Livestock breeding risk prediction and decision support system based on big data analysis

Abstract

The application discloses a kind of based on big data analysis's livestock breeding risk prediction and decision support system, it is related to livestock breeding environment monitoring technical field, including: environmental digital twin modeling module, for by fusing multisource sensor data and three-dimensional modeling technique, construct the three-dimensional pollutant diffusion model of farm, and carry out the construction of pollutant propagation path atlas, realize the dynamic migration and distribution visualization of pollutant in breeding environment.The application realizes the dynamic simulation and analysis of the migration, interaction and coupling effect of multiple pollutants such as ammonia nitrogen, chemical oxygen demand and veterinary antibiotics by constructing three-dimensional pollutant diffusion model and pollutant concentration map, carries out time series prediction combined with deep learning model, identifies composite pollution risk in advance and grades early warning, overcome the limitations of traditional single index monitoring, improve the prediction accuracy and response timeliness of pollution event.

Description

Technical Field

[0001] This invention relates to the field of livestock farming environment monitoring technology, specifically a livestock farming risk prediction and decision support system based on big data analysis. Background Technology

[0002] As an important part of agricultural production, livestock farming faces a variety of risks, including natural disasters, disease outbreaks, market fluctuations, and policy changes. These risks not only affect the economic benefits of the livestock industry but also pose potential threats to food safety and the ecological environment. Therefore, implementing an effective risk prediction and decision support system is particularly important.

[0003] For example, the method for monitoring sewage discharge data in livestock farms, published in Chinese Patent Publication No. CN119067479A, relates to the field of sewage discharge data monitoring technology and can accurately predict the water pollution situation of today and the next day.

[0004] In existing technologies, prediction methods based on the similarity of historical monitoring curves are prone to prediction distortion when faced with complex scenarios involving the nonlinear coupling and diffusion of multiple pollutants in livestock farms. In particular, they are unable to effectively address the complex water pollution problems caused by the sudden interaction of pollutants such as drug residues, disinfectant intervention, and feces. Furthermore, after identifying the risks of complex pollution, there is a lack of dynamic optimization capabilities for emission reduction strategies. It is difficult to adaptively adjust the wastewater treatment process based on real-time water quality fluctuations, equipment status, and multiple constraints, resulting in a disconnect between risk warning and disposal, and an inability to form closed-loop decision support. To address these issues, a livestock farming risk prediction and decision support system based on big data analysis is proposed. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a livestock farming risk prediction and decision support system based on big data analysis, comprising:

[0006] The environmental digital twin modeling module is used to construct a three-dimensional pollutant diffusion model of a farm by integrating multi-source sensor data and three-dimensional modeling technology, and to construct a pollutant propagation path map, so as to realize the dynamic migration and distribution visualization of pollutants in the aquaculture environment, and provide high-precision simulated environment support for subsequent risk prediction.

[0007] The multi-pollutant coupling prediction module is used to establish a pollutant concentration map by introducing knowledge graph technology on the basis of a three-dimensional pollutant diffusion model, and to integrate the nonlinear interaction between multiple pollutants and external intervention factors to construct a multi-pollutant compound pollution prediction model, analyze the trend of pollutant concentration changes, and achieve accurate prediction from single pollutant indicators to multi-pollutant compound pollution.

[0008] The emission reduction decision optimization module is used to generate adaptive emission reduction scheduling strategies and optimize the wastewater treatment process by using a deep Q-network-based reinforcement learning algorithm based on pollution prediction, with pollutant concentration maps and wastewater treatment equipment status as inputs.

[0009] The multi-constraint adaptive scheduling module is used to integrate multi-dimensional constraints such as the operating status of wastewater treatment equipment, treatment costs, and energy consumption costs, dynamically evaluate the feasibility and economy of emission reduction scheduling strategies, adjust the scheduling scheme in real time, generate the optimal emission reduction scheduling scheme, and ensure compliant and efficient decision-making.

[0010] The risk warning and response module generates risk warning information and disposal suggestions in real time based on pollution prediction and decision optimization results. It also monitors the implementation effect of emission reduction scheduling plans, collects feedback data, and continuously optimizes prediction models and decision-making strategies by combining historical execution records and environmental changes, forming a closed loop of warning-disposal-feedback-optimization, driving the system to continuously evolve.

[0011] Preferably, the environmental digital twin modeling module includes a pollutant three-dimensional diffusion simulation unit and a risk path mapping construction unit;

[0012] The pollutant three-dimensional diffusion simulation unit, based on the spatial geographic information, environmental parameters and pollutant property data of the farm, constructs a three-dimensional pollutant diffusion model, simulates the migration, diffusion and coupling process of multiple pollutants such as ammonia nitrogen, COD and drug residues under different environmental conditions in real time, outputs a dynamic pollution concentration field, and the simulation results accurately reflect the dynamic change process of multiple pollutants in a complex environment.

[0013] The risk path mapping construction unit is used to combine the pollutant diffusion simulation results analyzed by the three-dimensional pollutant diffusion model to construct a pollutant propagation path map, identify high-pollution propagation channels and key risk nodes, clearly present the main propagation paths and key nodes, and assist in precise deployment.

[0014] Preferably, the pollutant three-dimensional diffusion simulation unit includes the following steps:

[0015] We acquire three-dimensional spatial geographic information, environmental parameter data, and pollutant property data of the farm, and construct a three-dimensional pollutant diffusion model to provide a high-precision, multi-physics coupled computational foundation for subsequent dynamic simulation of pollutants.

[0016] Based on a three-dimensional pollutant diffusion model and combined with real-time sensor monitoring data, the dynamic migration process of multiple pollutants such as ammonia nitrogen, COD and drug residues under wind, water flow and temperature conditions is simulated, so as to accurately reproduce the spatiotemporal evolution process of pollutants in the actual environment.

[0017] It outputs dynamic pollution concentration fields at different altitudes and in different regions in real time, forming a spatiotemporal distribution map of pollutant concentrations, providing intuitive spatial data support for pollution diffusion visualization, risk analysis, and control decisions.

[0018] Preferably, the risk path mapping construction unit includes the following steps:

[0019] Based on the dynamic pollution concentration field, the main directions, speeds and concentration decay gradients of pollutant propagation are identified, and the spatial topology of pollution propagation is extracted, providing a structured and computable spatial network foundation for the quantitative analysis of pollutant propagation paths and risk tracing.

[0020] By combining the layout of farm facilities with the distribution of natural barriers, a pollutant transmission path map is constructed, marking the pollution intensity level and transmission risk weight of different transmission channels, intuitively revealing the interaction between pollution diffusion and physical facilities, and realizing the graded and classified refined management of risk channels;

[0021] By analyzing pollutant propagation path maps, high-pollution transmission channels and key risk nodes are identified, forming a visual layer of critical pollutant diffusion paths. This transforms abstract maps into intuitive decision-making maps, directly guiding the optimization of monitoring sites and the precise deployment of emergency interception facilities.

