learning decision model), execution (issuing prevention and control execution instructions), to feedback optimization (updating the model according to environmental feedback), realizing a closed-loop system from perception to execution.

[0198] Generally speaking, the designed multi-objective reward function comprehensively considers multiple key factors such as risk reduction, cost consumption, prevention and control delay time, and the accuracy of the coverage range of prevention and control measures. Through the setting of balancing weights, the deep reinforcement learning decision model can make reasonable trade-offs among multiple objectives and find the optimal prevention and control strategy.

[0199] Generally speaking, the Monte Carlo simulation module in the digital twin platform is used to randomly perturb environmental parameters to evaluate the robustness of the bacterial prevention and control strategy. Traditional methods may be difficult to comprehensively consider various uncertainty factors in the actual environment, while the present invention can more comprehensively test the effectiveness of the prevention and control strategy by simulating a variety of different environmental conditions, ensuring that the strategy can also play a good role in the complex and changeable actual environment.

[0200] As Figure 2 shown, it is a functional module diagram of the bacterial transmission risk assessment system in the chicken coop provided by an embodiment of the present invention.

[0201] The bacterial transmission risk assessment system 100 of the present invention can be installed in an electronic device. According to the functions implemented, the bacterial transmission risk assessment system 100 can include a data collection module 101, a feature extraction module 102, a dynamic simulation module 103, and a decision model module 104. The modules of the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by the processor of the electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0202] In this embodiment, the functions of each module / unit are as follows:

[0203] The data collection module 101 is used to collect multi-modal perception data of bacteria in the chicken coop in real time through a distributed sensor array;

[0204] The feature extraction module 102 is configured to perform spatio-temporal feature extraction on the multi-modal perception data by using a dual attention spatio-temporal graph neural network, and generate a bacterial transmission risk prediction map with confidence evaluation;

[0205] The dynamic simulation module 103 is configured to input the bacterial transmission risk prediction map into a digital twin platform, and obtain a CFD-BIM coupled dynamic simulation model by fusing a BIM model and CFD to simulate the dynamic simulation environment of the chicken coop;

[0206] The decision model module 104 is configured to construct a deep reinforcement learning decision model based on the CFD-BIM coupled dynamic simulation model, thereby generating a bacterial prevention and control execution instruction for the chicken coop. After the prevention and control execution instruction is fed back to the digital twin platform for virtual verification, a visual decision report is output.

[0207] In several embodiments provided by the present invention, it should be understood that the disclosed method and system can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0208] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0209] In addition, the functional modules in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware, or in the form of a hardware plus software functional module.

[0210] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0211] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence is a theory, method, technology and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for evaluating the risk of bacterial transmission in a chicken coop, characterized in that, The method includes: S1: Real-time collect multi-modal perception data of bacteria in the chicken coop through a distributed sensor array; S2: Use a dual attention spatio-temporal graph neural network to extract spatio-temporal features from the multi-modal perception data, and generate a bacteria transmission risk prediction map with confidence evaluation; S3: Input the bacteria transmission risk prediction map into the digital twin platform, and simulate the dynamic simulation environment of the chicken coop by fusing the BIM model and CFD to obtain a CFD-BIM coupled dynamic simulation model; S4: Build a deep reinforcement learning decision-making model based on the CFD-BIM coupled dynamic simulation model, where: Use the real-time airflow field data and bacteria diffusion data in the dynamic simulation environment as the basic input of the state space; Use the multi-dimensional prevention and control measure combination as the action space; Optimize the bacteria prevention and control strategy through a multi-objective reward function, so as to generate a bacteria prevention and control execution instruction for the chicken coop; After feeding back the prevention and control execution instruction to the digital twin platform for virtual verification, output a visual decision report.

2. The method for evaluating the risk of bacterial transmission in a chicken coop according to claim 1, wherein The distributed sensor array includes: A miniaturized surface-enhanced Raman spectroscopy sensor for real-time capturing the characteristic spectral data of bacteria in the air in the chicken coop; A microfluidic sampling chip modified with magnetic nanomaterials for adsorbing bacteria on the inner surface of the chicken coop and recording the spatio-temporal coordinates of the sampling points; An environmental parameter monitoring module integrated with a temperature and humidity sensor, an ammonia concentration sensor, and a wind speed sensor for collecting the environmental parameters in the chicken coop, and the data of each sensor is encrypted and transmitted through the LoRa self-organizing network protocol.

3. The method for evaluating the risk of bacterial transmission in a chicken coop according to claim 1, characterized in that, The dual attention spatio-temporal graph neural network includes: a spatial attention layer and a temporal attention layer, where: The spatial attention layer calculates the bacteria transmission path weights between sensor nodes in the distributed sensor array based on a path weight generation algorithm, where the path weight generation algorithm is as follows: ； Among them, is the weight of the bacterial transmission path, is the normalization function, is the value matrix, is the dimension scaling factor, is the query matrix, is the key matrix, is the transpose of the key matrix; The temporal attention layer extracts the temporal dependence relationship of bacteria transmission risk through a temporal dependence generation algorithm, where the temporal dependence generation algorithm is: ； Among them, is the temporal dependence relationship, is the long short-term memory network, is the Hadamard product, is the multi-modal perception data, is the weight matrix, is the bias vector, is the time step index, is the Sigmoid activation function.