[0022] Preferably, the multi-pollutant coupling prediction module includes a pollutant interaction knowledge graph unit and a composite pollution risk prediction unit;

[0023] The pollutant interaction knowledge graph unit, based on a three-dimensional pollutant diffusion model and a pollutant propagation path map, combines knowledge graph technology to establish a pollutant concentration map of physical, chemical and biological interactions among multiple pollutants. It integrates external intervention factors such as drug residues and disinfectant use, dynamically updates the synergistic, antagonistic and transformation mechanisms among pollutants, enhances the model's understanding of compound pollution, dynamically updates interaction relationships, and enhances the system's cognitive depth of compound pollution.

[0024] The composite pollution risk prediction unit constructs a multi-pollutant composite pollution prediction model based on pollutant concentration maps and historical data. It analyzes the trend of pollutant concentration changes in real time, provides early warning of potential pollution events, and enables early prediction of pollution events, thus gaining valuable time for emergency response.

[0025] Preferably, the pollutant interaction knowledge graph unit includes the following steps:

[0026] Based on a three-dimensional pollutant diffusion model and pollutant propagation path map, an initial knowledge graph structure with pollutants as nodes is established, including pollutant entities such as ammonia nitrogen, COD, and disinfectants. This allows pollutant entities and their attributes to be stored and associated in a structured manner, supporting the formal definition and efficient retrieval of subsequent interaction relationships.

[0027] By integrating historical monitoring data with external intervention information, the physical, chemical, and biological interaction relationships between pollutants are defined, and the synergistic, antagonistic, and transformation mechanisms in the initial knowledge graph structure are updated. This enables a dynamic characterization of the complex interaction mechanisms between pollutants, providing an interpretable relational network basis for the analysis of compound pollution effects. Furthermore, a dynamically updated pollutant concentration map is established to reflect the nonlinear interactions and compound pollution effects of multiple pollutants in complex environments in real time. This enhances the system's understanding and response capabilities to real-time pollution conditions and supports more accurate compound pollution risk assessment and early warning decisions.

[0028] Preferably, the composite pollution risk prediction unit includes the following steps:

[0029] Based on pollutant concentration maps, a multi-pollutant compound pollution prediction model is constructed by combining deep learning models. The model input includes the current concentration of pollutants, environmental parameters and historical change trends, so as to achieve high-precision dynamic simulation and risk quantification of the synergistic and antagonistic effects of multiple pollutants.

[0030] The concentration change trends and interaction effects of each pollutant are analyzed using a multi-pollutant composite pollution prediction model. The prediction results of multi-pollutant coupled pollution and the time series diagram of concentration distribution are generated. The multi-pollutant coupled pollution prediction results are represented by a composite pollution index, providing a spatial visualization prediction view of the pollutant diffusion path and concentration evolution in the next few hours.

[0031] Based on the comparison between the model prediction results and the preset warning threshold, the system provides real-time warnings of existing complex pollution events and outputs the pollution type and spatiotemporal range, achieving graded warnings and precise positioning, and providing clear spatiotemporal targets and decision-making basis for emergency response.

[0032] Preferably, the emission reduction decision optimization module includes the following steps:

[0033] Based on pollutant concentration maps and real-time status data of wastewater treatment equipment, a reinforcement learning environment is constructed, and emission reduction control actions are set as a set of equipment adjustment commands. This achieves multi-source data fusion, provides a high-fidelity simulation environment for intelligent control, and enhances the realism and reliability of strategy generation.

[0034] The deep Q-network algorithm is adopted, with the reduction of pollutant concentration and stable equipment operation as the reward function. The model is trained to generate an adaptive emission reduction scheduling strategy, which guides the model to find the optimal balance between emission compliance and stable operation, and outputs an adaptive control strategy that takes into account both environmental protection and economy.

[0035] Output real-time emission reduction scheduling strategies, including the start-up and shutdown of wastewater treatment equipment, flow regulation and chemical dosing schemes, optimize pollutant treatment processes, achieve second-level response from decision-making to execution, improve treatment efficiency, and reduce the risk of exceeding standards and operating costs.

[0036] Preferably, the multi-constraint adaptive scheduling module includes the following steps:

[0037] Construct a multi-constraint assessment model that includes equipment operating status, processing costs, energy consumption costs and environmental compliance constraints, and establish an objective optimization function to achieve quantitative trade-offs and comprehensive optimal decision-making among multiple objectives;

[0038] Based on emission reduction scheduling strategies and real-time operational data, the feasibility and economy of each emission reduction scheduling strategy are dynamically evaluated, a comprehensive execution score is calculated, infeasible solutions are dynamically eliminated, and economically efficient strategies are focused on.

[0039] The emission reduction scheduling plan is adjusted based on the assessment results to generate the optimal emission reduction scheduling plan that meets multiple constraints, and is simultaneously updated to the wastewater treatment control system to ensure that the final plan is compliant, economical and can be implemented immediately.

[0040] Preferably, the risk warning and response module includes the following steps:

[0041] Based on the results of compound pollution prediction and the optimal emission reduction scheduling scheme, risk warning information and emergency response suggestions are generated in real time and sent out through a visual interface and message push, realizing multi-level risk second-level warning and intelligent response suggestion push, improving the speed of emergency response and the pertinence of decision-making;

[0042] Monitor changes in pollutant concentrations and equipment status during the execution of emission reduction scheduling plans, collect feedback data on execution effectiveness, and form a closed-loop tracking of the entire process of strategy-execution-monitoring to provide real-time data support for strategy optimization;

[0043] By combining historical implementation records with environmental changes, we continuously optimize pollutant prediction models and emission reduction decision-making strategies, forming a closed-loop decision support and system self-learning mechanism to promote the continuous evolution of the system in real operation and gradually improve prediction accuracy and scheduling reliability.

[0044] This invention provides a livestock farming risk prediction and decision support system based on big data analysis. It has the following beneficial effects:

[0045] (I) This livestock farming risk prediction and decision support system based on big data analysis constructs a three-dimensional pollutant diffusion model and pollutant concentration map to realize the dynamic simulation and analysis of the migration, interaction and coupling effects of multiple pollutants such as ammonia nitrogen, chemical oxygen demand and veterinary antibiotics. Combined with a deep learning model for time series prediction, it can identify compound pollution risks in advance and issue graded warnings, overcome the limitations of traditional single indicator monitoring, and improve the prediction accuracy and response time of pollution events.

[0046] (II) This livestock farming risk prediction and decision support system based on big data analysis takes digital twins and reinforcement learning as its core, and organically integrates pollution prediction, path analysis, scheduling optimization and execution feedback. Based on real-time pollution load and equipment status, it generates adaptive emission reduction scheduling strategies through deep Q networks, and forms an executable optimal control scheme through dynamic screening and adjustment of multi-constraint evaluation models. It realizes the full-process automated closed-loop management from monitoring and early warning to decision execution, and improves the systematicness and initiative of pollution prevention and control.

[0047] (III) This livestock farming risk prediction and decision support system based on big data analysis comprehensively considers multiple factors such as equipment operating status, treatment costs, energy consumption costs and environmental compliance, constructs a weighted optimization objective function, dynamically evaluates and selects the most economically optimal scheduling scheme under the premise of meeting emission standards, effectively balances environmental benefits and operating costs, ensures that the sewage treatment process achieves energy conservation and consumption reduction on the basis of compliance, and improves the overall resource utilization efficiency and sustainable operation capability of the farm. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the workflow of a livestock breeding risk prediction and decision support system based on big data analysis according to the present invention.

[0049] Figure 2 This is a data flow diagram of a livestock breeding risk prediction and decision support system based on big data analysis according to the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: a livestock farming risk prediction and decision support system based on big data analysis, comprising:

[0052] The environmental digital twin modeling module is used to construct a three-dimensional pollutant diffusion model of a farm by integrating multi-source sensor data and three-dimensional modeling technology, and to construct a pollutant propagation path map. This enables the visualization of the dynamic migration and distribution of pollutants in the aquaculture environment, providing high-precision simulated environmental support for subsequent risk prediction, realizing the realistic restoration and trend prediction of the pollution process, and improving risk perception capabilities. The environmental digital twin modeling module includes a pollutant three-dimensional diffusion simulation unit and a risk path map construction unit.

[0053] Among them, the three-dimensional pollutant diffusion simulation unit, based on the spatial geographic information, environmental parameters and pollutant property data of the farm, constructs a three-dimensional pollutant diffusion model, simulates the migration, diffusion and coupling process of multiple pollutants such as ammonia nitrogen, COD, and drug residues under different environmental conditions in real time, and outputs a dynamic pollution concentration field. The simulation results accurately reflect the dynamic change process of multiple pollutants in a complex environment. It obtains the three-dimensional spatial geographic information, environmental parameter data and pollutant property data of the farm, and constructs a three-dimensional pollutant diffusion model, providing a high-precision, multi-physics field coupled calculation basis for subsequent dynamic pollutant simulation. Based on the three-dimensional pollutant diffusion model, combined with real-time sensor monitoring data, it simulates the dynamic migration process of multiple pollutants such as ammonia nitrogen, COD and drug residues under wind, water flow and temperature conditions, realizes the accurate reproduction of the spatiotemporal evolution process of pollutants in the actual environment, and outputs dynamic pollution concentration fields at different heights and regions in real time, forming a spatiotemporal distribution map of pollutant concentration, providing intuitive spatial data support for pollution diffusion visualization, risk analysis and control decisions.

[0054] The specific work involves collecting and integrating basic geospatial data, environmental parameter data, and pollutant property data of the aquaculture farm. The basic geospatial data includes a high-precision digital elevation model, a 3D model of the farm's buildings and facilities, and a vector distribution map of the sewage network and natural water system. Environmental parameter data, covering wind speed and direction, ambient temperature, water flow velocity and flow rate, and temperature and humidity, is acquired in real time through a deployed IoT sensor network. Pollutant property data mainly includes the diffusion coefficient of ammonia nitrogen (NH3-N) in the water (1.5 × 10⁻⁶). -5 cm 2 / s to 2.0×10 -5 cm 2Based on multi-source data, a non-steady-state, multiphase three-dimensional pollutant diffusion model was constructed using computational fluid dynamics and mass transport coupling principles. The model was based on the degradation half-life of chemical oxygen demand (COD) (depending on temperature and dissolved oxygen levels) and the hydrolysis and adsorption coefficients of typical veterinary antibiotics (sulfonamides). The governing equations included the Navier-Stokes equations, convection-diffusion equations, and reaction kinetic equations. Mesh generation (mesh size from 0.5 to 2 meters) and boundary condition settings were completed during model initialization. During model operation, real-time monitoring data from a distributed water quality sensor array was continuously received via a standardized data interface, including ammonia nitrogen concentration, COD concentration, and concentrations of specific drug residue markers. The three-dimensional pollutant diffusion model used real-time monitoring data as dynamic boundary conditions and validation benchmarks to drive the simulation calculations. The simulation process fully considered environmental driving factors: wind field data drove the exchange of pollutants at the gas-liquid interface; water flow field data determined the convection and dispersion processes of pollutants in the water body. Temperature data regulates chemical reaction rates and biodegradation activity. Through multi-physics coupling solutions, it simulates in real time the migration, transformation, and distribution dynamics of target pollutants such as ammonia nitrogen, COD, and drug residues in the aquaculture farm's waters and surrounding environment, outputting transient simulation results of the entire scenario every 15 minutes. After each round of simulation calculations, it automatically generates and updates dynamic pollution concentration field data. This pollution concentration field data is organized in a three-dimensional grid, covering different height layers from the surface to the groundwater layer and the near-surface atmosphere (vertically layered at 0.5-meter intervals), and covering all key spatial units in the aquaculture farm and downstream sensitive areas. For each grid unit, it outputs the corresponding concentration values ​​of ammonia nitrogen, COD, and each target drug residue. Through the built-in visualization engine, the data is rendered in real time into a spatiotemporal distribution map of pollutant concentrations. This map supports sliding viewing on the time axis, clearly showing the spatial gradient distribution of pollutant concentrations, the diffusion trend over time, and the concentration plume morphology formed under the influence of different environmental factors.

[0055] The Risk Path Mapping Unit is used to construct a pollutant propagation path map by combining the pollutant diffusion simulation results analyzed by the 3D pollutant diffusion model. It identifies high-pollution propagation channels and key risk nodes, clearly presents the main propagation paths and key nodes, and assists in precise deployment. Based on the dynamic pollution concentration field, it identifies the main direction, speed and concentration decay gradient of pollutant propagation, extracts the spatial topology of pollution propagation, and provides a structured and computable spatial network foundation for the quantitative analysis of pollutant propagation paths and risk tracing. Combining the layout of farm facilities and the distribution of natural barriers, it constructs a pollutant propagation path map, marks the pollution intensity level and propagation risk weight of different propagation channels, intuitively reveals the interaction between pollution diffusion and physical facilities, realizes the hierarchical and classified refined management of risk channels, and determines high-pollution propagation channels and key risk nodes through pollutant propagation path map analysis, forming a visual layer of key pollutant diffusion paths. It transforms the abstract map into an intuitive decision map, directly guiding the optimization of monitoring points and the precise deployment of emergency interception facilities.