4. The method for evaluating the risk of bacterial transmission in a chicken coop according to claim 3, wherein The spatial attention layer further includes: Construct a topological graph of the chicken coop, where the topological nodes are the deployment positions of the sensors in the distributed sensor array, and the edge weights are jointly determined by the connection relationship of the ventilation ducts in the chicken coop and the historical bacteria transmission event data; Perform weighted fusion on the bacteria transmission path weights and the graph convolution output features to obtain a high-risk path identification matrix of the connection paths between the sensor nodes.

5. The method for evaluating the risk of bacterial transmission in a chicken coop according to claim 1, wherein The simulation of the dynamic simulation environment of the chicken coop by fusing the BIM model and CFD includes: Discretize the three-dimensional geometric structure of the chicken coop in the BIM model into an unstructured CFD calculation grid, use the chicken cage layout in the chicken coop as the solid obstacle boundary, and the ventilation port position as the velocity inlet boundary and the pressure outlet boundary; Simulate the aerosol diffusion field in the chicken coop by solving the Navier-Stokes equation, and map the bacteria concentration in the bacteria transmission risk prediction map to the grid nodes in the unstructured CFD calculation grid; In the BIM visualization interface, the dynamic diffusion particle trajectories of the CFD simulation are superimposed on the BIM model, and the risk levels of different areas in the chicken house are rendered in the form of a heat map.

6. The method for evaluating the risk of bacterial transmission in a chicken coop according to claim 1, wherein The multi-objective reward function is designed as: ； Among them, is the multi-objective reward function value, is the balance weight for adjusting the reduction of risk and the consumption of cost, is the negative impact weight of the control delay time on the reward, is the reward weight for adjusting the accuracy of the coverage range of prevention and control measures, is the reduction amount of the risk level, is the cost of prevention and control measures, is the prevention and control delay time, is the actual biological pollution area, is the area where the prevention and control measures act, is the overlap degree between the actual biological pollution area and the area where the prevention and control measures act.

7. The method for evaluating the risk of bacterial transmission in a chicken coop according to claim 4, characterized in that The method for determining the edge weights includes: Based on the ventilation duct connection relationship, an initial adjacency matrix is constructed, where the initial edge weight between adjacent nodes in the initial adjacency matrix is the product of the duct cross-sectional area and the real-time wind speed in the ventilation duct; Based on the historical bacteria transmission event data, a historical transmission data correction factor for the initial edge weight is generated, and the initial edge weight is dynamically adjusted based on the historical transmission data correction factor, where the calculation formula for the historical transmission data correction factor is: ； Among them, is the balance factor for adjusting the physical properties of the ventilation duct and the data of the historical bacterial transmission event, is the cross-sectional area of the duct, is the real-time wind speed in the ventilation duct, is the number of times bacteria spread from node to node times, is the total number of transmission events, is the th topological node in the topological graph, is the th topological node in the topological graph.

8. The method for evaluating the risk of bacterial transmission in a chicken coop according to claim 1, wherein, The feedback of the prevention and control execution instruction to the digital twin platform for virtual verification includes: Setting a Monte Carlo simulation module in the digital twin platform to randomly perturb environmental parameters to evaluate the robustness of the bacteria prevention and control strategy; Through the OPC-UA protocol, the verified prevention and control execution instruction is sent to the fan and spray disinfection equipment controller in the chicken house, and the environmental feedback data after implementing the bacteria prevention and control strategy is collected in real time to update the input data of the dual attention spatio-temporal graph neural network in S2.

9. The method for assessing the risk of bacterial transmission in a chicken coop according to claim 5, wherein The obtaining of the CFD-BIM coupled dynamic simulation model by fusing the BIM model and the CFD simulation of the dynamic simulation environment of the chicken house further includes: Based on the aerosol diffusion field, the concentration gradient change rate of each grid node in the target time period is extracted, thereby constructing a spatio-temporal propagation tensor; Using the reverse particle tracking algorithm and the spatio-temporal propagation tensor to locate the coordinates of the most probable pollution source in the chicken house; Mapping the coordinates of the most probable pollution source located based on the grid nodes to the original BIM component through the local coordinate system transformation matrix of the BIM model; Retrieving the subset of operation logs associated with the original BIM component, and associating the coordinates of the most probable pollution source with the subset of operation logs to obtain a propagation traceability analysis report.

10. A bacterial transmission risk assessment system in a chicken coop, characterized in that, The system includes: A data acquisition module for real-time collecting multi-modal perception data of bacteria in the chicken house through a distributed sensor array; A feature extraction module for performing spatio-temporal feature extraction on the multi-modal perception data using a dual attention spatio-temporal graph neural network to generate a bacteria transmission risk prediction map with confidence evaluation; A dynamic simulation module for inputting the bacteria transmission risk prediction map into the digital twin platform, and obtaining a CFD-BIM coupled dynamic simulation model by fusing the BIM model and the CFD simulation of the dynamic simulation environment of the chicken house; A decision model module for constructing a deep reinforcement learning decision model based on the CFD-BIM coupled dynamic simulation model, where: Using the real-time airflow field data and bacteria diffusion data in the dynamic simulation environment as the basic input of the state space; Using a multi-dimensional prevention and control measure combination as the action space; Optimizing the bacteria prevention and control strategy through a multi-objective reward function to generate a bacteria prevention and control execution instruction for the chicken house. After feeding back the prevention and control execution instruction to the digital twin platform for virtual verification, a visual decision-making report is output.

Images