[0056] The specific work involves: after generating a dynamic pollution concentration field in each simulation, extracting pollutant propagation characteristics; based on three-dimensional raster concentration data, using gradient field analysis to calculate the main pollutant concentration gradient direction and magnitude for each grid cell, thereby identifying the main axis direction of overall pollutant migration; quantifying the migration velocity of the pollutant front by calculating the difference in concentration fields between adjacent time steps, combined with the preset grid size and simulation time interval (15 minutes), with the velocity ranging from 0.01 m / s to 0.5 m / s, depending on the driving intensity of the water flow and wind fields; simultaneously, calculating the concentration decay rate segmented along the main axis direction of pollutant migration according to the distance from the pollution source, generating an index of concentration change with distance. Alternatively, a power-law decay curve can be used, with the decay coefficient dynamically determined based on pollutant properties and environmental conditions. A directed weighted network for pollutant propagation is constructed based on direction, velocity, and decay gradient information. Nodes represent key spatial locations, and directed edges represent pollutant propagation paths. The edge weights include propagation velocity and decay gradient information, extracting a pollutant propagation structure with clear spatial topological relationships. Based on this extracted spatial topological structure, it is overlaid with precise spatial data of the farm's infrastructure. Infrastructure data includes, but is not limited to: 3D model coordinates of pollution sources such as livestock pens, manure treatment ponds, and oxidation ponds; vector distribution maps of sewage pipe networks (diameter DN200-DN500); and data on roads within the farm area. Geographic information of green belts, walls, and natural barriers outside the site is used to determine the interaction between transmission paths and physical elements through spatial correlation and buffer zone analysis. Each identified transmission channel is assigned a pollution intensity level (calculated based on the pollutant flux flowing through the channel, categorized into high, medium, and low levels) and a transmission risk weight (the weight calculation integrates the channel's diversion capacity, downstream environmental sensitivity, and barrier blocking effectiveness), thereby constructing a structured and attribute-rich pollutant transmission path map. In-depth analysis of the constructed pollutant transmission path map is then performed to support risk management decisions. Map analysis algorithms are used to identify high-pollution transmission channels (those with high pollution intensity and high risk) that play a dominant role in the entire pollution diffusion process. By extracting high-pollution transmission channels and key risk nodes from the map (such as pipeline junctions, sewage outlets, and barrier gaps) that have a critical impact on the collection, diversion, or blockage of pollutants, this map is used as a key control target. It is then combined with the base geographic information map to generate a thematic visualization layer. This layer uses arrows of different colors and widths to vectorize high-pollution transmission channels, highlights key risk nodes with symbols, and displays their pollution intensity, risk weight, and real-time simulated concentration attributes. This provides environmental managers with an intuitive spatial distribution map of key pollutant diffusion paths, which can be used to guide the precise deployment of monitoring points, optimize the layout of emergency interception facilities, and assess the effectiveness of risk prevention and control measures.

[0057] The multi-pollutant coupling prediction module is used to establish a pollutant concentration map based on a three-dimensional pollutant diffusion model by introducing knowledge graph technology. It integrates the nonlinear interaction between multiple pollutants and external intervention factors to construct a multi-pollutant compound pollution prediction model, analyze the trend of pollutant concentration changes, and achieve accurate prediction from single pollutant indicators to multi-pollutant compound pollution. By comprehensively considering the interaction, it effectively improves the accuracy of compound pollution early warning. The multi-pollutant coupling prediction module includes a pollutant interaction knowledge graph unit and a compound pollution risk prediction unit.

[0058] The pollutant interaction knowledge graph unit, based on a three-dimensional pollutant diffusion model and pollutant propagation path map, combines knowledge graph technology to establish a pollutant concentration map showing the physical, chemical, and biological interactions between multiple pollutants. It integrates external intervention factors such as drug residues and disinfectant use, dynamically updating the synergistic, antagonistic, and transformation mechanisms between pollutants. This enhances the model's understanding of complex pollution, dynamically updates interaction relationships, and deepens the system's understanding of complex pollution. Based on the three-dimensional pollutant diffusion model and pollutant propagation path map, it establishes an initial knowledge graph structure with pollutants as nodes, including pollutant entities such as ammonia nitrogen, COD, and disinfectants, enabling the pollutant entities and their... Attributes are stored and associated in a structured manner, supporting the formal definition and efficient retrieval of subsequent interaction relationships. By integrating historical monitoring data and external intervention information, physical, chemical and biological interaction relationships between pollutants are defined, and the synergistic, antagonistic and transformation mechanisms in the initial knowledge graph structure are updated. This enables a dynamic characterization of the complex interaction mechanisms between pollutants, providing an interpretable relational network basis for the analysis of compound pollution effects. Furthermore, a dynamically updated pollutant concentration map is established, which reflects the nonlinear interaction and compound pollution effects of multiple pollutants in complex environments in real time. This enhances the system's understanding and response capabilities to real-time pollution conditions and supports more accurate compound pollution risk assessment and early warning decisions.

[0059] The specific work involves constructing an initial knowledge graph structure centered on pollutant entities, based on the dynamic concentration field and pollutant propagation path map output by the three-dimensional pollutant diffusion model. This graph uses common key pollutants in the farm environment as nodes, including ammonia nitrogen (NH3-N), chemical oxygen demand (COD), typical veterinary antibiotics, and commonly used disinfectants (sodium hypochlorite, peracetic acid). Each pollutant node defines a series of attribute fields, including its molecular formula, molecular weight, water solubility, typical background concentration range in water, and diffusion coefficient (the diffusion coefficient of NH3-N in water at 20℃ is approximately 1.8 × 10⁻⁶). - 5 cm 2The initial architecture of the graph is organized using a resource description framework to ensure standardized representation and machine readability of entities, attributes, and subsequent relationship edges, based on the entity-relationship structure, degradation half-life, and distribution coefficients in different environmental phases (water, sediment, organisms). This ensures standardized representation and machine readability of entities, attributes, and subsequent relationship edges. Building upon the initial entity structure, interaction relationship edges between pollutant nodes are defined and dynamically updated by integrating historical monitoring data sequences and external structured intervention information (disinfectant dosing records, veterinary drug usage logs, feed composition change logs, etc.). Relationship edge types are categorized into physical adsorption, chemical transformation, and biodegradation based on their mechanisms of action. Each relationship edge is associated with an influence weight factor and a set of kinetic parameters to quantify the strength and direction of the interaction (synergistic, antagonistic, or transformative). The weights and parameters of the relationship edges are calibrated online using real-time sensor data streams. Based on the established entity-relationship structure, dynamic pollutant concentrations are constructed and continuously updated. The system generates a dynamic graph that statically describes pollutant properties and interactions. It simultaneously integrates real-time rasterized concentration predictions from a 3D pollutant diffusion model (spatiotemporal resolution: horizontal grid 0.5-2 meters, vertical stratification 0.5 meters, 15-minute time intervals) and key channel pollution flux data from the pollutant concentration graph. The update mechanism combines event-driven and periodic scanning: when the deviation between real-time monitoring data and model predictions exceeds a preset threshold, parameter reassessment of relevant nodes and edges is triggered. Furthermore, a periodic consistency check and knowledge reasoning of the entire graph is performed every 24 hours. The built-in rule engine (based on SWRL semantic rules) infers potential emerging interactions or complex pollution effects, ultimately generating a dynamic graph presented in a visual interface. This graph supports tracing the evolution history of pollutant concentrations, topological changes in interaction networks, and the spatial distribution of complex pollution risk indices by timestamp.

[0060] The compound pollution risk prediction unit constructs a multi-pollutant compound pollution prediction model based on pollutant concentration maps and historical data. It analyzes pollutant concentration trends in real time, provides early warnings of potential pollution events, and buys valuable time for emergency response. Based on pollutant concentration maps, the model combines deep learning to construct a multi-pollutant compound pollution prediction model. The model inputs include current pollutant concentrations, environmental parameters, and historical trends, enabling high-precision dynamic simulation and risk quantification of the synergistic and antagonistic effects of multiple pollutants. The model analyzes the concentration trends and interactions of each pollutant, generating multi-pollutant coupled pollution prediction results and a time-series concentration distribution map. The multi-pollutant coupled pollution prediction results are represented by a compound pollution index, providing a spatially visualized prediction view of pollutant diffusion paths and concentration evolution over the next few hours. Based on the model prediction results and a comparison with preset warning thresholds, it provides real-time warnings of potential compound pollution events and outputs the pollution type and spatiotemporal range, achieving tiered warnings and precise location, providing clear spatiotemporal targets and decision-making basis for emergency response.

[0061] The specific work involves constructing a multi-pollutant composite pollution prediction model based on the established dynamic pollutant concentration map. Deep learning is employed, using a sequence prediction architecture combining long short-term memory networks and attention mechanisms. The model's input dimension is designed as a multi-dimensional time series, including: a current pollutant concentration vector, covering real-time monitoring values ​​of ammonia nitrogen, COD, and at least three typical veterinary antibiotics (sulfamethoxazole, enrofloxacin, oxytetracycline, etc.), with each concentration data unit standardized in mg / L and the numerical range set based on historical monitoring statistics; an environmental parameter vector, including wind speed, wind direction, water temperature, water flow velocity, pH value, and dissolved oxygen concentration, with each environmental parameter acquired in real-time from the deployed IoT sensor network via a standardized interface; and historical trend characteristics, extracted through a sliding time window, showing pollutant concentrations at 15-minute intervals over the past 24 hours. The degree sequence, after first-order differencing and standardization, constitutes a historical trend feature vector with 96 time steps. All input data undergoes missing value imputation and outlier cleaning before entering the model to ensure data quality. After the multi-pollutant complex pollution prediction model is trained, it is deployed on a real-time inference server, automatically triggering a prediction calculation every 15 minutes, corresponding to the output cycle of the three-dimensional pollutant diffusion model. The prediction task includes the concentration changes of each target pollutant in the next 6 hours (24 time steps). The multi-pollutant complex pollution prediction model uses an internal attention mechanism weight to quantify the interaction strength between ammonia nitrogen, COD, and antibiotics. Positive values ​​indicate synergistic effects, and negative values ​​indicate antagonistic effects. The prediction results generate the concentration prediction values ​​of each pollutant at each future time point, and simultaneously generate multi-pollutant coupled pollution prediction results. This result is expressed as a complex pollution index. It is stated that it calculates and integrates the predicted concentrations of each pollutant with their corresponding interaction weights, and associates the predicted concentration sequence with spatial location. Combined with the farm's geographic information system, it generates a time series map of pollutant concentration distribution every 15 minutes for the next 6 hours. This map is overlaid on the farm's two-dimensional base map in the form of a heat map, clearly showing the migration path and diffusion trend of high pollutant concentration areas.

[0062] Composite Pollution Index The calculation formula is as follows:

[0063] ;

[0064] ;

[0065] In the formula: This is a composite pollution index, representing the overall pollution risk level of multiple pollutants. The total number of pollutant types involved in the assessment; For the first The weighting factors for each pollutant are determined comprehensively based on its degree of harm to the environment or health, environmental standard limits, and importance in the interaction network. For the first Pollutants in the predicted time Normalized concentration; To predict concentration; This refers to the standard limit for this pollutant; This is a coefficient for adjusting the interaction between pollutants, used to amplify or suppress the intensity of the interaction effect on the comprehensive index, and is set based on historical data or expert experience. The logarithm represents the number of pollutants that indicate a significant interaction. For the first The intensity coefficient of the interaction between pollutants is calculated by weighting the attention mechanism within the model. Positive values ​​indicate synergistic effects (enhancing pollution), while negative values ​​indicate antagonistic effects (reducing pollution).

[0066] Utilizing a built-in hierarchical early warning threshold system, which is set with reference to the national "Emission Standard of Pollutants for Livestock and Poultry Breeding" (GB18596-2001) and local environmental protection regulations, the early warning threshold for compound pollution events is defined based on the compound pollution index P, and is divided into four levels: blue (P>0.7), yellow (P>0.8), orange (P>0.9), and red (P>1.0). After each round of prediction, the predicted value of the compound pollution index is automatically compared with the early warning threshold at each level in real time. When the predicted value of the compound pollution index exceeds any early warning threshold, the corresponding level of early warning is immediately triggered. The early warning information is pushed to the monitoring center and the mobile terminals of management personnel in real time through the message queue. At the same time, the system automatically analyzes and outputs the detailed type of pollution and the precise spatiotemporal range. The spatiotemporal range is described by a GIS polygon area and a future time window. This area is composed of all grid cells whose predicted concentration exceeds the threshold of a single pollutant. All early warning events, prediction results, and output information are automatically recorded in the early warning log database.

[0067] The emission reduction decision optimization module is used to generate adaptive emission reduction scheduling strategies based on pollution prediction and deep Q-network reinforcement learning algorithm. It takes pollutant concentration map and sewage treatment equipment status as input, optimizes sewage treatment process, learns and outputs the optimal scheduling strategy, and realizes intelligent optimization of treatment process.

[0068] The multi-constraint adaptive scheduling module is used to integrate the multi-dimensional constraints of the operating status, treatment cost and energy consumption cost of sewage treatment equipment, dynamically evaluate the feasibility and economy of emission reduction scheduling strategies, adjust the scheduling scheme in real time, generate the optimal emission reduction scheduling scheme, ensure compliant and efficient decision-making, comprehensively consider multiple constraints, and dynamically output the economical, compliant and feasible optimal scheduling scheme.

[0069] The risk warning and response module generates risk warning information and disposal suggestions in real time based on pollution prediction and decision optimization results. It also monitors the implementation effect of emission reduction scheduling plans, collects feedback data, and continuously optimizes prediction models and decision-making strategies by combining historical execution records and environmental changes, forming a closed loop of warning-disposal-feedback-optimization, driving the system to continuously evolve.

[0070] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: the emission reduction decision optimization module includes the following steps: based on pollutant concentration maps and real-time status data of sewage treatment equipment, a reinforcement learning environment is constructed, emission reduction control actions are set as a set of equipment adjustment instructions, multi-source data fusion is realized, a high-fidelity simulation environment is provided for intelligent regulation, the authenticity and reliability of strategy generation are enhanced, a deep Q-network algorithm is adopted, with pollutant concentration reduction and stable equipment operation as reward functions, the model is trained to generate an adaptive emission reduction scheduling strategy, the model is guided to find the optimal balance between emission compliance and stable operation, an adaptive regulation strategy that takes into account both environmental protection and economy is output, and a real-time emission reduction scheduling strategy is output, including the opening and closing of sewage treatment equipment, flow regulation and chemical dosing schemes, optimizing the pollutant treatment process, realizing a second-level response from decision to execution, improving treatment efficiency, and reducing the risk of exceeding standards and operating costs;

[0071] The specific work content is as follows: In the reinforcement learning environment construction phase, based on the real-time pollution load data in the pollutant concentration spectrum and the operating status data of the wastewater treatment equipment, a state observation space is established. This state observation space includes three types of key parameters: First, pollution load parameters, including influent ammonia nitrogen concentration, COD concentration, and antibiotic concentration (sulfonamides 0-5 mg / L); second, equipment operating parameters, covering lift pump frequency, aeration fan air volume, chemical dosing pump flow rate, and sludge return ratio (30-100%); third, treatment efficiency parameters, including pollutant removal rate and effluent water quality data for each process stage. The action space is defined as a set of equipment adjustment commands, including discrete control commands: lift pump frequency with an adjustment step of 5 Hz, aeration fan air volume with an adjustment step of 10 m³ / s. 3 The adjustment unit is / min, and the dosage increment is 5L / h. The start-up and shutdown status of each piece of equipment is considered an independent control dimension. Environmental state transitions are achieved through a wastewater treatment process mechanism model, which is based on mass conservation and reaction kinetics, with the nitrification rate constant ranging from 0.15 to 0.35d. -1 The denitrification rate constant is taken as 0.05-0.15d. -1When using the deep Q-network algorithm for strategy training, a neural network structure with three fully connected layers is constructed, with 256, 128, and 64 hidden neurons respectively. The activation function is ReLU. The reward function design comprehensively considers environmental benefits and operating costs: when the effluent ammonia nitrogen concentration is below 15 mg / L and the COD concentration is below 100 mg / L, a positive reward of +10 is given; for every 1 mg / L decrease in ammonia nitrogen concentration, the reward is +0.5; for every 10 mg / L decrease in COD concentration, the reward is +0.3; for every unit change in equipment frequency or air volume, a penalty of -0.1 is applied; the cost of chemical dosing is calculated as a penalty of 0.5 yuan per liter. An experience replay mechanism is used during training, with a replay buffer capacity of 10,000 experience samples and a batch training sample size of 64. The exploration strategy uses the ε-greedy algorithm, with an initial exploration rate of 0.9 that linearly decays to 0.1 as training progresses. A discount factor is used. γ is set to 0.95, and each training cycle contains 1000 time steps, with the time step size corresponding to a 15-minute scheduling cycle in actual operation. In the strategy output and execution phase, the trained deep Q-network model is deployed on the real-time control server. Every 15 minutes, the optimal action sequence is generated based on the current state observations. The output strategies specifically include: pump frequency adjustment commands, aeration fan air volume control commands, PAC agent dosage commands, and sludge return pump start / stop commands. The control system sends the commands to the PLC execution layer through the OPCUA protocol, with the execution response time controlled within 5 seconds. At the same time, a strategy verification mechanism is established: after each batch of strategy execution, the actual effluent water quality data for the next two cycles (30 minutes) is collected and compared with the expected results. When the ammonia nitrogen or COD removal rate deviation exceeds 15%, the online fine-tuning program of the model is triggered. All scheduling strategies, execution results, and water quality data are stored in the historical database in real time.

[0072] The multi-constraint adaptive scheduling module includes the following steps: constructing a multi-constraint evaluation model that includes constraints such as equipment operating status, processing cost, energy consumption cost, and environmental compliance, and establishing an objective optimization function to achieve quantitative trade-offs and comprehensive optimal decision-making among multiple objectives; dynamically evaluating the feasibility and economy of each emission reduction scheduling strategy based on emission reduction scheduling strategies and real-time operating data; calculating a comprehensive execution score; dynamically eliminating infeasible solutions; focusing on economically efficient strategies; adjusting emission reduction scheduling schemes according to evaluation results; generating the optimal emission reduction scheduling scheme that meets multiple constraints; and synchronously updating it to the wastewater treatment control system to ensure that the final scheme is compliant, economical, and immediately executable.

[0073] The specific work content is as follows: In the stage of constructing a multi-constraint evaluation model, an objective optimization function is designed to quantitatively evaluate the comprehensive effectiveness of the emission reduction scheduling strategy. This model integrates four categories of constraints: equipment operating status, treatment cost, energy consumption cost, and environmental compliance. Equipment operating status constraints include the upper limit of the booster pump frequency, the range of aeration fan air volume, and the sludge return ratio. The treatment cost constraint sets the PAC agent dosage cost at 0.5 yuan / liter and the PAM agent cost at 2.0 yuan / kg. The energy consumption cost is calculated based on the real-time industrial electricity price and equipment power, where the booster pump power is 7.5–22kW and the aeration fan power is 15–75kW. The environmental compliance constraint follows the "Emission Standard of Pollutants for Livestock and Poultry Breeding Industry" (GB18596-2001), setting the effluent ammonia nitrogen concentration limit to 15mg / L, COD concentration limit to 100mg / L, and total nitrogen limit to 25mg / L. The objective optimization function adopts a weighted summation form, and its expression is: In the formula: To reduce processing costs, For energy consumption costs, For exceeding the standard penalty item (a fine of 50 yuan for every 1 mg / L of ammonia nitrogen or COD exceeding the standard), the weighting coefficient is... Set according to actual management needs and meet them The function optimization objective is to minimize The system ensures that the strategy operates economically while meeting compliance requirements. Based on emission reduction scheduling strategies and real-time operational data, the feasibility and economic efficiency of each strategy are dynamically evaluated, and a comprehensive execution score is calculated. Scheduling strategies are received from the reinforcement learning environment every 15 minutes. Equipment control parameters are extracted from the strategies, including the target frequency of booster pumps, the target airflow of aeration blowers, the PAC dosage, and the start / stop status of sludge return pumps. A process mechanism model is used for rapid simulation to predict effluent quality (ammonia nitrogen, COD) and cost data within 30 minutes of implementing the strategy. A multi-constraint evaluation model is used to calculate the treatment cost, energy cost, and compliance penalty of the strategy, and a weighted sum is used to obtain a preliminary score. Further introduce feasibility correction factors (Range 0–1), its value is determined by the current availability of the equipment, the load rate of the process section, and the operational safety boundary. The final comprehensive performance score is calculated as follows: The lower the score, the better the overall performance of the strategy. All strategies to be evaluated and their scores are stored in the strategy evaluation database in real time. Based on the evaluation results, the emission reduction scheduling scheme is automatically adjusted and the optimal emission reduction scheduling scheme that meets multiple constraints is generated. The scheme with the lowest overall execution score is selected from the strategy evaluation database. The strategy is used as the candidate optimal solution. If multiple strategies have similar scores (difference ≤ 5%), the solution with smaller equipment adjustment range and more stable operation is selected first. The generated optimal solution clearly includes the following executable instructions: increase pump frequency set value, aeration fan air volume set value, PAC dosing flow rate, sludge return pump target status (on / off) and the timing arrangement of each instruction execution. The solution is sent to the PLC execution layer of the sewage treatment control system in a structured data packet format via the OPCUA protocol to ensure that the instruction is received and parsed within 5 seconds. At the same time, the full text of the solution and the sending timestamp are updated to the central control database and pushed to the monitoring interface for visualization display, so that operators can confirm and trace. The evaluation-adjustment-sending process is repeated every 15 minutes to achieve dynamic closed-loop optimization of emission reduction scheduling.

[0074] The risk warning and response module includes the following steps: Based on the composite pollution prediction results and the optimal emission reduction scheduling plan, it generates risk warning information and emergency response suggestions in real time, and sends them out through a visual interface and message push, realizing multi-level risk second-level warning and intelligent response suggestion push, improving the speed of emergency response and the pertinence of decision-making; it monitors the changes in pollutant concentration and equipment status during the execution of the emission reduction scheduling plan, collects the execution effect feedback data, and forms a closed-loop tracking of the entire process of strategy-execution-monitoring, providing real-time data support for strategy optimization; combined with historical execution records and environmental changes, it continuously optimizes the pollutant prediction model and emission reduction decision-making strategy, forming a closed-loop decision support and system self-learning mechanism, promoting the continuous evolution of the system in real operation, and gradually improving the prediction accuracy and scheduling reliability;

[0075] The specific work involves the following: When the composite pollution index P output by the multi-pollutant composite pollution prediction model exceeds the preset warning threshold (blue: P>0.7, yellow: P>0.8, orange: P>0.9, red: P>1.0), or when the optimal emission reduction scheduling scheme simulation predicts that key water quality indicators (ammonia nitrogen, COD, total nitrogen) will exceed the limits of the "Emission Standard of Pollutants for Livestock and Poultry Breeding Industry" (GB18596-2001), the system immediately triggers a warning. The warning information is pushed in real time to the central monitoring screen, audible and visual alarms, and authorized management personnel's mobile terminal APP in structured JSON format via message middleware. The push content includes the warning level, trigger time, predicted pollutant type exceeding the standard, and predicted spatiotemporal range of exceeding the standard (described by GIS polygon coordinates and a time window of 1-6 hours in the future). Preliminary emergency response suggestions are automatically generated based on pollutant interaction relationships and contingency plan databases in a knowledge graph. All warning events and related information are written to the warning log database in real time, recording a unique event ID, timestamp, etc. Processing status and operation traces; After the optimal emission reduction scheduling plan is issued to the PLC execution layer via the OPCUA protocol, real-time monitoring of the plan execution process is initiated. The monitoring data sources include online water quality sensors (ammonia nitrogen, COD, total nitrogen, pH, dissolved oxygen, with a monitoring frequency of 5 minutes / time) deployed in each process section (inlet, biological tank, sedimentation tank outlet, total discharge outlet, etc.), as well as status feedback signals of the equipment control system (real-time frequency, current, and operating status of booster pumps, aeration blowers, and dosing pumps, with a collection frequency of 1 second / time). The system collects the actual change curve of pollutant concentration, the equipment instruction execution compliance rate (the deviation between the target value and the actual value must be controlled within ±5%), and the actual removal rate of key water quality parameters within a 15-minute monitoring cycle. After the data is verified and cleaned, it is compared with the predicted value before the plan is executed, the prediction deviation is calculated, and it is packaged into an execution effect feedback data package. In addition to the monitoring value, the feedback data package also marks the external environmental interference factors and is subsequently stored in the historical execution effect database.Based on accumulated historical execution records, environmental change data, and feedback effect data, a closed-loop optimization and self-learning mechanism is established. During the low-load period in the early morning every day, the routine task of model optimization is automatically started. This task first performs regression analysis and error statistics on the output results of the pollutant prediction model and the actual monitoring data in the past 24 hours. If the average absolute percentage error of the prediction of a certain type of pollutant (specific sulfonamide antibiotics) is consistently higher than 15%, the automatic parameter calibration process of the pollutant prediction sub-model is triggered. The LSTM-attention network model is trained using the latest data increment. At the same time, the optimization of emission reduction decision strategy is achieved through the periodic retraining of the reinforcement learning model: every two weeks, the state-action-reward-new state sequence stored in the historical execution records is used as experience samples to update the experience replay buffer of the deep Q network, and an offline training cycle is started (training steps 10,000, batch size 64) to generate new weight parameters for the policy network. The updated multi-pollutant composite pollution prediction model and decision strategy need to be verified in the simulation environment through backtesting based on historical scenarios. Only after ensuring its performance improvement can it be deployed to the online system in the next low-risk period, gradually replacing the old version and realizing the continuous autonomous evolution of the system in terms of environmental adaptability and decision effectiveness. ;

[0076] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A livestock farming risk prediction and decision support system based on big data analysis, characterized in that, include: The environmental digital twin modeling module is used to construct a three-dimensional pollutant diffusion model of a farm by integrating multi-source sensor data and three-dimensional modeling technology, and to construct a pollutant propagation path map, so as to realize the visualization of the dynamic migration and distribution of pollutants in the aquaculture environment. The multi-pollutant coupling prediction module is used to establish a pollutant concentration map based on a three-dimensional pollutant diffusion model by introducing knowledge graph technology, and to integrate the nonlinear interaction between multiple pollutants and external intervention factors to construct a multi-pollutant composite pollution prediction model and analyze the trend of pollutant concentration changes. The emission reduction decision optimization module is used to autonomously generate adaptive emission reduction scheduling strategies based on pollution prediction and a deep Q-network-based reinforcement learning algorithm, taking pollutant concentration maps and wastewater treatment equipment status as inputs. The multi-constraint adaptive scheduling module is used to integrate multi-dimensional constraints such as the operating status of wastewater treatment equipment, treatment costs, and energy consumption costs, dynamically evaluate the feasibility and economy of emission reduction scheduling strategies, adjust the scheduling scheme in real time, and generate the optimal emission reduction scheduling scheme. The risk warning and response module generates risk warning information and disposal suggestions in real time based on pollution prediction and decision optimization results. It also monitors the implementation effect of emission reduction scheduling plans, collects feedback data, and continuously optimizes prediction models and decision-making strategies by combining historical implementation records and environmental changes.

2. The livestock farming risk prediction and decision support system based on big data analysis according to claim 1, characterized in that: The environmental digital twin modeling module includes a pollutant three-dimensional diffusion simulation unit and a risk path mapping construction unit; The pollutant three-dimensional diffusion simulation unit, based on the spatial geographic information, environmental parameters and pollutant property data of the farm, constructs a three-dimensional pollutant diffusion model, simulates the migration, diffusion and coupling process of multiple pollutants under different environmental conditions in real time, and outputs a dynamic pollution concentration field. The risk path mapping construction unit is used to construct a pollutant propagation path map by combining the pollutant diffusion simulation results analyzed by the three-dimensional pollutant diffusion model, and to identify high-pollution propagation channels and key risk nodes.

3. The livestock farming risk prediction and decision support system based on big data analysis according to claim 2, characterized in that: The pollutant three-dimensional diffusion simulation unit includes the following steps: Acquire three-dimensional spatial geographic information, environmental parameter data, and pollutant property data of the farm, and construct a three-dimensional pollutant diffusion model; Based on a three-dimensional pollutant diffusion model and combined with real-time sensor monitoring data, the dynamic migration process of multiple pollutants such as ammonia nitrogen, COD and drug residues under wind, water flow and temperature conditions was simulated. It outputs dynamic pollution concentration fields at different altitudes and in different regions in real time, forming a spatiotemporal distribution map of pollutant concentration.

4. The livestock farming risk prediction and decision support system based on big data analysis according to claim 2, characterized in that: The risk path mapping construction unit includes the following steps: Based on the dynamic pollution concentration field, the main direction, speed and concentration decay gradient of pollutant propagation are identified, and the spatial topology of pollution propagation is extracted. By combining the layout of farm facilities and the distribution of natural barriers, a pollutant transmission path map is constructed, and the pollution intensity level and transmission risk weight of different transmission channels are marked. By analyzing pollutant propagation pathway maps, high-pollution transmission channels and key risk nodes are identified, forming a visual layer of key pollutant diffusion pathways.

5. The livestock farming risk prediction and decision support system based on big data analysis according to claim 2, characterized in that: The multi-pollutant coupling prediction module includes a pollutant interaction knowledge graph unit and a composite pollution risk prediction unit. The pollutant interaction knowledge graph unit, based on a three-dimensional pollutant diffusion model and a pollutant propagation path map, combines knowledge graph technology to establish a pollutant concentration map of physical, chemical and biological interactions among multiple pollutants, integrates external intervention factors, and dynamically updates the synergistic, antagonistic and transformation mechanisms among pollutants. The compound pollution risk prediction unit constructs a multi-pollutant compound pollution prediction model based on pollutant concentration maps and historical data, analyzes the trend of pollutant concentration changes in real time, and provides early warning of potential pollution events.

6. The livestock farming risk prediction and decision support system based on big data analysis according to claim 5, characterized in that: The pollutant interactive knowledge graph unit includes the following steps: Based on a three-dimensional pollutant diffusion model and pollutant propagation path map, an initial knowledge graph structure with pollutants as nodes is established, including pollutant entities such as ammonia nitrogen, COD, and disinfectants. By integrating historical monitoring data with external intervention information, we define the physical, chemical, and biological interaction relationships between pollutants, update the synergistic, antagonistic, and transformation mechanisms in the initial knowledge graph structure, and then establish a dynamically updated pollutant concentration map to reflect the nonlinear interactions and compound pollution effects of multiple pollutants in complex environments in real time.

7. The livestock farming risk prediction and decision support system based on big data analysis according to claim 5, characterized in that: The compound pollution risk prediction unit includes the following steps: Based on pollutant concentration maps, a multi-pollutant complex pollution prediction model is constructed by combining a deep learning model. The model input includes the current concentration of pollutants, environmental parameters, and historical change trends. The concentration change trends and interaction effects of each pollutant are analyzed using a multi-pollutant composite pollution prediction model. The prediction results of multi-pollutant coupled pollution and the time series diagram of concentration distribution are generated. The multi-pollutant coupled pollution prediction results are represented by the composite pollution index. Based on the comparison between the model prediction results and the preset warning threshold, the system provides real-time warnings of existing complex pollution events and outputs the pollution type and spatiotemporal range.

8. The livestock farming risk prediction and decision support system based on big data analysis according to claim 5, characterized in that: The emission reduction decision optimization module includes the following steps: Based on pollutant concentration maps and real-time status data of wastewater treatment equipment, a reinforcement learning environment is constructed, and emission reduction control actions are set as a set of equipment adjustment commands. A deep Q-network algorithm is used, with pollutant concentration reduction and stable equipment operation as reward functions, to train the model and generate an adaptive emission reduction scheduling strategy; Output real-time emission reduction scheduling strategies, including the start-up and shutdown of wastewater treatment equipment, flow regulation and chemical dosing schemes, to optimize the pollutant treatment process.

9. A livestock farming risk prediction and decision support system based on big data analysis according to claim 8, characterized in that: The multi-constraint adaptive scheduling module includes the following steps: Construct a multi-constraint assessment model that includes equipment operating status, processing costs, energy consumption costs, and environmental compliance constraints, and establish an objective optimization function; Based on emission reduction scheduling strategies and real-time operational data, the feasibility and economic efficiency of each emission reduction scheduling strategy are dynamically evaluated, and a comprehensive performance score is calculated. The emission reduction scheduling plan is adjusted based on the assessment results, and the optimal emission reduction scheduling plan that meets multiple constraints is generated and updated to the wastewater treatment control system simultaneously.

10. A livestock farming risk prediction and decision support system based on big data analysis according to claim 9, characterized in that: The risk warning and response module includes the following steps: Based on the results of complex pollution prediction and the optimal emission reduction scheduling scheme, risk warning information and emergency response suggestions are generated in real time and sent out through a visual interface and message push. Monitor changes in pollutant concentrations and equipment status during the implementation of emission reduction scheduling plans, and collect feedback data on implementation effectiveness; By combining historical implementation records with environmental changes, we continuously optimize pollutant prediction models and emission reduction decision-making strategies, forming a closed-loop decision support and system self-learning mechanism.

Images